@effect-uai/inworld 0.6.0 → 0.7.0

package/dist/InworldRealtimeTranscriber.d.mts

@@ -8,7 +8,7 @@ declare namespace InworldRealtimeTranscriber_d_exports {
   export { Config, InworldTranscribeRequest, InworldTranscriber, InworldTranscriberService, layer, make };
 }
 declare const make: (cfg: Config) => Effect.Effect<{
-  transcribe: (r: InworldTranscribeRequest) => Effect.Effect<any, any, never>;
+  transcribe: (r: InworldTranscribeRequest) => Effect.Effect<any, any, unknown>;
   streamTranscriptionFrom: (audioIn: any, request: any) => _$effect_Stream0.Stream<TranscriptEvent, any, unknown>;
 }, never, HttpClient.HttpClient>;
 /**

package/dist/InworldSynthesizer.d.mts.map

@@ -1 +1 @@
-{"version":3,"file":"InworldSynthesizer.d.mts","names":[],"sources":["../src/InworldSynthesizer.ts"],"mappings":";;;;;;;;;;;;;;;;;AA+CA;;;;KAAY,wBAAA,GAA2B,IAAA,CAAK,uBAAA;EAAA,SACjC,KAAA,EAAO,eAAA;EAAA,SACP,OAAA,EAAS,cAAA;EAAA,SACT,WAAA;EAAA,SACA,YAAA,GAAe,mBAAA;EAAA,SACf,sBAAA;AAAA;AAAA,KAGC,yBAAA;EAAA,SACD,UAAA,GAAa,CAAA,EAAG,wBAAA,KAA6B,MAAA,CAAO,MAAA,CAAO,SAAA,EAAW,OAAA,CAAQ,OAAA;EAAA,SAC9E,eAAA,GACP,CAAA,EAAG,wBAAA,KACA,MAAA,CAAO,MAAA,CAAO,UAAA,EAAY,OAAA,CAAQ,OAAA;EAAA,SAC9B,mBAAA,EAAqB,wBAAA;EAAA,SACrB,kBAAA,EAAoB,wBAAA;EAAA,SACpB,wBAAA,EAA0B,wBAAA;AAAA;AAAA,cACpC,uBAAA;cAEY,kBAAA,SAA2B,uBAAA;AAAA,KAK5B,MAAA;EAAA,SAAoB,MAAA,EAAQ,QAAA,CAAS,QAAA;EAAA,SAAmB,OAAA;AAAA;;cA2BvD,SAAA,GAAa,CAAA,EAAG,wBAAA,KAAwB,MAAA,CAAA,MAAA;;;;;;;;;;;;;;cA4CxC,cAAA,GAAkB,GAAA,EAAK,MAAA,MAAY,OAAA,EAAS,wBAAA,KAAwB,MAAA,CAAA,MAAA,WAAA,UAAA,CAAA,UAAA;;cAoCpE,mBAAA,GAAuB,GAAA,EAAK,MAAA,MAAY,OAAA,EAAS,wBAAA,KAAwB,MAAA,CAAA,MAAA,aAAA,UAAA,CAAA,UAAA;;cA2CzE,uBAAA,EAAyB,wBAAA;;cAUzB,6BAAA,EAA+B,wBAAA;AAAA,cAc/B,IAAA,GACX,GAAA,EAAK,MAAA,KACJ,MAAA,CAAO,MAAA,CAAO,yBAAA,SAAkC,UAAA,CAAW,UAAA;;;;;;cAgBjD,KAAA,GACX,GAAA,EAAK,MAAA,KACJ,KAAA,CAAM,KAAA,CAAM,kBAAA,GAAqB,iBAAA,SAA0B,UAAA,CAAW,UAAA"}
+{"version":3,"file":"InworldSynthesizer.d.mts","names":[],"sources":["../src/InworldSynthesizer.ts"],"mappings":";;;;;;;;;;;;;;;;;AA+CA;;;;KAAY,wBAAA,GAA2B,IAAA,CAAK,uBAAA;EAAA,SACjC,KAAA,EAAO,eAAA;EAAA,SACP,OAAA,EAAS,cAAA;EAAA,SACT,WAAA;EAAA,SACA,YAAA,GAAe,mBAAA;EAAA,SACf,sBAAA;AAAA;AAAA,KAGC,yBAAA;EAAA,SACD,UAAA,GAAa,CAAA,EAAG,wBAAA,KAA6B,MAAA,CAAO,MAAA,CAAO,SAAA,EAAW,OAAA,CAAQ,OAAA;EAAA,SAC9E,eAAA,GACP,CAAA,EAAG,wBAAA,KACA,MAAA,CAAO,MAAA,CAAO,UAAA,EAAY,OAAA,CAAQ,OAAA;EAAA,SAC9B,mBAAA,EAAqB,wBAAA;EAAA,SACrB,kBAAA,EAAoB,wBAAA;EAAA,SACpB,wBAAA,EAA0B,wBAAA;AAAA;AAAA,cACpC,uBAAA;cAEY,kBAAA,SAA2B,uBAAA;AAAA,KAK5B,MAAA;EAAA,SAAoB,MAAA,EAAQ,QAAA,CAAS,QAAA;EAAA,SAAmB,OAAA;AAAA;;cAyBvD,SAAA,GAAa,CAAA,EAAG,wBAAA,KAAwB,MAAA,CAAA,MAAA;;;;;;;;;;;;;;cA4CxC,cAAA,GAAkB,GAAA,EAAK,MAAA,MAAY,OAAA,EAAS,wBAAA,KAAwB,MAAA,CAAA,MAAA,WAAA,UAAA,CAAA,UAAA;;cAoCpE,mBAAA,GAAuB,GAAA,EAAK,MAAA,MAAY,OAAA,EAAS,wBAAA,KAAwB,MAAA,CAAA,MAAA,aAAA,UAAA,CAAA,UAAA;;cA2CzE,uBAAA,EAAyB,wBAAA;;cAUzB,6BAAA,EAA+B,wBAAA;AAAA,cAc/B,IAAA,GACX,GAAA,EAAK,MAAA,KACJ,MAAA,CAAO,MAAA,CAAO,yBAAA,SAAkC,UAAA,CAAW,UAAA;;;;;;cAgBjD,KAAA,GACX,GAAA,EAAK,MAAA,KACJ,KAAA,CAAM,KAAA,CAAM,kBAAA,GAAqB,iBAAA,SAA0B,UAAA,CAAW,UAAA"}

package/dist/InworldSynthesizer.mjs

@@ -31,14 +31,14 @@ var InworldSynthesizer_exports = /* @__PURE__ */ __exportAll({
 var InworldSynthesizer = class extends Context.Service()("@betalyra/effect-uai/providers/inworld/InworldSynthesizer") {};
 /**
 * Inworld accepts inline phoneme overrides as `/ipa/` tokens directly
-* in the text — no separate field. Only IPA is supported; entries with
-* other encodings are silently dropped (audio still renders with the
-* default pronunciation). Whole-word, case-insensitive replacement.
+* in the text, no separate field. `CustomPronunciation.pronunciation`
+* is always IPA, so every entry is inlined. Whole-word,
+* case-insensitive replacement.
 */
 const escapeForRegex = (s) => s.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
 const applyPronunciations = (text, pronunciations) => {
 	if (pronunciations === void 0) return text;
-	return Array.reduce(pronunciations, text, (acc, p) => p.encoding === "ipa" ? acc.replace(new RegExp(`\\b${escapeForRegex(p.phrase)}\\b`, "i"), `/${p.pronunciation}/`) : acc);
+	return Array.reduce(pronunciations, text, (acc, p) => acc.replace(new RegExp(`\\b${escapeForRegex(p.phrase)}\\b`, "i"), `/${p.pronunciation}/`));
 };
 /** Exported for reuse by the realtime variant's BOS `create` frame. */
 const buildBody = (r) => Effect.gen(function* () {

package/dist/InworldSynthesizer.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"InworldSynthesizer.mjs","names":["Arr"],"sources":["../src/InworldSynthesizer.ts"],"sourcesContent":["/**\n * Sync + chunked-HTTP TTS via Inworld's REST endpoints.\n *\n * - `synthesize` → `POST /tts/v1/voice` (JSON body, single base64 audio in\n *   response).\n * - `streamSynthesis` → `POST /tts/v1/voice:stream` (NDJSON response, one\n *   JSON object per line, each carrying a base64 audio chunk).\n *\n * Incremental text-in (`streamSynthesisFrom`) lives at the\n * `InworldRealtimeSynthesizer` subpath, which adds the WS path and the\n * `TtsIncrementalText` capability marker.\n */\nimport { Array as Arr, Context, Effect, Layer, Redacted, Schema, Stream } from \"effect\"\nimport { HttpClient, HttpClientRequest } from \"effect/unstable/http\"\nimport * as AiError from \"@effect-uai/core/AiError\"\nimport type { AudioBlob, AudioChunk } from \"@effect-uai/core/Audio\"\nimport * as JSONL from \"@effect-uai/core/JSONL\"\nimport {\n  type CommonStreamSynthesizeRequest,\n  type CommonSynthesizeRequest,\n  type CustomPronunciation,\n  SpeechSynthesizer,\n  type SpeechSynthesizerService,\n} from \"@effect-uai/core/SpeechSynthesizer\"\nimport {\n  audioConfigFor,\n  authHeader,\n  decodeAudioContent,\n  defaultFormat,\n  httpStatusError,\n  transportFailure,\n} from \"./codec.js\"\nimport type { InworldDeliveryMode, InworldTtsModel, InworldVoiceId } from \"./models.js\"\n\n// ---------------------------------------------------------------------------\n// Public types\n// ---------------------------------------------------------------------------\n\n/**\n * Inworld-typed synthesize request.\n *\n * - `deliveryMode` only honored by `inworld-tts-2` (`STABLE` / `BALANCED` /\n *   `CREATIVE` — older models ignore it).\n * - `temperature` valid range is `(0, 2]` per docs.\n * - `applyTextNormalization` defaults to `ON`; pass `\"OFF\"` to feed the\n *   model raw text (faster, but punctuation pacing is on you).\n */\nexport type InworldSynthesizeRequest = Omit<CommonSynthesizeRequest, \"model\" | \"voiceId\"> & {\n  readonly model: InworldTtsModel\n  readonly voiceId: InworldVoiceId\n  readonly temperature?: number\n  readonly deliveryMode?: InworldDeliveryMode\n  readonly applyTextNormalization?: \"ON\" | \"OFF\"\n}\n\nexport type InworldSynthesizerService = {\n  readonly synthesize: (r: InworldSynthesizeRequest) => Effect.Effect<AudioBlob, AiError.AiError>\n  readonly streamSynthesis: (\n    r: InworldSynthesizeRequest,\n  ) => Stream.Stream<AudioChunk, AiError.AiError>\n  readonly streamSynthesisFrom: SpeechSynthesizerService[\"streamSynthesisFrom\"]\n  readonly synthesizeDialogue: SpeechSynthesizerService[\"synthesizeDialogue\"]\n  readonly streamSynthesizeDialogue: SpeechSynthesizerService[\"streamSynthesizeDialogue\"]\n}\n\nexport class InworldSynthesizer extends Context.Service<\n  InworldSynthesizer,\n  InworldSynthesizerService\n>()(\"@betalyra/effect-uai/providers/inworld/InworldSynthesizer\") {}\n\nexport type Config = { readonly apiKey: Redacted.Redacted; readonly baseUrl?: string }\n\n// ---------------------------------------------------------------------------\n// Request body\n// ---------------------------------------------------------------------------\n\n/**\n * Inworld accepts inline phoneme overrides as `/ipa/` tokens directly\n * in the text — no separate field. Only IPA is supported; entries with\n * other encodings are silently dropped (audio still renders with the\n * default pronunciation). Whole-word, case-insensitive replacement.\n */\nconst escapeForRegex = (s: string): string => s.replace(/[.*+?^${}()|[\\]\\\\]/g, \"\\\\$&\")\n\nconst applyPronunciations = (\n  text: string,\n  pronunciations: ReadonlyArray<CustomPronunciation> | undefined,\n): string => {\n  if (pronunciations === undefined) return text\n  return Arr.reduce(pronunciations, text, (acc, p) =>\n    p.encoding === \"ipa\"\n      ? acc.replace(new RegExp(`\\\\b${escapeForRegex(p.phrase)}\\\\b`, \"i\"), `/${p.pronunciation}/`)\n      : acc,\n  )\n}\n\n/** Exported for reuse by the realtime variant's BOS `create` frame. */\nexport const buildBody = (r: InworldSynthesizeRequest) =>\n  Effect.gen(function* () {\n    const format = r.outputFormat ?? defaultFormat\n    const audioConfig = yield* audioConfigFor(format, r.speed)\n    const body = {\n      text: applyPronunciations(r.text, r.pronunciations),\n      voiceId: r.voiceId,\n      modelId: r.model,\n      audioConfig,\n      ...(r.languageCode !== undefined && { language: r.languageCode }),\n      ...(r.deliveryMode !== undefined && { deliveryMode: r.deliveryMode }),\n      ...(r.temperature !== undefined && { temperature: r.temperature }),\n      ...(r.applyTextNormalization !== undefined && {\n        applyTextNormalization: r.applyTextNormalization,\n      }),\n    }\n    return { body, format }\n  })\n\n// ---------------------------------------------------------------------------\n// Wire schemas\n// ---------------------------------------------------------------------------\n\nconst SyncResponse = Schema.Struct({ audioContent: Schema.String })\nconst decodeSyncResponse = Schema.decodeUnknownEffect(SyncResponse)\n\nconst StreamLine = Schema.Struct({\n  result: Schema.optional(Schema.Struct({ audioContent: Schema.optional(Schema.String) })),\n  error: Schema.optional(Schema.Unknown),\n})\n\n// ---------------------------------------------------------------------------\n// HTTP plumbing\n// ---------------------------------------------------------------------------\n\nconst baseUrl = (cfg: Config) => cfg.baseUrl ?? \"https://api.inworld.ai\"\n\nconst buildRequest = (cfg: Config, path: \"/tts/v1/voice\" | \"/tts/v1/voice:stream\", body: unknown) =>\n  HttpClientRequest.post(`${baseUrl(cfg)}${path}`).pipe(\n    HttpClientRequest.setHeader(\"Authorization\", authHeader(cfg.apiKey)),\n    HttpClientRequest.bodyJsonUnsafe(body),\n  )\n\n/** Exported for reuse by `InworldRealtimeSynthesizer`. */\nexport const synthesizeImpl = (cfg: Config) => (request: InworldSynthesizeRequest) =>\n  Effect.gen(function* () {\n    const client = yield* HttpClient.HttpClient\n    const { body, format } = yield* buildBody(request)\n    const httpRequest = buildRequest(cfg, \"/tts/v1/voice\", body)\n    const response = yield* client.execute(httpRequest).pipe(Effect.mapError(transportFailure))\n    if (response.status >= 400) {\n      const text = yield* response.text.pipe(Effect.orElseSucceed(() => \"\"))\n      return yield* Effect.fail(httpStatusError(response.status, text))\n    }\n    const json = yield* response.json.pipe(Effect.mapError(transportFailure))\n    const wire = yield* decodeSyncResponse(json).pipe(\n      Effect.mapError(\n        (cause) =>\n          new AiError.GenerationFailed({\n            provider: \"inworld\",\n            raw: { message: \"sync TTS response missing `audioContent`\", cause, json },\n          }),\n      ),\n    )\n    const bytes = yield* decodeAudioContent(wire.audioContent)\n    return { format, bytes } satisfies AudioBlob\n  })\n\nconst lineToStream = (line: typeof StreamLine.Type): Stream.Stream<AudioChunk, AiError.AiError> => {\n  if (line.error !== undefined) {\n    return Stream.fail(new AiError.GenerationFailed({ provider: \"inworld\", raw: line.error }))\n  }\n  const b64 = line.result?.audioContent\n  if (b64 === undefined) return Stream.empty\n  return Stream.fromEffect(\n    decodeAudioContent(b64).pipe(Effect.map((bytes): AudioChunk => ({ bytes }))),\n  )\n}\n\n/** Exported for reuse by `InworldRealtimeSynthesizer`. */\nexport const streamSynthesisImpl = (cfg: Config) => (request: InworldSynthesizeRequest) =>\n  Stream.unwrap(\n    Effect.gen(function* () {\n      const client = yield* HttpClient.HttpClient\n      const { body } = yield* buildBody(request)\n      const httpRequest = buildRequest(cfg, \"/tts/v1/voice:stream\", body)\n      const response = yield* client.execute(httpRequest).pipe(Effect.mapError(transportFailure))\n      if (response.status >= 400) {\n        const text = yield* response.text.pipe(Effect.orElseSucceed(() => \"\"))\n        return Stream.fail(httpStatusError(response.status, text))\n      }\n      return response.stream.pipe(\n        Stream.mapError(transportFailure),\n        JSONL.fromBytes,\n        JSONL.parse(StreamLine),\n        Stream.mapError(\n          (cause): AiError.AiError =>\n            new AiError.GenerationFailed({\n              provider: \"inworld\",\n              raw: { message: \"NDJSON parse failed\", cause },\n            }),\n        ),\n        Stream.flatMap(lineToStream),\n      )\n    }),\n  )\n\nconst streamUnsupported = <E, R>(\n  _textIn: Stream.Stream<string, E, R>,\n  _request: CommonStreamSynthesizeRequest,\n): Stream.Stream<AudioChunk, AiError.AiError | E, R> => {\n  const fail: Stream.Stream<AudioChunk, AiError.AiError | E, R> = Stream.fail(\n    new AiError.Unsupported({\n      provider: \"inworld\",\n      capability: \"streamSynthesisFrom\",\n      reason:\n        \"This Layer is sync-only. Import `@effect-uai/inworld/InworldRealtimeSynthesizer` to use the WS streamSynthesisFrom path.\",\n    }),\n  )\n  return fail\n}\n\n/** Exported for reuse by `InworldRealtimeSynthesizer`. */\nexport const dialogueUnsupportedImpl: SpeechSynthesizerService[\"synthesizeDialogue\"] = () =>\n  Effect.fail(\n    new AiError.Unsupported({\n      provider: \"inworld\",\n      capability: \"synthesizeDialogue\",\n      reason: \"Inworld TTS has no multi-speaker endpoint.\",\n    }),\n  )\n\n/** Exported for reuse by `InworldRealtimeSynthesizer`. */\nexport const streamDialogueUnsupportedImpl: SpeechSynthesizerService[\"streamSynthesizeDialogue\"] =\n  () =>\n    Stream.fail(\n      new AiError.Unsupported({\n        provider: \"inworld\",\n        capability: \"streamSynthesizeDialogue\",\n        reason: \"Inworld TTS has no multi-speaker endpoint.\",\n      }),\n    )\n\n// ---------------------------------------------------------------------------\n// Constructors\n// ---------------------------------------------------------------------------\n\nexport const make = (\n  cfg: Config,\n): Effect.Effect<InworldSynthesizerService, never, HttpClient.HttpClient> =>\n  Effect.map(HttpClient.HttpClient.asEffect(), (client) => ({\n    synthesize: (r) =>\n      synthesizeImpl(cfg)(r).pipe(Effect.provideService(HttpClient.HttpClient, client)),\n    streamSynthesis: (r) =>\n      streamSynthesisImpl(cfg)(r).pipe(Stream.provideService(HttpClient.HttpClient, client)),\n    streamSynthesisFrom: streamUnsupported,\n    synthesizeDialogue: dialogueUnsupportedImpl,\n    streamSynthesizeDialogue: streamDialogueUnsupportedImpl,\n  }))\n\n/**\n * Sync-only Layer. Registers `InworldSynthesizer` + the generic\n * `SpeechSynthesizer`. Does **not** register `TtsIncrementalText` — for\n * incremental text-in use `@effect-uai/inworld/InworldRealtimeSynthesizer`.\n */\nexport const layer = (\n  cfg: Config,\n): Layer.Layer<InworldSynthesizer | SpeechSynthesizer, never, HttpClient.HttpClient> =>\n  Layer.merge(\n    Layer.effect(InworldSynthesizer, make(cfg)),\n    Layer.effect(\n      SpeechSynthesizer,\n      Effect.map(\n        make(cfg),\n        (s): SpeechSynthesizerService => ({\n          synthesize: (req: CommonSynthesizeRequest) =>\n            s.synthesize(req as InworldSynthesizeRequest),\n          streamSynthesis: (req: CommonSynthesizeRequest) =>\n            s.streamSynthesis(req as InworldSynthesizeRequest),\n          streamSynthesisFrom: s.streamSynthesisFrom,\n          synthesizeDialogue: s.synthesizeDialogue,\n          streamSynthesizeDialogue: s.streamSynthesizeDialogue,\n        }),\n      ),\n    ),\n  )\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAiEA,IAAa,qBAAb,cAAwC,QAAQ,SAG7C,CAAC,4DAA4D,CAAC;;;;;;;AAcjE,MAAM,kBAAkB,MAAsB,EAAE,QAAQ,uBAAuB,OAAO;AAEtF,MAAM,uBACJ,MACA,mBACW;AACX,KAAI,mBAAmB,KAAA,EAAW,QAAO;AACzC,QAAOA,MAAI,OAAO,gBAAgB,OAAO,KAAK,MAC5C,EAAE,aAAa,QACX,IAAI,QAAQ,IAAI,OAAO,MAAM,eAAe,EAAE,OAAO,CAAC,MAAM,IAAI,EAAE,IAAI,EAAE,cAAc,GAAG,GACzF,IACL;;;AAIH,MAAa,aAAa,MACxB,OAAO,IAAI,aAAa;CACtB,MAAM,SAAS,EAAE,gBAAgB;CACjC,MAAM,cAAc,OAAO,eAAe,QAAQ,EAAE,MAAM;AAa1D,QAAO;EAAE,MAAA;GAXP,MAAM,oBAAoB,EAAE,MAAM,EAAE,eAAe;GACnD,SAAS,EAAE;GACX,SAAS,EAAE;GACX;GACA,GAAI,EAAE,iBAAiB,KAAA,KAAa,EAAE,UAAU,EAAE,cAAc;GAChE,GAAI,EAAE,iBAAiB,KAAA,KAAa,EAAE,cAAc,EAAE,cAAc;GACpE,GAAI,EAAE,gBAAgB,KAAA,KAAa,EAAE,aAAa,EAAE,aAAa;GACjE,GAAI,EAAE,2BAA2B,KAAA,KAAa,EAC5C,wBAAwB,EAAE,wBAC3B;GAEU;EAAE;EAAQ;EACvB;AAMJ,MAAM,eAAe,OAAO,OAAO,EAAE,cAAc,OAAO,QAAQ,CAAC;AACnE,MAAM,qBAAqB,OAAO,oBAAoB,aAAa;AAEnE,MAAM,aAAa,OAAO,OAAO;CAC/B,QAAQ,OAAO,SAAS,OAAO,OAAO,EAAE,cAAc,OAAO,SAAS,OAAO,OAAO,EAAE,CAAC,CAAC;CACxF,OAAO,OAAO,SAAS,OAAO,QAAQ;CACvC,CAAC;AAMF,MAAM,WAAW,QAAgB,IAAI,WAAW;AAEhD,MAAM,gBAAgB,KAAa,MAAgD,SACjF,kBAAkB,KAAK,GAAG,QAAQ,IAAI,GAAG,OAAO,CAAC,KAC/C,kBAAkB,UAAU,iBAAiB,WAAW,IAAI,OAAO,CAAC,EACpE,kBAAkB,eAAe,KAAK,CACvC;;AAGH,MAAa,kBAAkB,SAAiB,YAC9C,OAAO,IAAI,aAAa;CACtB,MAAM,SAAS,OAAO,WAAW;CACjC,MAAM,EAAE,MAAM,WAAW,OAAO,UAAU,QAAQ;CAClD,MAAM,cAAc,aAAa,KAAK,iBAAiB,KAAK;CAC5D,MAAM,WAAW,OAAO,OAAO,QAAQ,YAAY,CAAC,KAAK,OAAO,SAAS,iBAAiB,CAAC;AAC3F,KAAI,SAAS,UAAU,KAAK;EAC1B,MAAM,OAAO,OAAO,SAAS,KAAK,KAAK,OAAO,oBAAoB,GAAG,CAAC;AACtE,SAAO,OAAO,OAAO,KAAK,gBAAgB,SAAS,QAAQ,KAAK,CAAC;;CAEnE,MAAM,OAAO,OAAO,SAAS,KAAK,KAAK,OAAO,SAAS,iBAAiB,CAAC;AAWzE,QAAO;EAAE;EAAQ,OAAA,OADI,oBAAmB,OATpB,mBAAmB,KAAK,CAAC,KAC3C,OAAO,UACJ,UACC,IAAI,QAAQ,iBAAiB;GAC3B,UAAU;GACV,KAAK;IAAE,SAAS;IAA4C;IAAO;IAAM;GAC1E,CAAC,CACL,CACF,EAC4C,aAAa;EAClC;EACxB;AAEJ,MAAM,gBAAgB,SAA6E;AACjG,KAAI,KAAK,UAAU,KAAA,EACjB,QAAO,OAAO,KAAK,IAAI,QAAQ,iBAAiB;EAAE,UAAU;EAAW,KAAK,KAAK;EAAO,CAAC,CAAC;CAE5F,MAAM,MAAM,KAAK,QAAQ;AACzB,KAAI,QAAQ,KAAA,EAAW,QAAO,OAAO;AACrC,QAAO,OAAO,WACZ,mBAAmB,IAAI,CAAC,KAAK,OAAO,KAAK,WAAuB,EAAE,OAAO,EAAE,CAAC,CAC7E;;;AAIH,MAAa,uBAAuB,SAAiB,YACnD,OAAO,OACL,OAAO,IAAI,aAAa;CACtB,MAAM,SAAS,OAAO,WAAW;CACjC,MAAM,EAAE,SAAS,OAAO,UAAU,QAAQ;CAC1C,MAAM,cAAc,aAAa,KAAK,wBAAwB,KAAK;CACnE,MAAM,WAAW,OAAO,OAAO,QAAQ,YAAY,CAAC,KAAK,OAAO,SAAS,iBAAiB,CAAC;AAC3F,KAAI,SAAS,UAAU,KAAK;EAC1B,MAAM,OAAO,OAAO,SAAS,KAAK,KAAK,OAAO,oBAAoB,GAAG,CAAC;AACtE,SAAO,OAAO,KAAK,gBAAgB,SAAS,QAAQ,KAAK,CAAC;;AAE5D,QAAO,SAAS,OAAO,KACrB,OAAO,SAAS,iBAAiB,EACjC,MAAM,WACN,MAAM,MAAM,WAAW,EACvB,OAAO,UACJ,UACC,IAAI,QAAQ,iBAAiB;EAC3B,UAAU;EACV,KAAK;GAAE,SAAS;GAAuB;GAAO;EAC/C,CAAC,CACL,EACD,OAAO,QAAQ,aAAa,CAC7B;EACD,CACH;AAEH,MAAM,qBACJ,SACA,aACsD;AAStD,QARgE,OAAO,KACrE,IAAI,QAAQ,YAAY;EACtB,UAAU;EACV,YAAY;EACZ,QACE;EACH,CAAC,CAEO;;;AAIb,MAAa,gCACX,OAAO,KACL,IAAI,QAAQ,YAAY;CACtB,UAAU;CACV,YAAY;CACZ,QAAQ;CACT,CAAC,CACH;;AAGH,MAAa,sCAET,OAAO,KACL,IAAI,QAAQ,YAAY;CACtB,UAAU;CACV,YAAY;CACZ,QAAQ;CACT,CAAC,CACH;AAML,MAAa,QACX,QAEA,OAAO,IAAI,WAAW,WAAW,UAAU,GAAG,YAAY;CACxD,aAAa,MACX,eAAe,IAAI,CAAC,EAAE,CAAC,KAAK,OAAO,eAAe,WAAW,YAAY,OAAO,CAAC;CACnF,kBAAkB,MAChB,oBAAoB,IAAI,CAAC,EAAE,CAAC,KAAK,OAAO,eAAe,WAAW,YAAY,OAAO,CAAC;CACxF,qBAAqB;CACrB,oBAAoB;CACpB,0BAA0B;CAC3B,EAAE;;;;;;AAOL,MAAa,SACX,QAEA,MAAM,MACJ,MAAM,OAAO,oBAAoB,KAAK,IAAI,CAAC,EAC3C,MAAM,OACJ,mBACA,OAAO,IACL,KAAK,IAAI,GACR,OAAiC;CAChC,aAAa,QACX,EAAE,WAAW,IAAgC;CAC/C,kBAAkB,QAChB,EAAE,gBAAgB,IAAgC;CACpD,qBAAqB,EAAE;CACvB,oBAAoB,EAAE;CACtB,0BAA0B,EAAE;CAC7B,EACF,CACF,CACF"}
+{"version":3,"file":"InworldSynthesizer.mjs","names":["Arr"],"sources":["../src/InworldSynthesizer.ts"],"sourcesContent":["/**\n * Sync + chunked-HTTP TTS via Inworld's REST endpoints.\n *\n * - `synthesize` → `POST /tts/v1/voice` (JSON body, single base64 audio in\n *   response).\n * - `streamSynthesis` → `POST /tts/v1/voice:stream` (NDJSON response, one\n *   JSON object per line, each carrying a base64 audio chunk).\n *\n * Incremental text-in (`streamSynthesisFrom`) lives at the\n * `InworldRealtimeSynthesizer` subpath, which adds the WS path and the\n * `TtsIncrementalText` capability marker.\n */\nimport { Array as Arr, Context, Effect, Layer, Redacted, Schema, Stream } from \"effect\"\nimport { HttpClient, HttpClientRequest } from \"effect/unstable/http\"\nimport * as AiError from \"@effect-uai/core/AiError\"\nimport type { AudioBlob, AudioChunk } from \"@effect-uai/core/Audio\"\nimport * as JSONL from \"@effect-uai/core/JSONL\"\nimport {\n  type CommonStreamSynthesizeRequest,\n  type CommonSynthesizeRequest,\n  type CustomPronunciation,\n  SpeechSynthesizer,\n  type SpeechSynthesizerService,\n} from \"@effect-uai/core/SpeechSynthesizer\"\nimport {\n  audioConfigFor,\n  authHeader,\n  decodeAudioContent,\n  defaultFormat,\n  httpStatusError,\n  transportFailure,\n} from \"./codec.js\"\nimport type { InworldDeliveryMode, InworldTtsModel, InworldVoiceId } from \"./models.js\"\n\n// ---------------------------------------------------------------------------\n// Public types\n// ---------------------------------------------------------------------------\n\n/**\n * Inworld-typed synthesize request.\n *\n * - `deliveryMode` only honored by `inworld-tts-2` (`STABLE` / `BALANCED` /\n *   `CREATIVE` — older models ignore it).\n * - `temperature` valid range is `(0, 2]` per docs.\n * - `applyTextNormalization` defaults to `ON`; pass `\"OFF\"` to feed the\n *   model raw text (faster, but punctuation pacing is on you).\n */\nexport type InworldSynthesizeRequest = Omit<CommonSynthesizeRequest, \"model\" | \"voiceId\"> & {\n  readonly model: InworldTtsModel\n  readonly voiceId: InworldVoiceId\n  readonly temperature?: number\n  readonly deliveryMode?: InworldDeliveryMode\n  readonly applyTextNormalization?: \"ON\" | \"OFF\"\n}\n\nexport type InworldSynthesizerService = {\n  readonly synthesize: (r: InworldSynthesizeRequest) => Effect.Effect<AudioBlob, AiError.AiError>\n  readonly streamSynthesis: (\n    r: InworldSynthesizeRequest,\n  ) => Stream.Stream<AudioChunk, AiError.AiError>\n  readonly streamSynthesisFrom: SpeechSynthesizerService[\"streamSynthesisFrom\"]\n  readonly synthesizeDialogue: SpeechSynthesizerService[\"synthesizeDialogue\"]\n  readonly streamSynthesizeDialogue: SpeechSynthesizerService[\"streamSynthesizeDialogue\"]\n}\n\nexport class InworldSynthesizer extends Context.Service<\n  InworldSynthesizer,\n  InworldSynthesizerService\n>()(\"@betalyra/effect-uai/providers/inworld/InworldSynthesizer\") {}\n\nexport type Config = { readonly apiKey: Redacted.Redacted; readonly baseUrl?: string }\n\n// ---------------------------------------------------------------------------\n// Request body\n// ---------------------------------------------------------------------------\n\n/**\n * Inworld accepts inline phoneme overrides as `/ipa/` tokens directly\n * in the text, no separate field. `CustomPronunciation.pronunciation`\n * is always IPA, so every entry is inlined. Whole-word,\n * case-insensitive replacement.\n */\nconst escapeForRegex = (s: string): string => s.replace(/[.*+?^${}()|[\\]\\\\]/g, \"\\\\$&\")\n\nconst applyPronunciations = (\n  text: string,\n  pronunciations: ReadonlyArray<CustomPronunciation> | undefined,\n): string => {\n  if (pronunciations === undefined) return text\n  return Arr.reduce(pronunciations, text, (acc, p) =>\n    acc.replace(new RegExp(`\\\\b${escapeForRegex(p.phrase)}\\\\b`, \"i\"), `/${p.pronunciation}/`),\n  )\n}\n\n/** Exported for reuse by the realtime variant's BOS `create` frame. */\nexport const buildBody = (r: InworldSynthesizeRequest) =>\n  Effect.gen(function* () {\n    const format = r.outputFormat ?? defaultFormat\n    const audioConfig = yield* audioConfigFor(format, r.speed)\n    const body = {\n      text: applyPronunciations(r.text, r.pronunciations),\n      voiceId: r.voiceId,\n      modelId: r.model,\n      audioConfig,\n      ...(r.languageCode !== undefined && { language: r.languageCode }),\n      ...(r.deliveryMode !== undefined && { deliveryMode: r.deliveryMode }),\n      ...(r.temperature !== undefined && { temperature: r.temperature }),\n      ...(r.applyTextNormalization !== undefined && {\n        applyTextNormalization: r.applyTextNormalization,\n      }),\n    }\n    return { body, format }\n  })\n\n// ---------------------------------------------------------------------------\n// Wire schemas\n// ---------------------------------------------------------------------------\n\nconst SyncResponse = Schema.Struct({ audioContent: Schema.String })\nconst decodeSyncResponse = Schema.decodeUnknownEffect(SyncResponse)\n\nconst StreamLine = Schema.Struct({\n  result: Schema.optional(Schema.Struct({ audioContent: Schema.optional(Schema.String) })),\n  error: Schema.optional(Schema.Unknown),\n})\n\n// ---------------------------------------------------------------------------\n// HTTP plumbing\n// ---------------------------------------------------------------------------\n\nconst baseUrl = (cfg: Config) => cfg.baseUrl ?? \"https://api.inworld.ai\"\n\nconst buildRequest = (cfg: Config, path: \"/tts/v1/voice\" | \"/tts/v1/voice:stream\", body: unknown) =>\n  HttpClientRequest.post(`${baseUrl(cfg)}${path}`).pipe(\n    HttpClientRequest.setHeader(\"Authorization\", authHeader(cfg.apiKey)),\n    HttpClientRequest.bodyJsonUnsafe(body),\n  )\n\n/** Exported for reuse by `InworldRealtimeSynthesizer`. */\nexport const synthesizeImpl = (cfg: Config) => (request: InworldSynthesizeRequest) =>\n  Effect.gen(function* () {\n    const client = yield* HttpClient.HttpClient\n    const { body, format } = yield* buildBody(request)\n    const httpRequest = buildRequest(cfg, \"/tts/v1/voice\", body)\n    const response = yield* client.execute(httpRequest).pipe(Effect.mapError(transportFailure))\n    if (response.status >= 400) {\n      const text = yield* response.text.pipe(Effect.orElseSucceed(() => \"\"))\n      return yield* Effect.fail(httpStatusError(response.status, text))\n    }\n    const json = yield* response.json.pipe(Effect.mapError(transportFailure))\n    const wire = yield* decodeSyncResponse(json).pipe(\n      Effect.mapError(\n        (cause) =>\n          new AiError.GenerationFailed({\n            provider: \"inworld\",\n            raw: { message: \"sync TTS response missing `audioContent`\", cause, json },\n          }),\n      ),\n    )\n    const bytes = yield* decodeAudioContent(wire.audioContent)\n    return { format, bytes } satisfies AudioBlob\n  })\n\nconst lineToStream = (line: typeof StreamLine.Type): Stream.Stream<AudioChunk, AiError.AiError> => {\n  if (line.error !== undefined) {\n    return Stream.fail(new AiError.GenerationFailed({ provider: \"inworld\", raw: line.error }))\n  }\n  const b64 = line.result?.audioContent\n  if (b64 === undefined) return Stream.empty\n  return Stream.fromEffect(\n    decodeAudioContent(b64).pipe(Effect.map((bytes): AudioChunk => ({ bytes }))),\n  )\n}\n\n/** Exported for reuse by `InworldRealtimeSynthesizer`. */\nexport const streamSynthesisImpl = (cfg: Config) => (request: InworldSynthesizeRequest) =>\n  Stream.unwrap(\n    Effect.gen(function* () {\n      const client = yield* HttpClient.HttpClient\n      const { body } = yield* buildBody(request)\n      const httpRequest = buildRequest(cfg, \"/tts/v1/voice:stream\", body)\n      const response = yield* client.execute(httpRequest).pipe(Effect.mapError(transportFailure))\n      if (response.status >= 400) {\n        const text = yield* response.text.pipe(Effect.orElseSucceed(() => \"\"))\n        return Stream.fail(httpStatusError(response.status, text))\n      }\n      return response.stream.pipe(\n        Stream.mapError(transportFailure),\n        JSONL.fromBytes,\n        JSONL.parse(StreamLine),\n        Stream.mapError(\n          (cause): AiError.AiError =>\n            new AiError.GenerationFailed({\n              provider: \"inworld\",\n              raw: { message: \"NDJSON parse failed\", cause },\n            }),\n        ),\n        Stream.flatMap(lineToStream),\n      )\n    }),\n  )\n\nconst streamUnsupported = <E, R>(\n  _textIn: Stream.Stream<string, E, R>,\n  _request: CommonStreamSynthesizeRequest,\n): Stream.Stream<AudioChunk, AiError.AiError | E, R> => {\n  const fail: Stream.Stream<AudioChunk, AiError.AiError | E, R> = Stream.fail(\n    new AiError.Unsupported({\n      provider: \"inworld\",\n      capability: \"streamSynthesisFrom\",\n      reason:\n        \"This Layer is sync-only. Import `@effect-uai/inworld/InworldRealtimeSynthesizer` to use the WS streamSynthesisFrom path.\",\n    }),\n  )\n  return fail\n}\n\n/** Exported for reuse by `InworldRealtimeSynthesizer`. */\nexport const dialogueUnsupportedImpl: SpeechSynthesizerService[\"synthesizeDialogue\"] = () =>\n  Effect.fail(\n    new AiError.Unsupported({\n      provider: \"inworld\",\n      capability: \"synthesizeDialogue\",\n      reason: \"Inworld TTS has no multi-speaker endpoint.\",\n    }),\n  )\n\n/** Exported for reuse by `InworldRealtimeSynthesizer`. */\nexport const streamDialogueUnsupportedImpl: SpeechSynthesizerService[\"streamSynthesizeDialogue\"] =\n  () =>\n    Stream.fail(\n      new AiError.Unsupported({\n        provider: \"inworld\",\n        capability: \"streamSynthesizeDialogue\",\n        reason: \"Inworld TTS has no multi-speaker endpoint.\",\n      }),\n    )\n\n// ---------------------------------------------------------------------------\n// Constructors\n// ---------------------------------------------------------------------------\n\nexport const make = (\n  cfg: Config,\n): Effect.Effect<InworldSynthesizerService, never, HttpClient.HttpClient> =>\n  Effect.map(HttpClient.HttpClient.asEffect(), (client) => ({\n    synthesize: (r) =>\n      synthesizeImpl(cfg)(r).pipe(Effect.provideService(HttpClient.HttpClient, client)),\n    streamSynthesis: (r) =>\n      streamSynthesisImpl(cfg)(r).pipe(Stream.provideService(HttpClient.HttpClient, client)),\n    streamSynthesisFrom: streamUnsupported,\n    synthesizeDialogue: dialogueUnsupportedImpl,\n    streamSynthesizeDialogue: streamDialogueUnsupportedImpl,\n  }))\n\n/**\n * Sync-only Layer. Registers `InworldSynthesizer` + the generic\n * `SpeechSynthesizer`. Does **not** register `TtsIncrementalText` — for\n * incremental text-in use `@effect-uai/inworld/InworldRealtimeSynthesizer`.\n */\nexport const layer = (\n  cfg: Config,\n): Layer.Layer<InworldSynthesizer | SpeechSynthesizer, never, HttpClient.HttpClient> =>\n  Layer.merge(\n    Layer.effect(InworldSynthesizer, make(cfg)),\n    Layer.effect(\n      SpeechSynthesizer,\n      Effect.map(\n        make(cfg),\n        (s): SpeechSynthesizerService => ({\n          synthesize: (req: CommonSynthesizeRequest) =>\n            s.synthesize(req as InworldSynthesizeRequest),\n          streamSynthesis: (req: CommonSynthesizeRequest) =>\n            s.streamSynthesis(req as InworldSynthesizeRequest),\n          streamSynthesisFrom: s.streamSynthesisFrom,\n          synthesizeDialogue: s.synthesizeDialogue,\n          streamSynthesizeDialogue: s.streamSynthesizeDialogue,\n        }),\n      ),\n    ),\n  )\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAiEA,IAAa,qBAAb,cAAwC,QAAQ,SAG7C,CAAC,4DAA4D,CAAC;;;;;;;AAcjE,MAAM,kBAAkB,MAAsB,EAAE,QAAQ,uBAAuB,OAAO;AAEtF,MAAM,uBACJ,MACA,mBACW;AACX,KAAI,mBAAmB,KAAA,EAAW,QAAO;AACzC,QAAOA,MAAI,OAAO,gBAAgB,OAAO,KAAK,MAC5C,IAAI,QAAQ,IAAI,OAAO,MAAM,eAAe,EAAE,OAAO,CAAC,MAAM,IAAI,EAAE,IAAI,EAAE,cAAc,GAAG,CAC1F;;;AAIH,MAAa,aAAa,MACxB,OAAO,IAAI,aAAa;CACtB,MAAM,SAAS,EAAE,gBAAgB;CACjC,MAAM,cAAc,OAAO,eAAe,QAAQ,EAAE,MAAM;AAa1D,QAAO;EAAE,MAAA;GAXP,MAAM,oBAAoB,EAAE,MAAM,EAAE,eAAe;GACnD,SAAS,EAAE;GACX,SAAS,EAAE;GACX;GACA,GAAI,EAAE,iBAAiB,KAAA,KAAa,EAAE,UAAU,EAAE,cAAc;GAChE,GAAI,EAAE,iBAAiB,KAAA,KAAa,EAAE,cAAc,EAAE,cAAc;GACpE,GAAI,EAAE,gBAAgB,KAAA,KAAa,EAAE,aAAa,EAAE,aAAa;GACjE,GAAI,EAAE,2BAA2B,KAAA,KAAa,EAC5C,wBAAwB,EAAE,wBAC3B;GAEU;EAAE;EAAQ;EACvB;AAMJ,MAAM,eAAe,OAAO,OAAO,EAAE,cAAc,OAAO,QAAQ,CAAC;AACnE,MAAM,qBAAqB,OAAO,oBAAoB,aAAa;AAEnE,MAAM,aAAa,OAAO,OAAO;CAC/B,QAAQ,OAAO,SAAS,OAAO,OAAO,EAAE,cAAc,OAAO,SAAS,OAAO,OAAO,EAAE,CAAC,CAAC;CACxF,OAAO,OAAO,SAAS,OAAO,QAAQ;CACvC,CAAC;AAMF,MAAM,WAAW,QAAgB,IAAI,WAAW;AAEhD,MAAM,gBAAgB,KAAa,MAAgD,SACjF,kBAAkB,KAAK,GAAG,QAAQ,IAAI,GAAG,OAAO,CAAC,KAC/C,kBAAkB,UAAU,iBAAiB,WAAW,IAAI,OAAO,CAAC,EACpE,kBAAkB,eAAe,KAAK,CACvC;;AAGH,MAAa,kBAAkB,SAAiB,YAC9C,OAAO,IAAI,aAAa;CACtB,MAAM,SAAS,OAAO,WAAW;CACjC,MAAM,EAAE,MAAM,WAAW,OAAO,UAAU,QAAQ;CAClD,MAAM,cAAc,aAAa,KAAK,iBAAiB,KAAK;CAC5D,MAAM,WAAW,OAAO,OAAO,QAAQ,YAAY,CAAC,KAAK,OAAO,SAAS,iBAAiB,CAAC;AAC3F,KAAI,SAAS,UAAU,KAAK;EAC1B,MAAM,OAAO,OAAO,SAAS,KAAK,KAAK,OAAO,oBAAoB,GAAG,CAAC;AACtE,SAAO,OAAO,OAAO,KAAK,gBAAgB,SAAS,QAAQ,KAAK,CAAC;;CAEnE,MAAM,OAAO,OAAO,SAAS,KAAK,KAAK,OAAO,SAAS,iBAAiB,CAAC;AAWzE,QAAO;EAAE;EAAQ,OAAA,OADI,oBAAmB,OATpB,mBAAmB,KAAK,CAAC,KAC3C,OAAO,UACJ,UACC,IAAI,QAAQ,iBAAiB;GAC3B,UAAU;GACV,KAAK;IAAE,SAAS;IAA4C;IAAO;IAAM;GAC1E,CAAC,CACL,CACF,EAC4C,aAAa;EAClC;EACxB;AAEJ,MAAM,gBAAgB,SAA6E;AACjG,KAAI,KAAK,UAAU,KAAA,EACjB,QAAO,OAAO,KAAK,IAAI,QAAQ,iBAAiB;EAAE,UAAU;EAAW,KAAK,KAAK;EAAO,CAAC,CAAC;CAE5F,MAAM,MAAM,KAAK,QAAQ;AACzB,KAAI,QAAQ,KAAA,EAAW,QAAO,OAAO;AACrC,QAAO,OAAO,WACZ,mBAAmB,IAAI,CAAC,KAAK,OAAO,KAAK,WAAuB,EAAE,OAAO,EAAE,CAAC,CAC7E;;;AAIH,MAAa,uBAAuB,SAAiB,YACnD,OAAO,OACL,OAAO,IAAI,aAAa;CACtB,MAAM,SAAS,OAAO,WAAW;CACjC,MAAM,EAAE,SAAS,OAAO,UAAU,QAAQ;CAC1C,MAAM,cAAc,aAAa,KAAK,wBAAwB,KAAK;CACnE,MAAM,WAAW,OAAO,OAAO,QAAQ,YAAY,CAAC,KAAK,OAAO,SAAS,iBAAiB,CAAC;AAC3F,KAAI,SAAS,UAAU,KAAK;EAC1B,MAAM,OAAO,OAAO,SAAS,KAAK,KAAK,OAAO,oBAAoB,GAAG,CAAC;AACtE,SAAO,OAAO,KAAK,gBAAgB,SAAS,QAAQ,KAAK,CAAC;;AAE5D,QAAO,SAAS,OAAO,KACrB,OAAO,SAAS,iBAAiB,EACjC,MAAM,WACN,MAAM,MAAM,WAAW,EACvB,OAAO,UACJ,UACC,IAAI,QAAQ,iBAAiB;EAC3B,UAAU;EACV,KAAK;GAAE,SAAS;GAAuB;GAAO;EAC/C,CAAC,CACL,EACD,OAAO,QAAQ,aAAa,CAC7B;EACD,CACH;AAEH,MAAM,qBACJ,SACA,aACsD;AAStD,QARgE,OAAO,KACrE,IAAI,QAAQ,YAAY;EACtB,UAAU;EACV,YAAY;EACZ,QACE;EACH,CAAC,CAEO;;;AAIb,MAAa,gCACX,OAAO,KACL,IAAI,QAAQ,YAAY;CACtB,UAAU;CACV,YAAY;CACZ,QAAQ;CACT,CAAC,CACH;;AAGH,MAAa,sCAET,OAAO,KACL,IAAI,QAAQ,YAAY;CACtB,UAAU;CACV,YAAY;CACZ,QAAQ;CACT,CAAC,CACH;AAML,MAAa,QACX,QAEA,OAAO,IAAI,WAAW,WAAW,UAAU,GAAG,YAAY;CACxD,aAAa,MACX,eAAe,IAAI,CAAC,EAAE,CAAC,KAAK,OAAO,eAAe,WAAW,YAAY,OAAO,CAAC;CACnF,kBAAkB,MAChB,oBAAoB,IAAI,CAAC,EAAE,CAAC,KAAK,OAAO,eAAe,WAAW,YAAY,OAAO,CAAC;CACxF,qBAAqB;CACrB,oBAAoB;CACpB,0BAA0B;CAC3B,EAAE;;;;;;AAOL,MAAa,SACX,QAEA,MAAM,MACJ,MAAM,OAAO,oBAAoB,KAAK,IAAI,CAAC,EAC3C,MAAM,OACJ,mBACA,OAAO,IACL,KAAK,IAAI,GACR,OAAiC;CAChC,aAAa,QACX,EAAE,WAAW,IAAgC;CAC/C,kBAAkB,QAChB,EAAE,gBAAgB,IAAgC;CACpD,qBAAqB,EAAE;CACvB,oBAAoB,EAAE;CACtB,0BAA0B,EAAE;CAC7B,EACF,CACF,CACF"}

package/dist/InworldTranscriber.d.mts

@@ -31,7 +31,7 @@ type Config = {
   readonly baseUrl?: string;
 };
 /** Exported for reuse by `InworldRealtimeTranscriber`. */
-declare const transcribeImpl: (cfg: Config) => (request: InworldTranscribeRequest) => Effect.Effect<any, any, HttpClient.HttpClient>;
+declare const transcribeImpl: (cfg: Config) => (request: InworldTranscribeRequest) => Effect.Effect<any, any, unknown>;
 declare const make: (cfg: Config) => Effect.Effect<InworldTranscriberService, never, HttpClient.HttpClient>;
 /**
  * Sync-only Layer. Registers `InworldTranscriber` + the generic

package/dist/InworldTranscriber.d.mts.map

@@ -1 +1 @@
-{"version":3,"file":"InworldTranscriber.d.mts","names":[],"sources":["../src/InworldTranscriber.ts"],"mappings":";;;;;;;;;;;;;AAiCA;;;KAAY,wBAAA,GAA2B,IAAA,CAAK,uBAAA;EAAA,SACjC,KAAA,EAAO,eAAA;EAAA,SACP,aAAA;EAAA,SACA,eAAA;EAAA,SACA,gBAAA;AAAA;AAAA,KAGC,yBAAA;EAAA,SACD,UAAA,GACP,CAAA,EAAG,wBAAA,KACA,MAAA,CAAO,MAAA,CAAO,gBAAA,EAAkB,OAAA,CAAQ,OAAA;EAAA,SACpC,uBAAA,EAAyB,kBAAA;AAAA;AAAA,cACnC,uBAAA;cAEY,kBAAA,SAA2B,uBAAA;AAAA,KAK5B,MAAA;EAAA,SAAoB,MAAA,EAAQ,QAAA,CAAS,QAAA;EAAA,SAAmB,OAAA;AAAA;;cAsFvD,cAAA,GAAkB,GAAA,EAAK,MAAA,MAAY,OAAA,EAAS,wBAAA,KAAwB,MAAA,CAAA,MAAA,WAAA,UAAA,CAAA,UAAA;AAAA,cAwDpE,IAAA,GACX,GAAA,EAAK,MAAA,KACJ,MAAA,CAAO,MAAA,CAAO,yBAAA,SAAkC,UAAA,CAAW,UAAA;;;;;;cAYjD,KAAA,GACX,GAAA,EAAK,MAAA,KACJ,KAAA,CAAM,KAAA,CAAM,kBAAA,GAAqB,WAAA,SAAoB,UAAA,CAAW,UAAA"}
+{"version":3,"file":"InworldTranscriber.d.mts","names":[],"sources":["../src/InworldTranscriber.ts"],"mappings":";;;;;;;;;;;;;AAkCA;;;KAAY,wBAAA,GAA2B,IAAA,CAAK,uBAAA;EAAA,SACjC,KAAA,EAAO,eAAA;EAAA,SACP,aAAA;EAAA,SACA,eAAA;EAAA,SACA,gBAAA;AAAA;AAAA,KAGC,yBAAA;EAAA,SACD,UAAA,GACP,CAAA,EAAG,wBAAA,KACA,MAAA,CAAO,MAAA,CAAO,gBAAA,EAAkB,OAAA,CAAQ,OAAA;EAAA,SACpC,uBAAA,EAAyB,kBAAA;AAAA;AAAA,cACnC,uBAAA;cAEY,kBAAA,SAA2B,uBAAA;AAAA,KAK5B,MAAA;EAAA,SAAoB,MAAA,EAAQ,QAAA,CAAS,QAAA;EAAA,SAAmB,OAAA;AAAA;;cAwFvD,cAAA,GAAkB,GAAA,EAAK,MAAA,MAAY,OAAA,EAAS,wBAAA,KAAwB,MAAA,CAAA,MAAA;AAAA,cAwDpE,IAAA,GACX,GAAA,EAAK,MAAA,KACJ,MAAA,CAAO,MAAA,CAAO,yBAAA,SAAkC,UAAA,CAAW,UAAA;;;;;;cAYjD,KAAA,GACX,GAAA,EAAK,MAAA,KACJ,KAAA,CAAM,KAAA,CAAM,kBAAA,GAAqB,WAAA,SAAoB,UAAA,CAAW,UAAA"}

package/dist/InworldTranscriber.mjs

@@ -4,6 +4,7 @@ import { Context, Effect, Encoding, Layer, Match, Schema, Stream } from "effect"
 import { HttpClient, HttpClientRequest } from "effect/unstable/http";
 import * as AiError from "@effect-uai/core/AiError";
 import { Transcriber } from "@effect-uai/core/Transcriber";
+import * as Capabilities from "@effect-uai/core/Capabilities";
 //#region src/InworldTranscriber.ts
 /**
 * Sync transcription via Inworld's REST endpoint.
@@ -28,10 +29,14 @@ const urlNotSupported = new AiError.InvalidRequest({
 	raw: "Inworld accepts only inline base64. Fetch the URL yourself and pass `{ _tag: \"bytes\", bytes, mimeType }`."
 });
 const audioToBase64 = Match.type().pipe(Match.tag("bytes", (a) => Effect.succeed(Encoding.encodeBase64(a.bytes))), Match.tag("base64", (a) => Effect.succeed(a.base64)), Match.tag("url", () => Effect.fail(urlNotSupported)), Match.exhaustive);
-const promptToTerms = (prompt) => prompt === void 0 ? void 0 : typeof prompt === "string" ? [prompt] : prompt.terms;
 const buildBody = (request) => Effect.gen(function* () {
 	const content = yield* audioToBase64(request.audio);
-	const prompts = promptToTerms(request.prompt);
+	yield* Capabilities.warnDroppedWhen(request.prompt, {
+		provider: "inworld",
+		capability: "prompt",
+		field: "prompt",
+		reason: "Inworld STT has no free-form prompt field; bias via `biasingTerms`."
+	});
 	return {
 		transcribeConfig: {
 			modelId: request.model,
@@ -39,7 +44,7 @@ const buildBody = (request) => Effect.gen(function* () {
 			...request.sampleRateHertz !== void 0 && { sampleRateHertz: request.sampleRateHertz },
 			numberOfChannels: request.numberOfChannels ?? 1,
 			...request.language !== void 0 && { language: request.language },
-			...prompts !== void 0 && { prompts },
+			...request.biasingTerms !== void 0 && { prompts: request.biasingTerms },
 			...request.wordTimestamps === true && { includeWordTimestamps: true }
 		},
 		audioData: { content }

package/dist/InworldTranscriber.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"InworldTranscriber.mjs","names":[],"sources":["../src/InworldTranscriber.ts"],"sourcesContent":["/**\n * Sync transcription via Inworld's REST endpoint.\n *\n * - `transcribe` → `POST /stt/v1/transcribe` with audio as base64 in\n *   `audioData.content`. URL input is not accepted by Inworld — the\n *   adapter rejects `_tag: \"url\"` audio sources with `InvalidRequest`.\n *\n * Streaming transcription lives at the `InworldRealtimeTranscriber`\n * subpath, which adds the WS path and the `SttStreaming` capability marker.\n */\nimport { Context, Effect, Encoding, Layer, Match, Redacted, Schema, Stream } from \"effect\"\nimport { HttpClient, HttpClientRequest } from \"effect/unstable/http\"\nimport * as AiError from \"@effect-uai/core/AiError\"\nimport type { AudioSource } from \"@effect-uai/core/Audio\"\nimport type { TranscriptEvent, TranscriptResult, WordTimestamp } from \"@effect-uai/core/Transcript\"\nimport {\n  type CommonStreamTranscribeRequest,\n  type CommonTranscribeRequest,\n  Transcriber,\n  type TranscriberService,\n} from \"@effect-uai/core/Transcriber\"\nimport { authHeader, httpStatusError, transportFailure } from \"./codec.js\"\nimport type { InworldSttModel } from \"./models.js\"\n\n// ---------------------------------------------------------------------------\n// Public types\n// ---------------------------------------------------------------------------\n\n/**\n * Inworld-typed transcribe request. `numberOfChannels` defaults to 1.\n * `audioEncoding` is autodetected from the audio MIME by default; pass\n * an explicit value to override (useful for raw PCM bytes).\n */\nexport type InworldTranscribeRequest = Omit<CommonTranscribeRequest, \"model\"> & {\n  readonly model: InworldSttModel\n  readonly audioEncoding?: \"AUTO_DETECT\" | \"LINEAR16\" | \"MP3\" | \"OGG_OPUS\" | \"FLAC\"\n  readonly sampleRateHertz?: number\n  readonly numberOfChannels?: number\n}\n\nexport type InworldTranscriberService = {\n  readonly transcribe: (\n    r: InworldTranscribeRequest,\n  ) => Effect.Effect<TranscriptResult, AiError.AiError>\n  readonly streamTranscriptionFrom: TranscriberService[\"streamTranscriptionFrom\"]\n}\n\nexport class InworldTranscriber extends Context.Service<\n  InworldTranscriber,\n  InworldTranscriberService\n>()(\"@betalyra/effect-uai/providers/inworld/InworldTranscriber\") {}\n\nexport type Config = { readonly apiKey: Redacted.Redacted; readonly baseUrl?: string }\n\n// ---------------------------------------------------------------------------\n// AudioSource → base64. Inworld's sync endpoint takes inline base64 only.\n// ---------------------------------------------------------------------------\n\nconst urlNotSupported: AiError.AiError = new AiError.InvalidRequest({\n  provider: \"inworld\",\n  param: \"audio\",\n  raw: 'Inworld accepts only inline base64. Fetch the URL yourself and pass `{ _tag: \"bytes\", bytes, mimeType }`.',\n})\n\nconst audioToBase64: (audio: AudioSource) => Effect.Effect<string, AiError.AiError> =\n  Match.type<AudioSource>().pipe(\n    Match.tag(\"bytes\", (a) => Effect.succeed(Encoding.encodeBase64(a.bytes))),\n    Match.tag(\"base64\", (a) => Effect.succeed(a.base64)),\n    Match.tag(\"url\", () => Effect.fail(urlNotSupported)),\n    Match.exhaustive,\n  )\n\n// ---------------------------------------------------------------------------\n// Request body\n// ---------------------------------------------------------------------------\n\nconst promptToTerms = (\n  prompt: string | { readonly terms: ReadonlyArray<string> } | undefined,\n): ReadonlyArray<string> | undefined =>\n  prompt === undefined ? undefined : typeof prompt === \"string\" ? [prompt] : prompt.terms\n\nconst buildBody = (request: InworldTranscribeRequest) =>\n  Effect.gen(function* () {\n    const content = yield* audioToBase64(request.audio)\n    const prompts = promptToTerms(request.prompt)\n    return {\n      transcribeConfig: {\n        modelId: request.model,\n        audioEncoding: request.audioEncoding ?? \"AUTO_DETECT\",\n        ...(request.sampleRateHertz !== undefined && {\n          sampleRateHertz: request.sampleRateHertz,\n        }),\n        numberOfChannels: request.numberOfChannels ?? 1,\n        ...(request.language !== undefined && { language: request.language }),\n        ...(prompts !== undefined && { prompts }),\n        ...(request.wordTimestamps === true && { includeWordTimestamps: true }),\n      },\n      audioData: { content },\n    }\n  })\n\n// ---------------------------------------------------------------------------\n// Wire schema\n// ---------------------------------------------------------------------------\n\nconst WireWord = Schema.Struct({\n  word: Schema.String,\n  startTimeMs: Schema.optional(Schema.Number),\n  endTimeMs: Schema.optional(Schema.Number),\n  confidence: Schema.optional(Schema.Number),\n})\n\nconst WireResponse = Schema.Struct({\n  transcription: Schema.Struct({\n    transcript: Schema.String,\n    isFinal: Schema.optional(Schema.Boolean),\n    wordTimestamps: Schema.optional(Schema.NullOr(Schema.Array(WireWord))),\n  }),\n})\nconst decodeResponse = Schema.decodeUnknownEffect(WireResponse)\n\nconst wireWordToCommon = (w: typeof WireWord.Type): WordTimestamp | undefined =>\n  w.startTimeMs === undefined || w.endTimeMs === undefined\n    ? undefined\n    : {\n        text: w.word,\n        startSeconds: w.startTimeMs / 1000,\n        endSeconds: w.endTimeMs / 1000,\n        ...(w.confidence !== undefined && { confidence: w.confidence }),\n      }\n\n// ---------------------------------------------------------------------------\n// HTTP plumbing\n// ---------------------------------------------------------------------------\n\nconst baseUrl = (cfg: Config) => cfg.baseUrl ?? \"https://api.inworld.ai\"\n\n/** Exported for reuse by `InworldRealtimeTranscriber`. */\nexport const transcribeImpl = (cfg: Config) => (request: InworldTranscribeRequest) =>\n  Effect.gen(function* () {\n    const client = yield* HttpClient.HttpClient\n    const body = yield* buildBody(request)\n    const httpRequest = HttpClientRequest.post(`${baseUrl(cfg)}/stt/v1/transcribe`).pipe(\n      HttpClientRequest.setHeader(\"Authorization\", authHeader(cfg.apiKey)),\n      HttpClientRequest.bodyJsonUnsafe(body),\n    )\n    const response = yield* client.execute(httpRequest).pipe(Effect.mapError(transportFailure))\n    if (response.status >= 400) {\n      const text = yield* response.text.pipe(Effect.orElseSucceed(() => \"\"))\n      return yield* Effect.fail(httpStatusError(response.status, text))\n    }\n    const json = yield* response.json.pipe(Effect.mapError(transportFailure))\n    const wire = yield* decodeResponse(json).pipe(\n      Effect.mapError(\n        (cause) =>\n          new AiError.GenerationFailed({\n            provider: \"inworld\",\n            raw: { message: \"STT response missing `transcription.transcript`\", cause, json },\n          }),\n      ),\n    )\n    const words = wire.transcription.wordTimestamps\n      ?.map(wireWordToCommon)\n      .filter((w): w is WordTimestamp => w !== undefined)\n    return {\n      text: wire.transcription.transcript,\n      ...(words !== undefined && words.length > 0 && { words }),\n      raw: json,\n    } satisfies TranscriptResult\n  })\n\n// ---------------------------------------------------------------------------\n// streamTranscriptionFrom Unsupported on this Layer\n// ---------------------------------------------------------------------------\n\nconst streamUnsupported = <E, R>(\n  _audioIn: Stream.Stream<Uint8Array, E, R>,\n  _request: CommonStreamTranscribeRequest,\n): Stream.Stream<TranscriptEvent, AiError.AiError | E, R> => {\n  const fail: Stream.Stream<TranscriptEvent, AiError.AiError | E, R> = Stream.fail(\n    new AiError.Unsupported({\n      provider: \"inworld\",\n      capability: \"streamTranscriptionFrom\",\n      reason:\n        \"This Layer is sync-only. Import `@effect-uai/inworld/InworldRealtimeTranscriber` to use the WS streamTranscriptionFrom path.\",\n    }),\n  )\n  return fail\n}\n\n// ---------------------------------------------------------------------------\n// Constructors\n// ---------------------------------------------------------------------------\n\nexport const make = (\n  cfg: Config,\n): Effect.Effect<InworldTranscriberService, never, HttpClient.HttpClient> =>\n  Effect.map(HttpClient.HttpClient.asEffect(), (client) => ({\n    transcribe: (request) =>\n      transcribeImpl(cfg)(request).pipe(Effect.provideService(HttpClient.HttpClient, client)),\n    streamTranscriptionFrom: streamUnsupported,\n  }))\n\n/**\n * Sync-only Layer. Registers `InworldTranscriber` + the generic\n * `Transcriber`. Does **not** register `SttStreaming` — for live\n * transcription use `@effect-uai/inworld/InworldRealtimeTranscriber`.\n */\nexport const layer = (\n  cfg: Config,\n): Layer.Layer<InworldTranscriber | Transcriber, never, HttpClient.HttpClient> =>\n  Layer.merge(\n    Layer.effect(InworldTranscriber, make(cfg)),\n    Layer.effect(\n      Transcriber,\n      Effect.map(\n        make(cfg),\n        (s): TranscriberService => ({\n          transcribe: (req: CommonTranscribeRequest) =>\n            s.transcribe(req as InworldTranscribeRequest),\n          streamTranscriptionFrom: s.streamTranscriptionFrom,\n        }),\n      ),\n    ),\n  )\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;AA+CA,IAAa,qBAAb,cAAwC,QAAQ,SAG7C,CAAC,4DAA4D,CAAC;AAQjE,MAAM,kBAAmC,IAAI,QAAQ,eAAe;CAClE,UAAU;CACV,OAAO;CACP,KAAK;CACN,CAAC;AAEF,MAAM,gBACJ,MAAM,MAAmB,CAAC,KACxB,MAAM,IAAI,UAAU,MAAM,OAAO,QAAQ,SAAS,aAAa,EAAE,MAAM,CAAC,CAAC,EACzE,MAAM,IAAI,WAAW,MAAM,OAAO,QAAQ,EAAE,OAAO,CAAC,EACpD,MAAM,IAAI,aAAa,OAAO,KAAK,gBAAgB,CAAC,EACpD,MAAM,WACP;AAMH,MAAM,iBACJ,WAEA,WAAW,KAAA,IAAY,KAAA,IAAY,OAAO,WAAW,WAAW,CAAC,OAAO,GAAG,OAAO;AAEpF,MAAM,aAAa,YACjB,OAAO,IAAI,aAAa;CACtB,MAAM,UAAU,OAAO,cAAc,QAAQ,MAAM;CACnD,MAAM,UAAU,cAAc,QAAQ,OAAO;AAC7C,QAAO;EACL,kBAAkB;GAChB,SAAS,QAAQ;GACjB,eAAe,QAAQ,iBAAiB;GACxC,GAAI,QAAQ,oBAAoB,KAAA,KAAa,EAC3C,iBAAiB,QAAQ,iBAC1B;GACD,kBAAkB,QAAQ,oBAAoB;GAC9C,GAAI,QAAQ,aAAa,KAAA,KAAa,EAAE,UAAU,QAAQ,UAAU;GACpE,GAAI,YAAY,KAAA,KAAa,EAAE,SAAS;GACxC,GAAI,QAAQ,mBAAmB,QAAQ,EAAE,uBAAuB,MAAM;GACvE;EACD,WAAW,EAAE,SAAS;EACvB;EACD;AAMJ,MAAM,WAAW,OAAO,OAAO;CAC7B,MAAM,OAAO;CACb,aAAa,OAAO,SAAS,OAAO,OAAO;CAC3C,WAAW,OAAO,SAAS,OAAO,OAAO;CACzC,YAAY,OAAO,SAAS,OAAO,OAAO;CAC3C,CAAC;AAEF,MAAM,eAAe,OAAO,OAAO,EACjC,eAAe,OAAO,OAAO;CAC3B,YAAY,OAAO;CACnB,SAAS,OAAO,SAAS,OAAO,QAAQ;CACxC,gBAAgB,OAAO,SAAS,OAAO,OAAO,OAAO,MAAM,SAAS,CAAC,CAAC;CACvE,CAAC,EACH,CAAC;AACF,MAAM,iBAAiB,OAAO,oBAAoB,aAAa;AAE/D,MAAM,oBAAoB,MACxB,EAAE,gBAAgB,KAAA,KAAa,EAAE,cAAc,KAAA,IAC3C,KAAA,IACA;CACE,MAAM,EAAE;CACR,cAAc,EAAE,cAAc;CAC9B,YAAY,EAAE,YAAY;CAC1B,GAAI,EAAE,eAAe,KAAA,KAAa,EAAE,YAAY,EAAE,YAAY;CAC/D;AAMP,MAAM,WAAW,QAAgB,IAAI,WAAW;;AAGhD,MAAa,kBAAkB,SAAiB,YAC9C,OAAO,IAAI,aAAa;CACtB,MAAM,SAAS,OAAO,WAAW;CACjC,MAAM,OAAO,OAAO,UAAU,QAAQ;CACtC,MAAM,cAAc,kBAAkB,KAAK,GAAG,QAAQ,IAAI,CAAC,oBAAoB,CAAC,KAC9E,kBAAkB,UAAU,iBAAiB,WAAW,IAAI,OAAO,CAAC,EACpE,kBAAkB,eAAe,KAAK,CACvC;CACD,MAAM,WAAW,OAAO,OAAO,QAAQ,YAAY,CAAC,KAAK,OAAO,SAAS,iBAAiB,CAAC;AAC3F,KAAI,SAAS,UAAU,KAAK;EAC1B,MAAM,OAAO,OAAO,SAAS,KAAK,KAAK,OAAO,oBAAoB,GAAG,CAAC;AACtE,SAAO,OAAO,OAAO,KAAK,gBAAgB,SAAS,QAAQ,KAAK,CAAC;;CAEnE,MAAM,OAAO,OAAO,SAAS,KAAK,KAAK,OAAO,SAAS,iBAAiB,CAAC;CACzE,MAAM,OAAO,OAAO,eAAe,KAAK,CAAC,KACvC,OAAO,UACJ,UACC,IAAI,QAAQ,iBAAiB;EAC3B,UAAU;EACV,KAAK;GAAE,SAAS;GAAmD;GAAO;GAAM;EACjF,CAAC,CACL,CACF;CACD,MAAM,QAAQ,KAAK,cAAc,gBAC7B,IAAI,iBAAiB,CACtB,QAAQ,MAA0B,MAAM,KAAA,EAAU;AACrD,QAAO;EACL,MAAM,KAAK,cAAc;EACzB,GAAI,UAAU,KAAA,KAAa,MAAM,SAAS,KAAK,EAAE,OAAO;EACxD,KAAK;EACN;EACD;AAMJ,MAAM,qBACJ,UACA,aAC2D;AAS3D,QARqE,OAAO,KAC1E,IAAI,QAAQ,YAAY;EACtB,UAAU;EACV,YAAY;EACZ,QACE;EACH,CAAC,CAEO;;AAOb,MAAa,QACX,QAEA,OAAO,IAAI,WAAW,WAAW,UAAU,GAAG,YAAY;CACxD,aAAa,YACX,eAAe,IAAI,CAAC,QAAQ,CAAC,KAAK,OAAO,eAAe,WAAW,YAAY,OAAO,CAAC;CACzF,yBAAyB;CAC1B,EAAE;;;;;;AAOL,MAAa,SACX,QAEA,MAAM,MACJ,MAAM,OAAO,oBAAoB,KAAK,IAAI,CAAC,EAC3C,MAAM,OACJ,aACA,OAAO,IACL,KAAK,IAAI,GACR,OAA2B;CAC1B,aAAa,QACX,EAAE,WAAW,IAAgC;CAC/C,yBAAyB,EAAE;CAC5B,EACF,CACF,CACF"}
+{"version":3,"file":"InworldTranscriber.mjs","names":[],"sources":["../src/InworldTranscriber.ts"],"sourcesContent":["/**\n * Sync transcription via Inworld's REST endpoint.\n *\n * - `transcribe` → `POST /stt/v1/transcribe` with audio as base64 in\n *   `audioData.content`. URL input is not accepted by Inworld — the\n *   adapter rejects `_tag: \"url\"` audio sources with `InvalidRequest`.\n *\n * Streaming transcription lives at the `InworldRealtimeTranscriber`\n * subpath, which adds the WS path and the `SttStreaming` capability marker.\n */\nimport { Context, Effect, Encoding, Layer, Match, Redacted, Schema, Stream } from \"effect\"\nimport { HttpClient, HttpClientRequest } from \"effect/unstable/http\"\nimport * as AiError from \"@effect-uai/core/AiError\"\nimport type { AudioSource } from \"@effect-uai/core/Audio\"\nimport * as Capabilities from \"@effect-uai/core/Capabilities\"\nimport type { TranscriptEvent, TranscriptResult, WordTimestamp } from \"@effect-uai/core/Transcript\"\nimport {\n  type CommonStreamTranscribeRequest,\n  type CommonTranscribeRequest,\n  Transcriber,\n  type TranscriberService,\n} from \"@effect-uai/core/Transcriber\"\nimport { authHeader, httpStatusError, transportFailure } from \"./codec.js\"\nimport type { InworldSttModel } from \"./models.js\"\n\n// ---------------------------------------------------------------------------\n// Public types\n// ---------------------------------------------------------------------------\n\n/**\n * Inworld-typed transcribe request. `numberOfChannels` defaults to 1.\n * `audioEncoding` is autodetected from the audio MIME by default; pass\n * an explicit value to override (useful for raw PCM bytes).\n */\nexport type InworldTranscribeRequest = Omit<CommonTranscribeRequest, \"model\"> & {\n  readonly model: InworldSttModel\n  readonly audioEncoding?: \"AUTO_DETECT\" | \"LINEAR16\" | \"MP3\" | \"OGG_OPUS\" | \"FLAC\"\n  readonly sampleRateHertz?: number\n  readonly numberOfChannels?: number\n}\n\nexport type InworldTranscriberService = {\n  readonly transcribe: (\n    r: InworldTranscribeRequest,\n  ) => Effect.Effect<TranscriptResult, AiError.AiError>\n  readonly streamTranscriptionFrom: TranscriberService[\"streamTranscriptionFrom\"]\n}\n\nexport class InworldTranscriber extends Context.Service<\n  InworldTranscriber,\n  InworldTranscriberService\n>()(\"@betalyra/effect-uai/providers/inworld/InworldTranscriber\") {}\n\nexport type Config = { readonly apiKey: Redacted.Redacted; readonly baseUrl?: string }\n\n// ---------------------------------------------------------------------------\n// AudioSource → base64. Inworld's sync endpoint takes inline base64 only.\n// ---------------------------------------------------------------------------\n\nconst urlNotSupported: AiError.AiError = new AiError.InvalidRequest({\n  provider: \"inworld\",\n  param: \"audio\",\n  raw: 'Inworld accepts only inline base64. Fetch the URL yourself and pass `{ _tag: \"bytes\", bytes, mimeType }`.',\n})\n\nconst audioToBase64: (audio: AudioSource) => Effect.Effect<string, AiError.AiError> =\n  Match.type<AudioSource>().pipe(\n    Match.tag(\"bytes\", (a) => Effect.succeed(Encoding.encodeBase64(a.bytes))),\n    Match.tag(\"base64\", (a) => Effect.succeed(a.base64)),\n    Match.tag(\"url\", () => Effect.fail(urlNotSupported)),\n    Match.exhaustive,\n  )\n\n// ---------------------------------------------------------------------------\n// Request body\n// ---------------------------------------------------------------------------\n\nconst buildBody = (request: InworldTranscribeRequest) =>\n  Effect.gen(function* () {\n    const content = yield* audioToBase64(request.audio)\n    // Inworld's `prompts` is a vocab-biasing term list — maps from\n    // `biasingTerms`. There's no free-form prompt field, so `prompt` warns.\n    yield* Capabilities.warnDroppedWhen(request.prompt, {\n      provider: \"inworld\",\n      capability: \"prompt\",\n      field: \"prompt\",\n      reason: \"Inworld STT has no free-form prompt field; bias via `biasingTerms`.\",\n    })\n    return {\n      transcribeConfig: {\n        modelId: request.model,\n        audioEncoding: request.audioEncoding ?? \"AUTO_DETECT\",\n        ...(request.sampleRateHertz !== undefined && {\n          sampleRateHertz: request.sampleRateHertz,\n        }),\n        numberOfChannels: request.numberOfChannels ?? 1,\n        ...(request.language !== undefined && { language: request.language }),\n        ...(request.biasingTerms !== undefined && { prompts: request.biasingTerms }),\n        ...(request.wordTimestamps === true && { includeWordTimestamps: true }),\n      },\n      audioData: { content },\n    }\n  })\n\n// ---------------------------------------------------------------------------\n// Wire schema\n// ---------------------------------------------------------------------------\n\nconst WireWord = Schema.Struct({\n  word: Schema.String,\n  startTimeMs: Schema.optional(Schema.Number),\n  endTimeMs: Schema.optional(Schema.Number),\n  confidence: Schema.optional(Schema.Number),\n})\n\nconst WireResponse = Schema.Struct({\n  transcription: Schema.Struct({\n    transcript: Schema.String,\n    isFinal: Schema.optional(Schema.Boolean),\n    wordTimestamps: Schema.optional(Schema.NullOr(Schema.Array(WireWord))),\n  }),\n})\nconst decodeResponse = Schema.decodeUnknownEffect(WireResponse)\n\nconst wireWordToCommon = (w: typeof WireWord.Type): WordTimestamp | undefined =>\n  w.startTimeMs === undefined || w.endTimeMs === undefined\n    ? undefined\n    : {\n        text: w.word,\n        startSeconds: w.startTimeMs / 1000,\n        endSeconds: w.endTimeMs / 1000,\n        ...(w.confidence !== undefined && { confidence: w.confidence }),\n      }\n\n// ---------------------------------------------------------------------------\n// HTTP plumbing\n// ---------------------------------------------------------------------------\n\nconst baseUrl = (cfg: Config) => cfg.baseUrl ?? \"https://api.inworld.ai\"\n\n/** Exported for reuse by `InworldRealtimeTranscriber`. */\nexport const transcribeImpl = (cfg: Config) => (request: InworldTranscribeRequest) =>\n  Effect.gen(function* () {\n    const client = yield* HttpClient.HttpClient\n    const body = yield* buildBody(request)\n    const httpRequest = HttpClientRequest.post(`${baseUrl(cfg)}/stt/v1/transcribe`).pipe(\n      HttpClientRequest.setHeader(\"Authorization\", authHeader(cfg.apiKey)),\n      HttpClientRequest.bodyJsonUnsafe(body),\n    )\n    const response = yield* client.execute(httpRequest).pipe(Effect.mapError(transportFailure))\n    if (response.status >= 400) {\n      const text = yield* response.text.pipe(Effect.orElseSucceed(() => \"\"))\n      return yield* Effect.fail(httpStatusError(response.status, text))\n    }\n    const json = yield* response.json.pipe(Effect.mapError(transportFailure))\n    const wire = yield* decodeResponse(json).pipe(\n      Effect.mapError(\n        (cause) =>\n          new AiError.GenerationFailed({\n            provider: \"inworld\",\n            raw: { message: \"STT response missing `transcription.transcript`\", cause, json },\n          }),\n      ),\n    )\n    const words = wire.transcription.wordTimestamps\n      ?.map(wireWordToCommon)\n      .filter((w): w is WordTimestamp => w !== undefined)\n    return {\n      text: wire.transcription.transcript,\n      ...(words !== undefined && words.length > 0 && { words }),\n      raw: json,\n    } satisfies TranscriptResult\n  })\n\n// ---------------------------------------------------------------------------\n// streamTranscriptionFrom Unsupported on this Layer\n// ---------------------------------------------------------------------------\n\nconst streamUnsupported = <E, R>(\n  _audioIn: Stream.Stream<Uint8Array, E, R>,\n  _request: CommonStreamTranscribeRequest,\n): Stream.Stream<TranscriptEvent, AiError.AiError | E, R> => {\n  const fail: Stream.Stream<TranscriptEvent, AiError.AiError | E, R> = Stream.fail(\n    new AiError.Unsupported({\n      provider: \"inworld\",\n      capability: \"streamTranscriptionFrom\",\n      reason:\n        \"This Layer is sync-only. Import `@effect-uai/inworld/InworldRealtimeTranscriber` to use the WS streamTranscriptionFrom path.\",\n    }),\n  )\n  return fail\n}\n\n// ---------------------------------------------------------------------------\n// Constructors\n// ---------------------------------------------------------------------------\n\nexport const make = (\n  cfg: Config,\n): Effect.Effect<InworldTranscriberService, never, HttpClient.HttpClient> =>\n  Effect.map(HttpClient.HttpClient.asEffect(), (client) => ({\n    transcribe: (request) =>\n      transcribeImpl(cfg)(request).pipe(Effect.provideService(HttpClient.HttpClient, client)),\n    streamTranscriptionFrom: streamUnsupported,\n  }))\n\n/**\n * Sync-only Layer. Registers `InworldTranscriber` + the generic\n * `Transcriber`. Does **not** register `SttStreaming` — for live\n * transcription use `@effect-uai/inworld/InworldRealtimeTranscriber`.\n */\nexport const layer = (\n  cfg: Config,\n): Layer.Layer<InworldTranscriber | Transcriber, never, HttpClient.HttpClient> =>\n  Layer.merge(\n    Layer.effect(InworldTranscriber, make(cfg)),\n    Layer.effect(\n      Transcriber,\n      Effect.map(\n        make(cfg),\n        (s): TranscriberService => ({\n          transcribe: (req: CommonTranscribeRequest) =>\n            s.transcribe(req as InworldTranscribeRequest),\n          streamTranscriptionFrom: s.streamTranscriptionFrom,\n        }),\n      ),\n    ),\n  )\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;AAgDA,IAAa,qBAAb,cAAwC,QAAQ,SAG7C,CAAC,4DAA4D,CAAC;AAQjE,MAAM,kBAAmC,IAAI,QAAQ,eAAe;CAClE,UAAU;CACV,OAAO;CACP,KAAK;CACN,CAAC;AAEF,MAAM,gBACJ,MAAM,MAAmB,CAAC,KACxB,MAAM,IAAI,UAAU,MAAM,OAAO,QAAQ,SAAS,aAAa,EAAE,MAAM,CAAC,CAAC,EACzE,MAAM,IAAI,WAAW,MAAM,OAAO,QAAQ,EAAE,OAAO,CAAC,EACpD,MAAM,IAAI,aAAa,OAAO,KAAK,gBAAgB,CAAC,EACpD,MAAM,WACP;AAMH,MAAM,aAAa,YACjB,OAAO,IAAI,aAAa;CACtB,MAAM,UAAU,OAAO,cAAc,QAAQ,MAAM;AAGnD,QAAO,aAAa,gBAAgB,QAAQ,QAAQ;EAClD,UAAU;EACV,YAAY;EACZ,OAAO;EACP,QAAQ;EACT,CAAC;AACF,QAAO;EACL,kBAAkB;GAChB,SAAS,QAAQ;GACjB,eAAe,QAAQ,iBAAiB;GACxC,GAAI,QAAQ,oBAAoB,KAAA,KAAa,EAC3C,iBAAiB,QAAQ,iBAC1B;GACD,kBAAkB,QAAQ,oBAAoB;GAC9C,GAAI,QAAQ,aAAa,KAAA,KAAa,EAAE,UAAU,QAAQ,UAAU;GACpE,GAAI,QAAQ,iBAAiB,KAAA,KAAa,EAAE,SAAS,QAAQ,cAAc;GAC3E,GAAI,QAAQ,mBAAmB,QAAQ,EAAE,uBAAuB,MAAM;GACvE;EACD,WAAW,EAAE,SAAS;EACvB;EACD;AAMJ,MAAM,WAAW,OAAO,OAAO;CAC7B,MAAM,OAAO;CACb,aAAa,OAAO,SAAS,OAAO,OAAO;CAC3C,WAAW,OAAO,SAAS,OAAO,OAAO;CACzC,YAAY,OAAO,SAAS,OAAO,OAAO;CAC3C,CAAC;AAEF,MAAM,eAAe,OAAO,OAAO,EACjC,eAAe,OAAO,OAAO;CAC3B,YAAY,OAAO;CACnB,SAAS,OAAO,SAAS,OAAO,QAAQ;CACxC,gBAAgB,OAAO,SAAS,OAAO,OAAO,OAAO,MAAM,SAAS,CAAC,CAAC;CACvE,CAAC,EACH,CAAC;AACF,MAAM,iBAAiB,OAAO,oBAAoB,aAAa;AAE/D,MAAM,oBAAoB,MACxB,EAAE,gBAAgB,KAAA,KAAa,EAAE,cAAc,KAAA,IAC3C,KAAA,IACA;CACE,MAAM,EAAE;CACR,cAAc,EAAE,cAAc;CAC9B,YAAY,EAAE,YAAY;CAC1B,GAAI,EAAE,eAAe,KAAA,KAAa,EAAE,YAAY,EAAE,YAAY;CAC/D;AAMP,MAAM,WAAW,QAAgB,IAAI,WAAW;;AAGhD,MAAa,kBAAkB,SAAiB,YAC9C,OAAO,IAAI,aAAa;CACtB,MAAM,SAAS,OAAO,WAAW;CACjC,MAAM,OAAO,OAAO,UAAU,QAAQ;CACtC,MAAM,cAAc,kBAAkB,KAAK,GAAG,QAAQ,IAAI,CAAC,oBAAoB,CAAC,KAC9E,kBAAkB,UAAU,iBAAiB,WAAW,IAAI,OAAO,CAAC,EACpE,kBAAkB,eAAe,KAAK,CACvC;CACD,MAAM,WAAW,OAAO,OAAO,QAAQ,YAAY,CAAC,KAAK,OAAO,SAAS,iBAAiB,CAAC;AAC3F,KAAI,SAAS,UAAU,KAAK;EAC1B,MAAM,OAAO,OAAO,SAAS,KAAK,KAAK,OAAO,oBAAoB,GAAG,CAAC;AACtE,SAAO,OAAO,OAAO,KAAK,gBAAgB,SAAS,QAAQ,KAAK,CAAC;;CAEnE,MAAM,OAAO,OAAO,SAAS,KAAK,KAAK,OAAO,SAAS,iBAAiB,CAAC;CACzE,MAAM,OAAO,OAAO,eAAe,KAAK,CAAC,KACvC,OAAO,UACJ,UACC,IAAI,QAAQ,iBAAiB;EAC3B,UAAU;EACV,KAAK;GAAE,SAAS;GAAmD;GAAO;GAAM;EACjF,CAAC,CACL,CACF;CACD,MAAM,QAAQ,KAAK,cAAc,gBAC7B,IAAI,iBAAiB,CACtB,QAAQ,MAA0B,MAAM,KAAA,EAAU;AACrD,QAAO;EACL,MAAM,KAAK,cAAc;EACzB,GAAI,UAAU,KAAA,KAAa,MAAM,SAAS,KAAK,EAAE,OAAO;EACxD,KAAK;EACN;EACD;AAMJ,MAAM,qBACJ,UACA,aAC2D;AAS3D,QARqE,OAAO,KAC1E,IAAI,QAAQ,YAAY;EACtB,UAAU;EACV,YAAY;EACZ,QACE;EACH,CAAC,CAEO;;AAOb,MAAa,QACX,QAEA,OAAO,IAAI,WAAW,WAAW,UAAU,GAAG,YAAY;CACxD,aAAa,YACX,eAAe,IAAI,CAAC,QAAQ,CAAC,KAAK,OAAO,eAAe,WAAW,YAAY,OAAO,CAAC;CACzF,yBAAyB;CAC1B,EAAE;;;;;;AAOL,MAAa,SACX,QAEA,MAAM,MACJ,MAAM,OAAO,oBAAoB,KAAK,IAAI,CAAC,EAC3C,MAAM,OACJ,aACA,OAAO,IACL,KAAK,IAAI,GACR,OAA2B;CAC1B,aAAa,QACX,EAAE,WAAW,IAAgC;CAC/C,yBAAyB,EAAE;CAC5B,EACF,CACF,CACF"}

package/dist/realtimeStt.d.mts.map

@@ -1 +1 @@
-{"version":3,"file":"realtimeStt.d.mts","names":[],"sources":["../src/realtimeStt.ts"],"mappings":";;;;;;KA6BY,MAAA;EAAA,SAAoB,MAAA,EAAQ,QAAA,CAAS,QAAA;EAAA,SAAmB,OAAA;AAAA;AAAA,cAqG9D,WAAA,EAAW,MAAA,CAAA,MAAA;EAAA;;;;;;;;;;;;;;;;;;;;;;cAgBJ,WAAA,GAAe,KAAA,SAAc,WAAA,CAAY,IAAA,KAAO,eAAA;AAAA,cAoDhD,mBAAA,GACV,GAAA,EAAK,MAAA,YAEJ,OAAA,EAAS,MAAA,CAAO,MAAA,CAAO,UAAA,EAAY,CAAA,EAAG,CAAA,GACtC,OAAA,EAAS,6BAAA,KACR,MAAA,CAAO,MAAA,CAAO,eAAA,EAAiB,OAAA,CAAQ,OAAA,GAAU,CAAA,EAAG,CAAA"}
+{"version":3,"file":"realtimeStt.d.mts","names":[],"sources":["../src/realtimeStt.ts"],"mappings":";;;;;;KA8BY,MAAA;EAAA,SAAoB,MAAA,EAAQ,QAAA,CAAS,QAAA;EAAA,SAAmB,OAAA;AAAA;AAAA,cA+F9D,WAAA,EAAW,MAAA,CAAA,MAAA;EAAA;;;;;;;;;;;;;;;;;;;;;;cAgBJ,WAAA,GAAe,KAAA,SAAc,WAAA,CAAY,IAAA,KAAO,eAAA;AAAA,cAoDhD,mBAAA,GACV,GAAA,EAAK,MAAA,YAEJ,OAAA,EAAS,MAAA,CAAO,MAAA,CAAO,UAAA,EAAY,CAAA,EAAG,CAAA,GACtC,OAAA,EAAS,6BAAA,KACR,MAAA,CAAO,MAAA,CAAO,eAAA,EAAiB,OAAA,CAAQ,OAAA,GAAU,CAAA,EAAG,CAAA"}

package/dist/realtimeStt.mjs

@@ -3,6 +3,7 @@ import { Effect, Encoding, Match, Queue, Schema, Stream } from "effect";
 import * as AiError from "@effect-uai/core/AiError";
 import * as JSONL from "@effect-uai/core/JSONL";
 import * as Socket from "effect/unstable/socket/Socket";
+import * as Capabilities from "@effect-uai/core/Capabilities";
 //#region src/realtimeStt.ts
 /**
 * Inworld Realtime STT — `wss://api.inworld.ai/stt/v1/transcribe:streamBidirectional`.
@@ -42,19 +43,15 @@ const inputFormatToWire = Match.type().pipe(Match.when({
 const buildWsUrl = (cfg) => {
 	return `${(cfg.baseUrl ?? "https://api.inworld.ai").replace(/^http/, "ws")}/stt/v1/transcribe:streamBidirectional`;
 };
-const promptToTerms = (prompt) => prompt === void 0 ? void 0 : typeof prompt === "string" ? [prompt] : prompt.terms;
-const configFrame = (encoding, request) => {
-	const prompts = promptToTerms(request.prompt);
-	return JSON.stringify({ transcribeConfig: {
-		modelId: request.model,
-		audioEncoding: encoding,
-		sampleRateHertz: request.inputFormat.sampleRate,
-		numberOfChannels: request.inputFormat.channels ?? 1,
-		language: request.language ?? "en-US",
-		...prompts !== void 0 && { prompts },
-		...request.wordTimestamps === true && { includeWordTimestamps: true }
-	} });
-};
+const configFrame = (encoding, request) => JSON.stringify({ transcribeConfig: {
+	modelId: request.model,
+	audioEncoding: encoding,
+	sampleRateHertz: request.inputFormat.sampleRate,
+	numberOfChannels: request.inputFormat.channels ?? 1,
+	language: request.language ?? "en-US",
+	...request.biasingTerms !== void 0 && { prompts: request.biasingTerms },
+	...request.wordTimestamps === true && { includeWordTimestamps: true }
+} });
 const audioChunkFrame = (bytes) => JSON.stringify({ audioChunk: { content: Encoding.encodeBase64(bytes) } });
 const endTurnFrame = JSON.stringify({ endTurn: {} });
 const closeStreamFrame = JSON.stringify({ closeStream: {} });
@@ -126,6 +123,12 @@ const handleServerMessage = (queue) => (raw) => Effect.gen(function* () {
 	if (event !== void 0) yield* Queue.offer(queue, event);
 });
 const streamTranscription = (cfg) => (audioIn, request) => Stream.unwrap(Effect.gen(function* () {
+	yield* Capabilities.warnDroppedWhen(request.prompt, {
+		provider: "inworld",
+		capability: "prompt",
+		field: "prompt",
+		reason: "Inworld STT has no free-form prompt field; bias via `biasingTerms`."
+	});
 	const encoding = yield* inputFormatToWire(request.inputFormat);
 	const socket = yield* Socket.makeWebSocket(buildWsUrl(cfg), { closeCodeIsError: (code) => code !== 1e3 && code !== 1001 && code !== 1005 }).pipe(Effect.provideService(Socket.WebSocketConstructor, authedWsConstructor(cfg.apiKey)));
 	const queue = yield* Queue.bounded(64);

package/dist/realtimeStt.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"realtimeStt.mjs","names":[],"sources":["../src/realtimeStt.ts"],"sourcesContent":["/**\n * Inworld Realtime STT — `wss://api.inworld.ai/stt/v1/transcribe:streamBidirectional`.\n *\n * Auth: `Authorization: Basic <API_KEY>` on the WS upgrade header (matches\n * Inworld's own JS sample). Setting headers needs the `ws` peer dep\n * (Node/Bun); see `./wsAuth.ts`.\n *\n * Wire shape (per [inworld-ai/inworld-api-examples](https://github.com/inworld-ai/inworld-api-examples)):\n *   client → server:\n *     `{ \"transcribeConfig\": { modelId, audioEncoding, sampleRateHertz, language, ... } }`  (first frame)\n *     `{ \"audioChunk\": { \"content\": \"<base64 audio>\" } }`                                  (×N)\n *     `{ \"closeStream\": {} }`                                                              (end-of-input)\n *   server → client:\n *     `{ \"result\": { \"transcription\": { \"transcript\": \"...\", \"isFinal\": true|false } } }`\n *     `{ \"result\": { \"speechStarted\": { \"startTimeMs\": ... } } }`\n *     `{ \"result\": { \"speechStopped\": { \"silenceDurationMs\": ... } } }`\n *\n * Inworld's STT WS sends audio as base64 inside JSON (NOT binary frames),\n * matching the rest of the Inworld API style.\n */\nimport { Cause, Effect, Encoding, Match, Queue, Redacted, Schema, Stream } from \"effect\"\nimport * as Socket from \"effect/unstable/socket/Socket\"\nimport * as AiError from \"@effect-uai/core/AiError\"\nimport type { AudioFormat } from \"@effect-uai/core/Audio\"\nimport * as JSONL from \"@effect-uai/core/JSONL\"\nimport type { TranscriptEvent, WordTimestamp } from \"@effect-uai/core/Transcript\"\nimport type { CommonStreamTranscribeRequest } from \"@effect-uai/core/Transcriber\"\nimport { authedWsConstructor } from \"./wsAuth.js\"\n\nexport type Config = { readonly apiKey: Redacted.Redacted; readonly baseUrl?: string }\n\n// ---------------------------------------------------------------------------\n// AudioFormat → `audioEncoding` slug for Inworld STT\n// ---------------------------------------------------------------------------\n\ntype WireEncoding = \"LINEAR16\" | \"MP3\" | \"OGG_OPUS\" | \"FLAC\"\n\nconst unsupportedFormat = (format: AudioFormat) =>\n  new AiError.Unsupported({\n    provider: \"inworld\",\n    capability: \"inputFormat\",\n    reason: `Inworld realtime STT accepts pcm_s16le (LINEAR16), mp3, ogg/opus, or flac. Got ${JSON.stringify(format)}.`,\n  })\n\nconst inputFormatToWire: (format: AudioFormat) => Effect.Effect<WireEncoding, AiError.AiError> =\n  Match.type<AudioFormat>().pipe(\n    Match.when({ container: \"raw\", encoding: \"pcm_s16le\" }, () =>\n      Effect.succeed<WireEncoding>(\"LINEAR16\"),\n    ),\n    Match.when({ container: \"wav\", encoding: \"pcm_s16le\" }, () =>\n      Effect.succeed<WireEncoding>(\"LINEAR16\"),\n    ),\n    Match.when({ container: \"mp3\" }, () => Effect.succeed<WireEncoding>(\"MP3\")),\n    Match.when({ container: \"ogg\", encoding: \"opus\" }, () =>\n      Effect.succeed<WireEncoding>(\"OGG_OPUS\"),\n    ),\n    Match.when({ container: \"flac\" }, () => Effect.succeed<WireEncoding>(\"FLAC\")),\n    Match.orElse((f) => Effect.fail(unsupportedFormat(f))),\n  )\n\n// ---------------------------------------------------------------------------\n// URL + frame builders\n// ---------------------------------------------------------------------------\n\nconst buildWsUrl = (cfg: Config) => {\n  const wsBase = (cfg.baseUrl ?? \"https://api.inworld.ai\").replace(/^http/, \"ws\")\n  return `${wsBase}/stt/v1/transcribe:streamBidirectional`\n}\n\nconst promptToTerms = (\n  prompt: string | { readonly terms: ReadonlyArray<string> } | undefined,\n): ReadonlyArray<string> | undefined =>\n  prompt === undefined ? undefined : typeof prompt === \"string\" ? [prompt] : prompt.terms\n\nconst configFrame = (encoding: WireEncoding, request: CommonStreamTranscribeRequest) => {\n  const prompts = promptToTerms(request.prompt)\n  return JSON.stringify({\n    transcribeConfig: {\n      modelId: request.model,\n      audioEncoding: encoding,\n      sampleRateHertz: request.inputFormat.sampleRate,\n      numberOfChannels: request.inputFormat.channels ?? 1,\n      // Inworld's sample includes `language` even though docs mark it optional.\n      // Default to en-US to match the sample's behavior; caller can override.\n      language: request.language ?? \"en-US\",\n      ...(prompts !== undefined && { prompts }),\n      ...(request.wordTimestamps === true && { includeWordTimestamps: true }),\n    },\n  })\n}\n\nconst audioChunkFrame = (bytes: Uint8Array) =>\n  JSON.stringify({ audioChunk: { content: Encoding.encodeBase64(bytes) } })\n\nconst endTurnFrame = JSON.stringify({ endTurn: {} })\nconst closeStreamFrame = JSON.stringify({ closeStream: {} })\n\n// ---------------------------------------------------------------------------\n// Wire schemas (server → client)\n// ---------------------------------------------------------------------------\n\nconst WireWord = Schema.Struct({\n  word: Schema.String,\n  startTimeMs: Schema.optional(Schema.Number),\n  endTimeMs: Schema.optional(Schema.Number),\n  confidence: Schema.optional(Schema.Number),\n})\n\nconst ResultBody = Schema.Struct({\n  transcription: Schema.optional(\n    Schema.Struct({\n      transcript: Schema.String,\n      isFinal: Schema.optional(Schema.Boolean),\n      wordTimestamps: Schema.optional(Schema.NullOr(Schema.Array(WireWord))),\n    }),\n  ),\n  speechStarted: Schema.optional(\n    Schema.Struct({\n      startTimeMs: Schema.optional(Schema.Number),\n      confidence: Schema.optional(Schema.Number),\n    }),\n  ),\n  speechStopped: Schema.optional(\n    Schema.Struct({\n      silenceDurationMs: Schema.optional(Schema.Number),\n    }),\n  ),\n  status: Schema.optional(Schema.Unknown),\n})\n\nconst ServerFrame = Schema.Struct({\n  result: Schema.optional(ResultBody),\n  error: Schema.optional(Schema.Unknown),\n})\nconst decodeServerFrame = Schema.decodeUnknownEffect(ServerFrame)\n\nconst wireWordToCommon = (w: typeof WireWord.Type): WordTimestamp | undefined =>\n  w.startTimeMs === undefined || w.endTimeMs === undefined\n    ? undefined\n    : {\n        text: w.word,\n        startSeconds: w.startTimeMs / 1000,\n        endSeconds: w.endTimeMs / 1000,\n        ...(w.confidence !== undefined && { confidence: w.confidence }),\n      }\n\nexport const wireToEvent = (frame: typeof ServerFrame.Type): TranscriptEvent | undefined => {\n  if (frame.error !== undefined) {\n    return {\n      _tag: \"error\",\n      message: typeof frame.error === \"string\" ? frame.error : JSON.stringify(frame.error),\n    }\n  }\n  const result = frame.result\n  if (result === undefined) return undefined\n  if (result.transcription !== undefined) {\n    const t = result.transcription\n    const words = t.wordTimestamps\n      ?.map(wireWordToCommon)\n      .filter((w): w is WordTimestamp => w !== undefined)\n    return t.isFinal === true\n      ? {\n          _tag: \"final\",\n          text: t.transcript,\n          ...(words !== undefined && words.length > 0 && { words }),\n        }\n      : {\n          _tag: \"partial\",\n          text: t.transcript,\n          ...(words !== undefined && words.length > 0 && { words }),\n        }\n  }\n  if (result.speechStarted !== undefined) {\n    return {\n      _tag: \"speech-started\",\n      atSeconds: (result.speechStarted.startTimeMs ?? 0) / 1000,\n    }\n  }\n  if (result.speechStopped !== undefined) {\n    return { _tag: \"utterance-ended\", atSeconds: 0 }\n  }\n  return undefined\n}\n\nconst handleServerMessage = (queue: Queue.Queue<TranscriptEvent, Cause.Done>) => (raw: string) =>\n  Effect.gen(function* () {\n    const json = yield* JSONL.parseSafe(raw)\n    if (json === undefined) return\n    const decoded = yield* decodeServerFrame(json).pipe(Effect.option)\n    if (decoded._tag === \"None\") return\n    const event = wireToEvent(decoded.value)\n    if (event !== undefined) yield* Queue.offer(queue, event)\n  })\n\n// ---------------------------------------------------------------------------\n// Stream<Uint8Array> → Stream<TranscriptEvent>\n// ---------------------------------------------------------------------------\n\nexport const streamTranscription =\n  (cfg: Config) =>\n  <E, R>(\n    audioIn: Stream.Stream<Uint8Array, E, R>,\n    request: CommonStreamTranscribeRequest,\n  ): Stream.Stream<TranscriptEvent, AiError.AiError | E, R> =>\n    Stream.unwrap(\n      Effect.gen(function* () {\n        const encoding = yield* inputFormatToWire(request.inputFormat)\n        const socket = yield* Socket.makeWebSocket(buildWsUrl(cfg), {\n          // Effect's Socket treats all close codes as errors by default —\n          // whitelist standard clean-close codes (1000 / 1001 / 1005).\n          closeCodeIsError: (code) => code !== 1000 && code !== 1001 && code !== 1005,\n        }).pipe(Effect.provideService(Socket.WebSocketConstructor, authedWsConstructor(cfg.apiKey)))\n        const queue = yield* Queue.bounded<TranscriptEvent, Cause.Done>(64)\n        const write = yield* socket.writer\n\n        // Writer: config → drain audio → endTurn → closeStream. Inworld\n        // emits `result.transcription` partials in real-time via VAD;\n        // `endTurn` is sent on input-stream end to flush any tail audio\n        // into a final transcript, then `closeStream` for graceful close.\n        yield* Effect.gen(function* () {\n          yield* write(configFrame(encoding, request))\n          yield* Stream.runForEach(audioIn, (bytes) => write(audioChunkFrame(bytes)))\n          yield* write(endTurnFrame)\n          yield* write(closeStreamFrame)\n        }).pipe(Effect.ignore, Effect.forkScoped)\n\n        // `Queue.end` flushes pending events then fails the next take with\n        // `Done` — clean stream end. `Queue.shutdown` would CLEAR queued\n        // items and interrupt pending takes (wrong for graceful teardown).\n        yield* socket\n          .runString(handleServerMessage(queue))\n          .pipe(Effect.ensuring(Queue.end(queue)), Effect.forkScoped)\n\n        return Stream.fromQueue(queue)\n      }),\n    )\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;AAqCA,MAAM,qBAAqB,WACzB,IAAI,QAAQ,YAAY;CACtB,UAAU;CACV,YAAY;CACZ,QAAQ,kFAAkF,KAAK,UAAU,OAAO,CAAC;CAClH,CAAC;AAEJ,MAAM,oBACJ,MAAM,MAAmB,CAAC,KACxB,MAAM,KAAK;CAAE,WAAW;CAAO,UAAU;CAAa,QACpD,OAAO,QAAsB,WAAW,CACzC,EACD,MAAM,KAAK;CAAE,WAAW;CAAO,UAAU;CAAa,QACpD,OAAO,QAAsB,WAAW,CACzC,EACD,MAAM,KAAK,EAAE,WAAW,OAAO,QAAQ,OAAO,QAAsB,MAAM,CAAC,EAC3E,MAAM,KAAK;CAAE,WAAW;CAAO,UAAU;CAAQ,QAC/C,OAAO,QAAsB,WAAW,CACzC,EACD,MAAM,KAAK,EAAE,WAAW,QAAQ,QAAQ,OAAO,QAAsB,OAAO,CAAC,EAC7E,MAAM,QAAQ,MAAM,OAAO,KAAK,kBAAkB,EAAE,CAAC,CAAC,CACvD;AAMH,MAAM,cAAc,QAAgB;AAElC,QAAO,IADS,IAAI,WAAW,0BAA0B,QAAQ,SAAS,KAC1D,CAAC;;AAGnB,MAAM,iBACJ,WAEA,WAAW,KAAA,IAAY,KAAA,IAAY,OAAO,WAAW,WAAW,CAAC,OAAO,GAAG,OAAO;AAEpF,MAAM,eAAe,UAAwB,YAA2C;CACtF,MAAM,UAAU,cAAc,QAAQ,OAAO;AAC7C,QAAO,KAAK,UAAU,EACpB,kBAAkB;EAChB,SAAS,QAAQ;EACjB,eAAe;EACf,iBAAiB,QAAQ,YAAY;EACrC,kBAAkB,QAAQ,YAAY,YAAY;EAGlD,UAAU,QAAQ,YAAY;EAC9B,GAAI,YAAY,KAAA,KAAa,EAAE,SAAS;EACxC,GAAI,QAAQ,mBAAmB,QAAQ,EAAE,uBAAuB,MAAM;EACvE,EACF,CAAC;;AAGJ,MAAM,mBAAmB,UACvB,KAAK,UAAU,EAAE,YAAY,EAAE,SAAS,SAAS,aAAa,MAAM,EAAE,EAAE,CAAC;AAE3E,MAAM,eAAe,KAAK,UAAU,EAAE,SAAS,EAAE,EAAE,CAAC;AACpD,MAAM,mBAAmB,KAAK,UAAU,EAAE,aAAa,EAAE,EAAE,CAAC;AAM5D,MAAM,WAAW,OAAO,OAAO;CAC7B,MAAM,OAAO;CACb,aAAa,OAAO,SAAS,OAAO,OAAO;CAC3C,WAAW,OAAO,SAAS,OAAO,OAAO;CACzC,YAAY,OAAO,SAAS,OAAO,OAAO;CAC3C,CAAC;AAEF,MAAM,aAAa,OAAO,OAAO;CAC/B,eAAe,OAAO,SACpB,OAAO,OAAO;EACZ,YAAY,OAAO;EACnB,SAAS,OAAO,SAAS,OAAO,QAAQ;EACxC,gBAAgB,OAAO,SAAS,OAAO,OAAO,OAAO,MAAM,SAAS,CAAC,CAAC;EACvE,CAAC,CACH;CACD,eAAe,OAAO,SACpB,OAAO,OAAO;EACZ,aAAa,OAAO,SAAS,OAAO,OAAO;EAC3C,YAAY,OAAO,SAAS,OAAO,OAAO;EAC3C,CAAC,CACH;CACD,eAAe,OAAO,SACpB,OAAO,OAAO,EACZ,mBAAmB,OAAO,SAAS,OAAO,OAAO,EAClD,CAAC,CACH;CACD,QAAQ,OAAO,SAAS,OAAO,QAAQ;CACxC,CAAC;AAEF,MAAM,cAAc,OAAO,OAAO;CAChC,QAAQ,OAAO,SAAS,WAAW;CACnC,OAAO,OAAO,SAAS,OAAO,QAAQ;CACvC,CAAC;AACF,MAAM,oBAAoB,OAAO,oBAAoB,YAAY;AAEjE,MAAM,oBAAoB,MACxB,EAAE,gBAAgB,KAAA,KAAa,EAAE,cAAc,KAAA,IAC3C,KAAA,IACA;CACE,MAAM,EAAE;CACR,cAAc,EAAE,cAAc;CAC9B,YAAY,EAAE,YAAY;CAC1B,GAAI,EAAE,eAAe,KAAA,KAAa,EAAE,YAAY,EAAE,YAAY;CAC/D;AAEP,MAAa,eAAe,UAAgE;AAC1F,KAAI,MAAM,UAAU,KAAA,EAClB,QAAO;EACL,MAAM;EACN,SAAS,OAAO,MAAM,UAAU,WAAW,MAAM,QAAQ,KAAK,UAAU,MAAM,MAAM;EACrF;CAEH,MAAM,SAAS,MAAM;AACrB,KAAI,WAAW,KAAA,EAAW,QAAO,KAAA;AACjC,KAAI,OAAO,kBAAkB,KAAA,GAAW;EACtC,MAAM,IAAI,OAAO;EACjB,MAAM,QAAQ,EAAE,gBACZ,IAAI,iBAAiB,CACtB,QAAQ,MAA0B,MAAM,KAAA,EAAU;AACrD,SAAO,EAAE,YAAY,OACjB;GACE,MAAM;GACN,MAAM,EAAE;GACR,GAAI,UAAU,KAAA,KAAa,MAAM,SAAS,KAAK,EAAE,OAAO;GACzD,GACD;GACE,MAAM;GACN,MAAM,EAAE;GACR,GAAI,UAAU,KAAA,KAAa,MAAM,SAAS,KAAK,EAAE,OAAO;GACzD;;AAEP,KAAI,OAAO,kBAAkB,KAAA,EAC3B,QAAO;EACL,MAAM;EACN,YAAY,OAAO,cAAc,eAAe,KAAK;EACtD;AAEH,KAAI,OAAO,kBAAkB,KAAA,EAC3B,QAAO;EAAE,MAAM;EAAmB,WAAW;EAAG;;AAKpD,MAAM,uBAAuB,WAAqD,QAChF,OAAO,IAAI,aAAa;CACtB,MAAM,OAAO,OAAO,MAAM,UAAU,IAAI;AACxC,KAAI,SAAS,KAAA,EAAW;CACxB,MAAM,UAAU,OAAO,kBAAkB,KAAK,CAAC,KAAK,OAAO,OAAO;AAClE,KAAI,QAAQ,SAAS,OAAQ;CAC7B,MAAM,QAAQ,YAAY,QAAQ,MAAM;AACxC,KAAI,UAAU,KAAA,EAAW,QAAO,MAAM,MAAM,OAAO,MAAM;EACzD;AAMJ,MAAa,uBACV,SAEC,SACA,YAEA,OAAO,OACL,OAAO,IAAI,aAAa;CACtB,MAAM,WAAW,OAAO,kBAAkB,QAAQ,YAAY;CAC9D,MAAM,SAAS,OAAO,OAAO,cAAc,WAAW,IAAI,EAAE,EAG1D,mBAAmB,SAAS,SAAS,OAAQ,SAAS,QAAQ,SAAS,MACxE,CAAC,CAAC,KAAK,OAAO,eAAe,OAAO,sBAAsB,oBAAoB,IAAI,OAAO,CAAC,CAAC;CAC5F,MAAM,QAAQ,OAAO,MAAM,QAAqC,GAAG;CACnE,MAAM,QAAQ,OAAO,OAAO;AAM5B,QAAO,OAAO,IAAI,aAAa;AAC7B,SAAO,MAAM,YAAY,UAAU,QAAQ,CAAC;AAC5C,SAAO,OAAO,WAAW,UAAU,UAAU,MAAM,gBAAgB,MAAM,CAAC,CAAC;AAC3E,SAAO,MAAM,aAAa;AAC1B,SAAO,MAAM,iBAAiB;GAC9B,CAAC,KAAK,OAAO,QAAQ,OAAO,WAAW;AAKzC,QAAO,OACJ,UAAU,oBAAoB,MAAM,CAAC,CACrC,KAAK,OAAO,SAAS,MAAM,IAAI,MAAM,CAAC,EAAE,OAAO,WAAW;AAE7D,QAAO,OAAO,UAAU,MAAM;EAC9B,CACH"}
+{"version":3,"file":"realtimeStt.mjs","names":[],"sources":["../src/realtimeStt.ts"],"sourcesContent":["/**\n * Inworld Realtime STT — `wss://api.inworld.ai/stt/v1/transcribe:streamBidirectional`.\n *\n * Auth: `Authorization: Basic <API_KEY>` on the WS upgrade header (matches\n * Inworld's own JS sample). Setting headers needs the `ws` peer dep\n * (Node/Bun); see `./wsAuth.ts`.\n *\n * Wire shape (per [inworld-ai/inworld-api-examples](https://github.com/inworld-ai/inworld-api-examples)):\n *   client → server:\n *     `{ \"transcribeConfig\": { modelId, audioEncoding, sampleRateHertz, language, ... } }`  (first frame)\n *     `{ \"audioChunk\": { \"content\": \"<base64 audio>\" } }`                                  (×N)\n *     `{ \"closeStream\": {} }`                                                              (end-of-input)\n *   server → client:\n *     `{ \"result\": { \"transcription\": { \"transcript\": \"...\", \"isFinal\": true|false } } }`\n *     `{ \"result\": { \"speechStarted\": { \"startTimeMs\": ... } } }`\n *     `{ \"result\": { \"speechStopped\": { \"silenceDurationMs\": ... } } }`\n *\n * Inworld's STT WS sends audio as base64 inside JSON (NOT binary frames),\n * matching the rest of the Inworld API style.\n */\nimport { Cause, Effect, Encoding, Match, Queue, Redacted, Schema, Stream } from \"effect\"\nimport * as Socket from \"effect/unstable/socket/Socket\"\nimport * as AiError from \"@effect-uai/core/AiError\"\nimport type { AudioFormat } from \"@effect-uai/core/Audio\"\nimport * as Capabilities from \"@effect-uai/core/Capabilities\"\nimport * as JSONL from \"@effect-uai/core/JSONL\"\nimport type { TranscriptEvent, WordTimestamp } from \"@effect-uai/core/Transcript\"\nimport type { CommonStreamTranscribeRequest } from \"@effect-uai/core/Transcriber\"\nimport { authedWsConstructor } from \"./wsAuth.js\"\n\nexport type Config = { readonly apiKey: Redacted.Redacted; readonly baseUrl?: string }\n\n// ---------------------------------------------------------------------------\n// AudioFormat → `audioEncoding` slug for Inworld STT\n// ---------------------------------------------------------------------------\n\ntype WireEncoding = \"LINEAR16\" | \"MP3\" | \"OGG_OPUS\" | \"FLAC\"\n\nconst unsupportedFormat = (format: AudioFormat) =>\n  new AiError.Unsupported({\n    provider: \"inworld\",\n    capability: \"inputFormat\",\n    reason: `Inworld realtime STT accepts pcm_s16le (LINEAR16), mp3, ogg/opus, or flac. Got ${JSON.stringify(format)}.`,\n  })\n\nconst inputFormatToWire: (format: AudioFormat) => Effect.Effect<WireEncoding, AiError.AiError> =\n  Match.type<AudioFormat>().pipe(\n    Match.when({ container: \"raw\", encoding: \"pcm_s16le\" }, () =>\n      Effect.succeed<WireEncoding>(\"LINEAR16\"),\n    ),\n    Match.when({ container: \"wav\", encoding: \"pcm_s16le\" }, () =>\n      Effect.succeed<WireEncoding>(\"LINEAR16\"),\n    ),\n    Match.when({ container: \"mp3\" }, () => Effect.succeed<WireEncoding>(\"MP3\")),\n    Match.when({ container: \"ogg\", encoding: \"opus\" }, () =>\n      Effect.succeed<WireEncoding>(\"OGG_OPUS\"),\n    ),\n    Match.when({ container: \"flac\" }, () => Effect.succeed<WireEncoding>(\"FLAC\")),\n    Match.orElse((f) => Effect.fail(unsupportedFormat(f))),\n  )\n\n// ---------------------------------------------------------------------------\n// URL + frame builders\n// ---------------------------------------------------------------------------\n\nconst buildWsUrl = (cfg: Config) => {\n  const wsBase = (cfg.baseUrl ?? \"https://api.inworld.ai\").replace(/^http/, \"ws\")\n  return `${wsBase}/stt/v1/transcribe:streamBidirectional`\n}\n\nconst configFrame = (encoding: WireEncoding, request: CommonStreamTranscribeRequest) =>\n  JSON.stringify({\n    transcribeConfig: {\n      modelId: request.model,\n      audioEncoding: encoding,\n      sampleRateHertz: request.inputFormat.sampleRate,\n      numberOfChannels: request.inputFormat.channels ?? 1,\n      // Inworld's sample includes `language` even though docs mark it optional.\n      // Default to en-US to match the sample's behavior; caller can override.\n      language: request.language ?? \"en-US\",\n      // `prompts` is a vocab-biasing term list — maps from `biasingTerms`.\n      ...(request.biasingTerms !== undefined && { prompts: request.biasingTerms }),\n      ...(request.wordTimestamps === true && { includeWordTimestamps: true }),\n    },\n  })\n\nconst audioChunkFrame = (bytes: Uint8Array) =>\n  JSON.stringify({ audioChunk: { content: Encoding.encodeBase64(bytes) } })\n\nconst endTurnFrame = JSON.stringify({ endTurn: {} })\nconst closeStreamFrame = JSON.stringify({ closeStream: {} })\n\n// ---------------------------------------------------------------------------\n// Wire schemas (server → client)\n// ---------------------------------------------------------------------------\n\nconst WireWord = Schema.Struct({\n  word: Schema.String,\n  startTimeMs: Schema.optional(Schema.Number),\n  endTimeMs: Schema.optional(Schema.Number),\n  confidence: Schema.optional(Schema.Number),\n})\n\nconst ResultBody = Schema.Struct({\n  transcription: Schema.optional(\n    Schema.Struct({\n      transcript: Schema.String,\n      isFinal: Schema.optional(Schema.Boolean),\n      wordTimestamps: Schema.optional(Schema.NullOr(Schema.Array(WireWord))),\n    }),\n  ),\n  speechStarted: Schema.optional(\n    Schema.Struct({\n      startTimeMs: Schema.optional(Schema.Number),\n      confidence: Schema.optional(Schema.Number),\n    }),\n  ),\n  speechStopped: Schema.optional(\n    Schema.Struct({\n      silenceDurationMs: Schema.optional(Schema.Number),\n    }),\n  ),\n  status: Schema.optional(Schema.Unknown),\n})\n\nconst ServerFrame = Schema.Struct({\n  result: Schema.optional(ResultBody),\n  error: Schema.optional(Schema.Unknown),\n})\nconst decodeServerFrame = Schema.decodeUnknownEffect(ServerFrame)\n\nconst wireWordToCommon = (w: typeof WireWord.Type): WordTimestamp | undefined =>\n  w.startTimeMs === undefined || w.endTimeMs === undefined\n    ? undefined\n    : {\n        text: w.word,\n        startSeconds: w.startTimeMs / 1000,\n        endSeconds: w.endTimeMs / 1000,\n        ...(w.confidence !== undefined && { confidence: w.confidence }),\n      }\n\nexport const wireToEvent = (frame: typeof ServerFrame.Type): TranscriptEvent | undefined => {\n  if (frame.error !== undefined) {\n    return {\n      _tag: \"error\",\n      message: typeof frame.error === \"string\" ? frame.error : JSON.stringify(frame.error),\n    }\n  }\n  const result = frame.result\n  if (result === undefined) return undefined\n  if (result.transcription !== undefined) {\n    const t = result.transcription\n    const words = t.wordTimestamps\n      ?.map(wireWordToCommon)\n      .filter((w): w is WordTimestamp => w !== undefined)\n    return t.isFinal === true\n      ? {\n          _tag: \"final\",\n          text: t.transcript,\n          ...(words !== undefined && words.length > 0 && { words }),\n        }\n      : {\n          _tag: \"partial\",\n          text: t.transcript,\n          ...(words !== undefined && words.length > 0 && { words }),\n        }\n  }\n  if (result.speechStarted !== undefined) {\n    return {\n      _tag: \"speech-started\",\n      atSeconds: (result.speechStarted.startTimeMs ?? 0) / 1000,\n    }\n  }\n  if (result.speechStopped !== undefined) {\n    return { _tag: \"utterance-ended\", atSeconds: 0 }\n  }\n  return undefined\n}\n\nconst handleServerMessage = (queue: Queue.Queue<TranscriptEvent, Cause.Done>) => (raw: string) =>\n  Effect.gen(function* () {\n    const json = yield* JSONL.parseSafe(raw)\n    if (json === undefined) return\n    const decoded = yield* decodeServerFrame(json).pipe(Effect.option)\n    if (decoded._tag === \"None\") return\n    const event = wireToEvent(decoded.value)\n    if (event !== undefined) yield* Queue.offer(queue, event)\n  })\n\n// ---------------------------------------------------------------------------\n// Stream<Uint8Array> → Stream<TranscriptEvent>\n// ---------------------------------------------------------------------------\n\nexport const streamTranscription =\n  (cfg: Config) =>\n  <E, R>(\n    audioIn: Stream.Stream<Uint8Array, E, R>,\n    request: CommonStreamTranscribeRequest,\n  ): Stream.Stream<TranscriptEvent, AiError.AiError | E, R> =>\n    Stream.unwrap(\n      Effect.gen(function* () {\n        yield* Capabilities.warnDroppedWhen(request.prompt, {\n          provider: \"inworld\",\n          capability: \"prompt\",\n          field: \"prompt\",\n          reason: \"Inworld STT has no free-form prompt field; bias via `biasingTerms`.\",\n        })\n        const encoding = yield* inputFormatToWire(request.inputFormat)\n        const socket = yield* Socket.makeWebSocket(buildWsUrl(cfg), {\n          // Effect's Socket treats all close codes as errors by default —\n          // whitelist standard clean-close codes (1000 / 1001 / 1005).\n          closeCodeIsError: (code) => code !== 1000 && code !== 1001 && code !== 1005,\n        }).pipe(Effect.provideService(Socket.WebSocketConstructor, authedWsConstructor(cfg.apiKey)))\n        const queue = yield* Queue.bounded<TranscriptEvent, Cause.Done>(64)\n        const write = yield* socket.writer\n\n        // Writer: config → drain audio → endTurn → closeStream. Inworld\n        // emits `result.transcription` partials in real-time via VAD;\n        // `endTurn` is sent on input-stream end to flush any tail audio\n        // into a final transcript, then `closeStream` for graceful close.\n        yield* Effect.gen(function* () {\n          yield* write(configFrame(encoding, request))\n          yield* Stream.runForEach(audioIn, (bytes) => write(audioChunkFrame(bytes)))\n          yield* write(endTurnFrame)\n          yield* write(closeStreamFrame)\n        }).pipe(Effect.ignore, Effect.forkScoped)\n\n        // `Queue.end` flushes pending events then fails the next take with\n        // `Done` — clean stream end. `Queue.shutdown` would CLEAR queued\n        // items and interrupt pending takes (wrong for graceful teardown).\n        yield* socket\n          .runString(handleServerMessage(queue))\n          .pipe(Effect.ensuring(Queue.end(queue)), Effect.forkScoped)\n\n        return Stream.fromQueue(queue)\n      }),\n    )\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;AAsCA,MAAM,qBAAqB,WACzB,IAAI,QAAQ,YAAY;CACtB,UAAU;CACV,YAAY;CACZ,QAAQ,kFAAkF,KAAK,UAAU,OAAO,CAAC;CAClH,CAAC;AAEJ,MAAM,oBACJ,MAAM,MAAmB,CAAC,KACxB,MAAM,KAAK;CAAE,WAAW;CAAO,UAAU;CAAa,QACpD,OAAO,QAAsB,WAAW,CACzC,EACD,MAAM,KAAK;CAAE,WAAW;CAAO,UAAU;CAAa,QACpD,OAAO,QAAsB,WAAW,CACzC,EACD,MAAM,KAAK,EAAE,WAAW,OAAO,QAAQ,OAAO,QAAsB,MAAM,CAAC,EAC3E,MAAM,KAAK;CAAE,WAAW;CAAO,UAAU;CAAQ,QAC/C,OAAO,QAAsB,WAAW,CACzC,EACD,MAAM,KAAK,EAAE,WAAW,QAAQ,QAAQ,OAAO,QAAsB,OAAO,CAAC,EAC7E,MAAM,QAAQ,MAAM,OAAO,KAAK,kBAAkB,EAAE,CAAC,CAAC,CACvD;AAMH,MAAM,cAAc,QAAgB;AAElC,QAAO,IADS,IAAI,WAAW,0BAA0B,QAAQ,SAAS,KAC1D,CAAC;;AAGnB,MAAM,eAAe,UAAwB,YAC3C,KAAK,UAAU,EACb,kBAAkB;CAChB,SAAS,QAAQ;CACjB,eAAe;CACf,iBAAiB,QAAQ,YAAY;CACrC,kBAAkB,QAAQ,YAAY,YAAY;CAGlD,UAAU,QAAQ,YAAY;CAE9B,GAAI,QAAQ,iBAAiB,KAAA,KAAa,EAAE,SAAS,QAAQ,cAAc;CAC3E,GAAI,QAAQ,mBAAmB,QAAQ,EAAE,uBAAuB,MAAM;CACvE,EACF,CAAC;AAEJ,MAAM,mBAAmB,UACvB,KAAK,UAAU,EAAE,YAAY,EAAE,SAAS,SAAS,aAAa,MAAM,EAAE,EAAE,CAAC;AAE3E,MAAM,eAAe,KAAK,UAAU,EAAE,SAAS,EAAE,EAAE,CAAC;AACpD,MAAM,mBAAmB,KAAK,UAAU,EAAE,aAAa,EAAE,EAAE,CAAC;AAM5D,MAAM,WAAW,OAAO,OAAO;CAC7B,MAAM,OAAO;CACb,aAAa,OAAO,SAAS,OAAO,OAAO;CAC3C,WAAW,OAAO,SAAS,OAAO,OAAO;CACzC,YAAY,OAAO,SAAS,OAAO,OAAO;CAC3C,CAAC;AAEF,MAAM,aAAa,OAAO,OAAO;CAC/B,eAAe,OAAO,SACpB,OAAO,OAAO;EACZ,YAAY,OAAO;EACnB,SAAS,OAAO,SAAS,OAAO,QAAQ;EACxC,gBAAgB,OAAO,SAAS,OAAO,OAAO,OAAO,MAAM,SAAS,CAAC,CAAC;EACvE,CAAC,CACH;CACD,eAAe,OAAO,SACpB,OAAO,OAAO;EACZ,aAAa,OAAO,SAAS,OAAO,OAAO;EAC3C,YAAY,OAAO,SAAS,OAAO,OAAO;EAC3C,CAAC,CACH;CACD,eAAe,OAAO,SACpB,OAAO,OAAO,EACZ,mBAAmB,OAAO,SAAS,OAAO,OAAO,EAClD,CAAC,CACH;CACD,QAAQ,OAAO,SAAS,OAAO,QAAQ;CACxC,CAAC;AAEF,MAAM,cAAc,OAAO,OAAO;CAChC,QAAQ,OAAO,SAAS,WAAW;CACnC,OAAO,OAAO,SAAS,OAAO,QAAQ;CACvC,CAAC;AACF,MAAM,oBAAoB,OAAO,oBAAoB,YAAY;AAEjE,MAAM,oBAAoB,MACxB,EAAE,gBAAgB,KAAA,KAAa,EAAE,cAAc,KAAA,IAC3C,KAAA,IACA;CACE,MAAM,EAAE;CACR,cAAc,EAAE,cAAc;CAC9B,YAAY,EAAE,YAAY;CAC1B,GAAI,EAAE,eAAe,KAAA,KAAa,EAAE,YAAY,EAAE,YAAY;CAC/D;AAEP,MAAa,eAAe,UAAgE;AAC1F,KAAI,MAAM,UAAU,KAAA,EAClB,QAAO;EACL,MAAM;EACN,SAAS,OAAO,MAAM,UAAU,WAAW,MAAM,QAAQ,KAAK,UAAU,MAAM,MAAM;EACrF;CAEH,MAAM,SAAS,MAAM;AACrB,KAAI,WAAW,KAAA,EAAW,QAAO,KAAA;AACjC,KAAI,OAAO,kBAAkB,KAAA,GAAW;EACtC,MAAM,IAAI,OAAO;EACjB,MAAM,QAAQ,EAAE,gBACZ,IAAI,iBAAiB,CACtB,QAAQ,MAA0B,MAAM,KAAA,EAAU;AACrD,SAAO,EAAE,YAAY,OACjB;GACE,MAAM;GACN,MAAM,EAAE;GACR,GAAI,UAAU,KAAA,KAAa,MAAM,SAAS,KAAK,EAAE,OAAO;GACzD,GACD;GACE,MAAM;GACN,MAAM,EAAE;GACR,GAAI,UAAU,KAAA,KAAa,MAAM,SAAS,KAAK,EAAE,OAAO;GACzD;;AAEP,KAAI,OAAO,kBAAkB,KAAA,EAC3B,QAAO;EACL,MAAM;EACN,YAAY,OAAO,cAAc,eAAe,KAAK;EACtD;AAEH,KAAI,OAAO,kBAAkB,KAAA,EAC3B,QAAO;EAAE,MAAM;EAAmB,WAAW;EAAG;;AAKpD,MAAM,uBAAuB,WAAqD,QAChF,OAAO,IAAI,aAAa;CACtB,MAAM,OAAO,OAAO,MAAM,UAAU,IAAI;AACxC,KAAI,SAAS,KAAA,EAAW;CACxB,MAAM,UAAU,OAAO,kBAAkB,KAAK,CAAC,KAAK,OAAO,OAAO;AAClE,KAAI,QAAQ,SAAS,OAAQ;CAC7B,MAAM,QAAQ,YAAY,QAAQ,MAAM;AACxC,KAAI,UAAU,KAAA,EAAW,QAAO,MAAM,MAAM,OAAO,MAAM;EACzD;AAMJ,MAAa,uBACV,SAEC,SACA,YAEA,OAAO,OACL,OAAO,IAAI,aAAa;AACtB,QAAO,aAAa,gBAAgB,QAAQ,QAAQ;EAClD,UAAU;EACV,YAAY;EACZ,OAAO;EACP,QAAQ;EACT,CAAC;CACF,MAAM,WAAW,OAAO,kBAAkB,QAAQ,YAAY;CAC9D,MAAM,SAAS,OAAO,OAAO,cAAc,WAAW,IAAI,EAAE,EAG1D,mBAAmB,SAAS,SAAS,OAAQ,SAAS,QAAQ,SAAS,MACxE,CAAC,CAAC,KAAK,OAAO,eAAe,OAAO,sBAAsB,oBAAoB,IAAI,OAAO,CAAC,CAAC;CAC5F,MAAM,QAAQ,OAAO,MAAM,QAAqC,GAAG;CACnE,MAAM,QAAQ,OAAO,OAAO;AAM5B,QAAO,OAAO,IAAI,aAAa;AAC7B,SAAO,MAAM,YAAY,UAAU,QAAQ,CAAC;AAC5C,SAAO,OAAO,WAAW,UAAU,UAAU,MAAM,gBAAgB,MAAM,CAAC,CAAC;AAC3E,SAAO,MAAM,aAAa;AAC1B,SAAO,MAAM,iBAAiB;GAC9B,CAAC,KAAK,OAAO,QAAQ,OAAO,WAAW;AAKzC,QAAO,OACJ,UAAU,oBAAoB,MAAM,CAAC,CACrC,KAAK,OAAO,SAAS,MAAM,IAAI,MAAM,CAAC,EAAE,OAAO,WAAW;AAE7D,QAAO,OAAO,UAAU,MAAM;EAC9B,CACH"}

package/dist/realtimeTts.d.mts

@@ -1,5 +1,6 @@
 import { a as InworldVoiceId, i as InworldTtsModel, n as InworldDeliveryMode } from "./models-CPLZDcgt.mjs";
 import { Redacted, Stream } from "effect";
+import { CustomPronunciation } from "@effect-uai/core/SpeechSynthesizer";
 import * as AiError from "@effect-uai/core/AiError";
 import { AudioChunk, AudioFormat } from "@effect-uai/core/Audio";
 
@@ -21,6 +22,9 @@ type StreamSynthesizeRequest = {
   readonly deliveryMode?: InworldDeliveryMode;
   readonly applyTextNormalization?: "ON" | "OFF";
   readonly speed?: number;
+  /** Carried from the Common request so the WS path can reject inline
+   *  pronunciations (no place to rewrite into a chunked text stream). */
+  readonly pronunciations?: ReadonlyArray<CustomPronunciation>;
 };
 declare const buildWsUrl: (cfg: Config) => string;
 declare const streamSynthesis: (cfg: Config) => <E, R>(textIn: Stream.Stream<string, E, R>, request: StreamSynthesizeRequest) => Stream.Stream<AudioChunk, AiError.AiError | E, R>;

package/dist/realtimeTts.d.mts.map

@@ -1 +1 @@
-{"version":3,"file":"realtimeTts.d.mts","names":[],"sources":["../src/realtimeTts.ts"],"mappings":";;;;;;KA8BY,MAAA;EAAA,SAAoB,MAAA,EAAQ,QAAA,CAAS,QAAA;EAAA,SAAmB,OAAA;AAAA;;;;;KAMxD,uBAAA;EAAA,SACD,KAAA,EAAO,eAAA;EAAA,SACP,OAAA,EAAS,cAAA;EAAA,SACT,YAAA,GAAe,WAAA;EAAA,SACf,YAAA;EAAA,SACA,WAAA;EAAA,SACA,YAAA,GAAe,mBAAA;EAAA,SACf,sBAAA;EAAA,SACA,KAAA;AAAA;AAAA,cAOE,UAAA,GAAc,GAAA,EAAK,MAAA;AAAA,cA8EnB,eAAA,GACV,GAAA,EAAK,MAAA,YAEJ,MAAA,EAAQ,MAAA,CAAO,MAAA,SAAe,CAAA,EAAG,CAAA,GACjC,OAAA,EAAS,uBAAA,KACR,MAAA,CAAO,MAAA,CAAO,UAAA,EAAY,OAAA,CAAQ,OAAA,GAAU,CAAA,EAAG,CAAA"}
+{"version":3,"file":"realtimeTts.d.mts","names":[],"sources":["../src/realtimeTts.ts"],"mappings":";;;;;;;KA+BY,MAAA;EAAA,SAAoB,MAAA,EAAQ,QAAA,CAAS,QAAA;EAAA,SAAmB,OAAA;AAAA;;;;;KAMxD,uBAAA;EAAA,SACD,KAAA,EAAO,eAAA;EAAA,SACP,OAAA,EAAS,cAAA;EAAA,SACT,YAAA,GAAe,WAAA;EAAA,SACf,YAAA;EAAA,SACA,WAAA;EAAA,SACA,YAAA,GAAe,mBAAA;EAAA,SACf,sBAAA;EAAA,SACA,KAAA;EAGkD;;EAAA,SAAlD,cAAA,GAAiB,aAAA,CAAc,mBAAA;AAAA;AAAA,cAO7B,UAAA,GAAc,GAAA,EAAK,MAAA;AAAA,cA8EnB,eAAA,GACV,GAAA,EAAK,MAAA,YAEJ,MAAA,EAAQ,MAAA,CAAO,MAAA,SAAe,CAAA,EAAG,CAAA,GACjC,OAAA,EAAS,uBAAA,KACR,MAAA,CAAO,MAAA,CAAO,UAAA,EAAY,OAAA,CAAQ,OAAA,GAAU,CAAA,EAAG,CAAA"}

package/dist/realtimeTts.mjs

@@ -1,6 +1,7 @@
 import { audioConfigFor, decodeAudioContent, defaultFormat } from "./codec.mjs";
 import { authedWsConstructor } from "./wsAuth.mjs";
 import { Effect, Queue, Schema, Stream } from "effect";
+import * as AiError from "@effect-uai/core/AiError";
 import * as JSONL from "@effect-uai/core/JSONL";
 import * as Socket from "effect/unstable/socket/Socket";
 //#region src/realtimeTts.ts
@@ -76,6 +77,11 @@ const handleServerFrame = (queue) => (raw) => Effect.gen(function* () {
 	if (bytes._tag === "Some") yield* Queue.offer(queue, { bytes: bytes.value });
 });
 const streamSynthesis = (cfg) => (textIn, request) => Stream.unwrap(Effect.gen(function* () {
+	if (request.pronunciations !== void 0 && request.pronunciations.length > 0) return Stream.fail(new AiError.Unsupported({
+		provider: "inworld",
+		capability: "pronunciations",
+		reason: "Inworld applies inline /ipa/ pronunciations on sync `synthesize` only; the incremental WS path streams text in chunks with no place to rewrite. Use `synthesize` for pronunciation overrides."
+	}));
 	const create = yield* createFrame(request);
 	const socket = yield* Socket.makeWebSocket(buildWsUrl(cfg), { closeCodeIsError: (code) => code !== 1e3 && code !== 1001 && code !== 1005 }).pipe(Effect.provideService(Socket.WebSocketConstructor, authedWsConstructor(cfg.apiKey)));
 	const queue = yield* Queue.bounded(64);

package/dist/realtimeTts.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"realtimeTts.mjs","names":[],"sources":["../src/realtimeTts.ts"],"sourcesContent":["/**\n * Inworld Realtime TTS — `wss://api.inworld.ai/tts/v1/voice:streamBidirectional`.\n *\n * Auth: `Authorization: Basic <API_KEY>` on the WS upgrade header (matches\n * Inworld's own JS sample). The docs mention a `?Authorization=…` query\n * variant but the server rejects it in practice. Setting headers needs the\n * `ws` peer dep (Node/Bun); see `./wsAuth.ts`.\n *\n * Wire shape (single-context per call):\n *   client → server:\n *     `{ \"create\": { voiceId, modelId, audioConfig, ... } }`\n *     `{ \"send_text\": { \"text\": \"...\" } }`             (repeated)\n *     `{ \"close_context\": {} }`                        (text stream end)\n *   server → client:\n *     `{ \"result\": { \"contextCreated\": {...} } }`      (handshake ack)\n *     `{ \"result\": { \"audioChunk\": { \"audioContent\": \"<b64>\" } } }` (×N)\n *     `{ \"result\": { \"contextClosed\": {...} } }`       (final)\n *\n * Multi-context (`contextId`) is not surfaced here — one logical\n * utterance per call.\n */\nimport { Cause, Effect, Queue, Redacted, Schema, Stream } from \"effect\"\nimport * as Socket from \"effect/unstable/socket/Socket\"\nimport * as AiError from \"@effect-uai/core/AiError\"\nimport type { AudioChunk, AudioFormat } from \"@effect-uai/core/Audio\"\nimport * as JSONL from \"@effect-uai/core/JSONL\"\nimport { audioConfigFor, decodeAudioContent, defaultFormat } from \"./codec.js\"\nimport type { InworldDeliveryMode, InworldTtsModel, InworldVoiceId } from \"./models.js\"\nimport { authedWsConstructor } from \"./wsAuth.js\"\n\nexport type Config = { readonly apiKey: Redacted.Redacted; readonly baseUrl?: string }\n\n/**\n * Incremental-text-in request. Mirrors the sync request minus `text` (which\n * comes from the input stream).\n */\nexport type StreamSynthesizeRequest = {\n  readonly model: InworldTtsModel\n  readonly voiceId: InworldVoiceId\n  readonly outputFormat?: AudioFormat\n  readonly languageCode?: string\n  readonly temperature?: number\n  readonly deliveryMode?: InworldDeliveryMode\n  readonly applyTextNormalization?: \"ON\" | \"OFF\"\n  readonly speed?: number\n}\n\n// ---------------------------------------------------------------------------\n// URL + frame builders\n// ---------------------------------------------------------------------------\n\nexport const buildWsUrl = (cfg: Config) => {\n  const wsBase = (cfg.baseUrl ?? \"https://api.inworld.ai\").replace(/^http/, \"ws\")\n  return `${wsBase}/tts/v1/voice:streamBidirectional`\n}\n\n// TTS WS uses snake_case outbound (per Inworld's own JS sample). The\n// camelCase helper from `codec.ts` is built for the REST/JSON sync paths;\n// rewire the field names here for the WS path. Server responses come back\n// camelCase regardless — see `ResultBody` below.\nconst audioConfigSnake = (request: StreamSynthesizeRequest) =>\n  Effect.map(audioConfigFor(request.outputFormat ?? defaultFormat, request.speed), (cfg) => ({\n    audio_encoding: cfg.audioEncoding,\n    ...(cfg.sampleRateHertz !== undefined && { sample_rate_hertz: cfg.sampleRateHertz }),\n    ...(cfg.bitRate !== undefined && { bit_rate: cfg.bitRate }),\n    ...(cfg.speakingRate !== undefined && { speaking_rate: cfg.speakingRate }),\n  }))\n\nconst createFrame = (request: StreamSynthesizeRequest) =>\n  Effect.map(audioConfigSnake(request), (audio_config) =>\n    JSON.stringify({\n      create: {\n        voice_id: request.voiceId,\n        model_id: request.model,\n        audio_config,\n        ...(request.languageCode !== undefined && { language: request.languageCode }),\n        ...(request.deliveryMode !== undefined && { delivery_mode: request.deliveryMode }),\n        ...(request.temperature !== undefined && { temperature: request.temperature }),\n        ...(request.applyTextNormalization !== undefined && {\n          apply_text_normalization: request.applyTextNormalization,\n        }),\n      },\n    }),\n  )\n\n// `flush_context: {}` inside `send_text` nudges the server to flush\n// pending audio promptly — matches the low-latency sample.\nconst sendTextFrame = (text: string) => JSON.stringify({ send_text: { text, flush_context: {} } })\nconst closeContextFrame = JSON.stringify({ close_context: {} })\n\n// ---------------------------------------------------------------------------\n// Wire schema (server → client)\n// ---------------------------------------------------------------------------\n\nconst ResultBody = Schema.Struct({\n  audioChunk: Schema.optional(Schema.Struct({ audioContent: Schema.String })),\n  contextCreated: Schema.optional(Schema.Unknown),\n  flushCompleted: Schema.optional(Schema.Unknown),\n  contextClosed: Schema.optional(Schema.Unknown),\n  status: Schema.optional(Schema.Unknown),\n})\n\nconst ServerFrame = Schema.Struct({\n  result: Schema.optional(ResultBody),\n  error: Schema.optional(Schema.Unknown),\n})\nconst decodeServerFrame = Schema.decodeUnknownEffect(ServerFrame)\n\nconst handleServerFrame = (queue: Queue.Queue<AudioChunk, Cause.Done>) => (raw: string) =>\n  Effect.gen(function* () {\n    const json = yield* JSONL.parseSafe(raw)\n    if (json === undefined) return\n    const decoded = yield* decodeServerFrame(json).pipe(Effect.option)\n    if (decoded._tag === \"None\") return\n    const frame = decoded.value\n    if (frame.error !== undefined) {\n      yield* Effect.logWarning(\"[inworld-tts] server error frame\", { error: frame.error })\n      return\n    }\n    const audio = frame.result?.audioChunk?.audioContent\n    if (audio === undefined || audio === \"\") return\n    const bytes = yield* decodeAudioContent(audio).pipe(Effect.option)\n    if (bytes._tag === \"Some\") yield* Queue.offer(queue, { bytes: bytes.value })\n  })\n\n// ---------------------------------------------------------------------------\n// Stream<string> → Stream<AudioChunk>\n// ---------------------------------------------------------------------------\n\nexport const streamSynthesis =\n  (cfg: Config) =>\n  <E, R>(\n    textIn: Stream.Stream<string, E, R>,\n    request: StreamSynthesizeRequest,\n  ): Stream.Stream<AudioChunk, AiError.AiError | E, R> =>\n    Stream.unwrap(\n      Effect.gen(function* () {\n        const create = yield* createFrame(request)\n        const socket = yield* Socket.makeWebSocket(buildWsUrl(cfg), {\n          // Effect's Socket treats all close codes as errors by default —\n          // whitelist standard clean-close codes (1000 / 1001 / 1005).\n          closeCodeIsError: (code) => code !== 1000 && code !== 1001 && code !== 1005,\n        }).pipe(Effect.provideService(Socket.WebSocketConstructor, authedWsConstructor(cfg.apiKey)))\n        const queue = yield* Queue.bounded<AudioChunk, Cause.Done>(64)\n        const write = yield* socket.writer\n\n        // Writer: BOS `create` → drain text as `send_text` → `close_context`.\n        // The reader fiber surfaces the remaining audio drained by the server\n        // post-close and shuts the queue when the WS upstream closes.\n        yield* Effect.gen(function* () {\n          yield* write(create)\n          yield* Stream.runForEach(textIn, (text) =>\n            text.length === 0 ? Effect.void : write(sendTextFrame(text)),\n          )\n          yield* write(closeContextFrame)\n        }).pipe(Effect.ignore, Effect.forkScoped)\n\n        // `Queue.end` flushes pending chunks then fails the next take with\n        // `Done` — clean stream end. `Queue.shutdown` would CLEAR queued items\n        // and interrupt pending takes (wrong for graceful teardown).\n        yield* socket\n          .runString(handleServerFrame(queue))\n          .pipe(Effect.ensuring(Queue.end(queue)), Effect.forkScoped)\n\n        return Stream.fromQueue(queue)\n      }),\n    )\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;AAmDA,MAAa,cAAc,QAAgB;AAEzC,QAAO,IADS,IAAI,WAAW,0BAA0B,QAAQ,SAAS,KAC1D,CAAC;;AAOnB,MAAM,oBAAoB,YACxB,OAAO,IAAI,eAAe,QAAQ,gBAAgB,eAAe,QAAQ,MAAM,GAAG,SAAS;CACzF,gBAAgB,IAAI;CACpB,GAAI,IAAI,oBAAoB,KAAA,KAAa,EAAE,mBAAmB,IAAI,iBAAiB;CACnF,GAAI,IAAI,YAAY,KAAA,KAAa,EAAE,UAAU,IAAI,SAAS;CAC1D,GAAI,IAAI,iBAAiB,KAAA,KAAa,EAAE,eAAe,IAAI,cAAc;CAC1E,EAAE;AAEL,MAAM,eAAe,YACnB,OAAO,IAAI,iBAAiB,QAAQ,GAAG,iBACrC,KAAK,UAAU,EACb,QAAQ;CACN,UAAU,QAAQ;CAClB,UAAU,QAAQ;CAClB;CACA,GAAI,QAAQ,iBAAiB,KAAA,KAAa,EAAE,UAAU,QAAQ,cAAc;CAC5E,GAAI,QAAQ,iBAAiB,KAAA,KAAa,EAAE,eAAe,QAAQ,cAAc;CACjF,GAAI,QAAQ,gBAAgB,KAAA,KAAa,EAAE,aAAa,QAAQ,aAAa;CAC7E,GAAI,QAAQ,2BAA2B,KAAA,KAAa,EAClD,0BAA0B,QAAQ,wBACnC;CACF,EACF,CAAC,CACH;AAIH,MAAM,iBAAiB,SAAiB,KAAK,UAAU,EAAE,WAAW;CAAE;CAAM,eAAe,EAAE;CAAE,EAAE,CAAC;AAClG,MAAM,oBAAoB,KAAK,UAAU,EAAE,eAAe,EAAE,EAAE,CAAC;AAM/D,MAAM,aAAa,OAAO,OAAO;CAC/B,YAAY,OAAO,SAAS,OAAO,OAAO,EAAE,cAAc,OAAO,QAAQ,CAAC,CAAC;CAC3E,gBAAgB,OAAO,SAAS,OAAO,QAAQ;CAC/C,gBAAgB,OAAO,SAAS,OAAO,QAAQ;CAC/C,eAAe,OAAO,SAAS,OAAO,QAAQ;CAC9C,QAAQ,OAAO,SAAS,OAAO,QAAQ;CACxC,CAAC;AAEF,MAAM,cAAc,OAAO,OAAO;CAChC,QAAQ,OAAO,SAAS,WAAW;CACnC,OAAO,OAAO,SAAS,OAAO,QAAQ;CACvC,CAAC;AACF,MAAM,oBAAoB,OAAO,oBAAoB,YAAY;AAEjE,MAAM,qBAAqB,WAAgD,QACzE,OAAO,IAAI,aAAa;CACtB,MAAM,OAAO,OAAO,MAAM,UAAU,IAAI;AACxC,KAAI,SAAS,KAAA,EAAW;CACxB,MAAM,UAAU,OAAO,kBAAkB,KAAK,CAAC,KAAK,OAAO,OAAO;AAClE,KAAI,QAAQ,SAAS,OAAQ;CAC7B,MAAM,QAAQ,QAAQ;AACtB,KAAI,MAAM,UAAU,KAAA,GAAW;AAC7B,SAAO,OAAO,WAAW,oCAAoC,EAAE,OAAO,MAAM,OAAO,CAAC;AACpF;;CAEF,MAAM,QAAQ,MAAM,QAAQ,YAAY;AACxC,KAAI,UAAU,KAAA,KAAa,UAAU,GAAI;CACzC,MAAM,QAAQ,OAAO,mBAAmB,MAAM,CAAC,KAAK,OAAO,OAAO;AAClE,KAAI,MAAM,SAAS,OAAQ,QAAO,MAAM,MAAM,OAAO,EAAE,OAAO,MAAM,OAAO,CAAC;EAC5E;AAMJ,MAAa,mBACV,SAEC,QACA,YAEA,OAAO,OACL,OAAO,IAAI,aAAa;CACtB,MAAM,SAAS,OAAO,YAAY,QAAQ;CAC1C,MAAM,SAAS,OAAO,OAAO,cAAc,WAAW,IAAI,EAAE,EAG1D,mBAAmB,SAAS,SAAS,OAAQ,SAAS,QAAQ,SAAS,MACxE,CAAC,CAAC,KAAK,OAAO,eAAe,OAAO,sBAAsB,oBAAoB,IAAI,OAAO,CAAC,CAAC;CAC5F,MAAM,QAAQ,OAAO,MAAM,QAAgC,GAAG;CAC9D,MAAM,QAAQ,OAAO,OAAO;AAK5B,QAAO,OAAO,IAAI,aAAa;AAC7B,SAAO,MAAM,OAAO;AACpB,SAAO,OAAO,WAAW,SAAS,SAChC,KAAK,WAAW,IAAI,OAAO,OAAO,MAAM,cAAc,KAAK,CAAC,CAC7D;AACD,SAAO,MAAM,kBAAkB;GAC/B,CAAC,KAAK,OAAO,QAAQ,OAAO,WAAW;AAKzC,QAAO,OACJ,UAAU,kBAAkB,MAAM,CAAC,CACnC,KAAK,OAAO,SAAS,MAAM,IAAI,MAAM,CAAC,EAAE,OAAO,WAAW;AAE7D,QAAO,OAAO,UAAU,MAAM;EAC9B,CACH"}
+{"version":3,"file":"realtimeTts.mjs","names":[],"sources":["../src/realtimeTts.ts"],"sourcesContent":["/**\n * Inworld Realtime TTS — `wss://api.inworld.ai/tts/v1/voice:streamBidirectional`.\n *\n * Auth: `Authorization: Basic <API_KEY>` on the WS upgrade header (matches\n * Inworld's own JS sample). The docs mention a `?Authorization=…` query\n * variant but the server rejects it in practice. Setting headers needs the\n * `ws` peer dep (Node/Bun); see `./wsAuth.ts`.\n *\n * Wire shape (single-context per call):\n *   client → server:\n *     `{ \"create\": { voiceId, modelId, audioConfig, ... } }`\n *     `{ \"send_text\": { \"text\": \"...\" } }`             (repeated)\n *     `{ \"close_context\": {} }`                        (text stream end)\n *   server → client:\n *     `{ \"result\": { \"contextCreated\": {...} } }`      (handshake ack)\n *     `{ \"result\": { \"audioChunk\": { \"audioContent\": \"<b64>\" } } }` (×N)\n *     `{ \"result\": { \"contextClosed\": {...} } }`       (final)\n *\n * Multi-context (`contextId`) is not surfaced here — one logical\n * utterance per call.\n */\nimport { Cause, Effect, Queue, Redacted, Schema, Stream } from \"effect\"\nimport * as Socket from \"effect/unstable/socket/Socket\"\nimport * as AiError from \"@effect-uai/core/AiError\"\nimport type { AudioChunk, AudioFormat } from \"@effect-uai/core/Audio\"\nimport * as JSONL from \"@effect-uai/core/JSONL\"\nimport type { CustomPronunciation } from \"@effect-uai/core/SpeechSynthesizer\"\nimport { audioConfigFor, decodeAudioContent, defaultFormat } from \"./codec.js\"\nimport type { InworldDeliveryMode, InworldTtsModel, InworldVoiceId } from \"./models.js\"\nimport { authedWsConstructor } from \"./wsAuth.js\"\n\nexport type Config = { readonly apiKey: Redacted.Redacted; readonly baseUrl?: string }\n\n/**\n * Incremental-text-in request. Mirrors the sync request minus `text` (which\n * comes from the input stream).\n */\nexport type StreamSynthesizeRequest = {\n  readonly model: InworldTtsModel\n  readonly voiceId: InworldVoiceId\n  readonly outputFormat?: AudioFormat\n  readonly languageCode?: string\n  readonly temperature?: number\n  readonly deliveryMode?: InworldDeliveryMode\n  readonly applyTextNormalization?: \"ON\" | \"OFF\"\n  readonly speed?: number\n  /** Carried from the Common request so the WS path can reject inline\n   *  pronunciations (no place to rewrite into a chunked text stream). */\n  readonly pronunciations?: ReadonlyArray<CustomPronunciation>\n}\n\n// ---------------------------------------------------------------------------\n// URL + frame builders\n// ---------------------------------------------------------------------------\n\nexport const buildWsUrl = (cfg: Config) => {\n  const wsBase = (cfg.baseUrl ?? \"https://api.inworld.ai\").replace(/^http/, \"ws\")\n  return `${wsBase}/tts/v1/voice:streamBidirectional`\n}\n\n// TTS WS uses snake_case outbound (per Inworld's own JS sample). The\n// camelCase helper from `codec.ts` is built for the REST/JSON sync paths;\n// rewire the field names here for the WS path. Server responses come back\n// camelCase regardless — see `ResultBody` below.\nconst audioConfigSnake = (request: StreamSynthesizeRequest) =>\n  Effect.map(audioConfigFor(request.outputFormat ?? defaultFormat, request.speed), (cfg) => ({\n    audio_encoding: cfg.audioEncoding,\n    ...(cfg.sampleRateHertz !== undefined && { sample_rate_hertz: cfg.sampleRateHertz }),\n    ...(cfg.bitRate !== undefined && { bit_rate: cfg.bitRate }),\n    ...(cfg.speakingRate !== undefined && { speaking_rate: cfg.speakingRate }),\n  }))\n\nconst createFrame = (request: StreamSynthesizeRequest) =>\n  Effect.map(audioConfigSnake(request), (audio_config) =>\n    JSON.stringify({\n      create: {\n        voice_id: request.voiceId,\n        model_id: request.model,\n        audio_config,\n        ...(request.languageCode !== undefined && { language: request.languageCode }),\n        ...(request.deliveryMode !== undefined && { delivery_mode: request.deliveryMode }),\n        ...(request.temperature !== undefined && { temperature: request.temperature }),\n        ...(request.applyTextNormalization !== undefined && {\n          apply_text_normalization: request.applyTextNormalization,\n        }),\n      },\n    }),\n  )\n\n// `flush_context: {}` inside `send_text` nudges the server to flush\n// pending audio promptly — matches the low-latency sample.\nconst sendTextFrame = (text: string) => JSON.stringify({ send_text: { text, flush_context: {} } })\nconst closeContextFrame = JSON.stringify({ close_context: {} })\n\n// ---------------------------------------------------------------------------\n// Wire schema (server → client)\n// ---------------------------------------------------------------------------\n\nconst ResultBody = Schema.Struct({\n  audioChunk: Schema.optional(Schema.Struct({ audioContent: Schema.String })),\n  contextCreated: Schema.optional(Schema.Unknown),\n  flushCompleted: Schema.optional(Schema.Unknown),\n  contextClosed: Schema.optional(Schema.Unknown),\n  status: Schema.optional(Schema.Unknown),\n})\n\nconst ServerFrame = Schema.Struct({\n  result: Schema.optional(ResultBody),\n  error: Schema.optional(Schema.Unknown),\n})\nconst decodeServerFrame = Schema.decodeUnknownEffect(ServerFrame)\n\nconst handleServerFrame = (queue: Queue.Queue<AudioChunk, Cause.Done>) => (raw: string) =>\n  Effect.gen(function* () {\n    const json = yield* JSONL.parseSafe(raw)\n    if (json === undefined) return\n    const decoded = yield* decodeServerFrame(json).pipe(Effect.option)\n    if (decoded._tag === \"None\") return\n    const frame = decoded.value\n    if (frame.error !== undefined) {\n      yield* Effect.logWarning(\"[inworld-tts] server error frame\", { error: frame.error })\n      return\n    }\n    const audio = frame.result?.audioChunk?.audioContent\n    if (audio === undefined || audio === \"\") return\n    const bytes = yield* decodeAudioContent(audio).pipe(Effect.option)\n    if (bytes._tag === \"Some\") yield* Queue.offer(queue, { bytes: bytes.value })\n  })\n\n// ---------------------------------------------------------------------------\n// Stream<string> → Stream<AudioChunk>\n// ---------------------------------------------------------------------------\n\nexport const streamSynthesis =\n  (cfg: Config) =>\n  <E, R>(\n    textIn: Stream.Stream<string, E, R>,\n    request: StreamSynthesizeRequest,\n  ): Stream.Stream<AudioChunk, AiError.AiError | E, R> =>\n    Stream.unwrap(\n      Effect.gen(function* () {\n        // Inline pronunciations are load-bearing (bucket 1) but there is no\n        // single text to rewrite `/ipa/` into on a chunked stream. Reject\n        // rather than silently drop; use sync `synthesize` for overrides.\n        if (request.pronunciations !== undefined && request.pronunciations.length > 0) {\n          return Stream.fail(\n            new AiError.Unsupported({\n              provider: \"inworld\",\n              capability: \"pronunciations\",\n              reason:\n                \"Inworld applies inline /ipa/ pronunciations on sync `synthesize` only; the incremental WS path streams text in chunks with no place to rewrite. Use `synthesize` for pronunciation overrides.\",\n            }),\n          )\n        }\n        const create = yield* createFrame(request)\n        const socket = yield* Socket.makeWebSocket(buildWsUrl(cfg), {\n          // Effect's Socket treats all close codes as errors by default —\n          // whitelist standard clean-close codes (1000 / 1001 / 1005).\n          closeCodeIsError: (code) => code !== 1000 && code !== 1001 && code !== 1005,\n        }).pipe(Effect.provideService(Socket.WebSocketConstructor, authedWsConstructor(cfg.apiKey)))\n        const queue = yield* Queue.bounded<AudioChunk, Cause.Done>(64)\n        const write = yield* socket.writer\n\n        // Writer: BOS `create` → drain text as `send_text` → `close_context`.\n        // The reader fiber surfaces the remaining audio drained by the server\n        // post-close and shuts the queue when the WS upstream closes.\n        yield* Effect.gen(function* () {\n          yield* write(create)\n          yield* Stream.runForEach(textIn, (text) =>\n            text.length === 0 ? Effect.void : write(sendTextFrame(text)),\n          )\n          yield* write(closeContextFrame)\n        }).pipe(Effect.ignore, Effect.forkScoped)\n\n        // `Queue.end` flushes pending chunks then fails the next take with\n        // `Done` — clean stream end. `Queue.shutdown` would CLEAR queued items\n        // and interrupt pending takes (wrong for graceful teardown).\n        yield* socket\n          .runString(handleServerFrame(queue))\n          .pipe(Effect.ensuring(Queue.end(queue)), Effect.forkScoped)\n\n        return Stream.fromQueue(queue)\n      }),\n    )\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;AAuDA,MAAa,cAAc,QAAgB;AAEzC,QAAO,IADS,IAAI,WAAW,0BAA0B,QAAQ,SAAS,KAC1D,CAAC;;AAOnB,MAAM,oBAAoB,YACxB,OAAO,IAAI,eAAe,QAAQ,gBAAgB,eAAe,QAAQ,MAAM,GAAG,SAAS;CACzF,gBAAgB,IAAI;CACpB,GAAI,IAAI,oBAAoB,KAAA,KAAa,EAAE,mBAAmB,IAAI,iBAAiB;CACnF,GAAI,IAAI,YAAY,KAAA,KAAa,EAAE,UAAU,IAAI,SAAS;CAC1D,GAAI,IAAI,iBAAiB,KAAA,KAAa,EAAE,eAAe,IAAI,cAAc;CAC1E,EAAE;AAEL,MAAM,eAAe,YACnB,OAAO,IAAI,iBAAiB,QAAQ,GAAG,iBACrC,KAAK,UAAU,EACb,QAAQ;CACN,UAAU,QAAQ;CAClB,UAAU,QAAQ;CAClB;CACA,GAAI,QAAQ,iBAAiB,KAAA,KAAa,EAAE,UAAU,QAAQ,cAAc;CAC5E,GAAI,QAAQ,iBAAiB,KAAA,KAAa,EAAE,eAAe,QAAQ,cAAc;CACjF,GAAI,QAAQ,gBAAgB,KAAA,KAAa,EAAE,aAAa,QAAQ,aAAa;CAC7E,GAAI,QAAQ,2BAA2B,KAAA,KAAa,EAClD,0BAA0B,QAAQ,wBACnC;CACF,EACF,CAAC,CACH;AAIH,MAAM,iBAAiB,SAAiB,KAAK,UAAU,EAAE,WAAW;CAAE;CAAM,eAAe,EAAE;CAAE,EAAE,CAAC;AAClG,MAAM,oBAAoB,KAAK,UAAU,EAAE,eAAe,EAAE,EAAE,CAAC;AAM/D,MAAM,aAAa,OAAO,OAAO;CAC/B,YAAY,OAAO,SAAS,OAAO,OAAO,EAAE,cAAc,OAAO,QAAQ,CAAC,CAAC;CAC3E,gBAAgB,OAAO,SAAS,OAAO,QAAQ;CAC/C,gBAAgB,OAAO,SAAS,OAAO,QAAQ;CAC/C,eAAe,OAAO,SAAS,OAAO,QAAQ;CAC9C,QAAQ,OAAO,SAAS,OAAO,QAAQ;CACxC,CAAC;AAEF,MAAM,cAAc,OAAO,OAAO;CAChC,QAAQ,OAAO,SAAS,WAAW;CACnC,OAAO,OAAO,SAAS,OAAO,QAAQ;CACvC,CAAC;AACF,MAAM,oBAAoB,OAAO,oBAAoB,YAAY;AAEjE,MAAM,qBAAqB,WAAgD,QACzE,OAAO,IAAI,aAAa;CACtB,MAAM,OAAO,OAAO,MAAM,UAAU,IAAI;AACxC,KAAI,SAAS,KAAA,EAAW;CACxB,MAAM,UAAU,OAAO,kBAAkB,KAAK,CAAC,KAAK,OAAO,OAAO;AAClE,KAAI,QAAQ,SAAS,OAAQ;CAC7B,MAAM,QAAQ,QAAQ;AACtB,KAAI,MAAM,UAAU,KAAA,GAAW;AAC7B,SAAO,OAAO,WAAW,oCAAoC,EAAE,OAAO,MAAM,OAAO,CAAC;AACpF;;CAEF,MAAM,QAAQ,MAAM,QAAQ,YAAY;AACxC,KAAI,UAAU,KAAA,KAAa,UAAU,GAAI;CACzC,MAAM,QAAQ,OAAO,mBAAmB,MAAM,CAAC,KAAK,OAAO,OAAO;AAClE,KAAI,MAAM,SAAS,OAAQ,QAAO,MAAM,MAAM,OAAO,EAAE,OAAO,MAAM,OAAO,CAAC;EAC5E;AAMJ,MAAa,mBACV,SAEC,QACA,YAEA,OAAO,OACL,OAAO,IAAI,aAAa;AAItB,KAAI,QAAQ,mBAAmB,KAAA,KAAa,QAAQ,eAAe,SAAS,EAC1E,QAAO,OAAO,KACZ,IAAI,QAAQ,YAAY;EACtB,UAAU;EACV,YAAY;EACZ,QACE;EACH,CAAC,CACH;CAEH,MAAM,SAAS,OAAO,YAAY,QAAQ;CAC1C,MAAM,SAAS,OAAO,OAAO,cAAc,WAAW,IAAI,EAAE,EAG1D,mBAAmB,SAAS,SAAS,OAAQ,SAAS,QAAQ,SAAS,MACxE,CAAC,CAAC,KAAK,OAAO,eAAe,OAAO,sBAAsB,oBAAoB,IAAI,OAAO,CAAC,CAAC;CAC5F,MAAM,QAAQ,OAAO,MAAM,QAAgC,GAAG;CAC9D,MAAM,QAAQ,OAAO,OAAO;AAK5B,QAAO,OAAO,IAAI,aAAa;AAC7B,SAAO,MAAM,OAAO;AACpB,SAAO,OAAO,WAAW,SAAS,SAChC,KAAK,WAAW,IAAI,OAAO,OAAO,MAAM,cAAc,KAAK,CAAC,CAC7D;AACD,SAAO,MAAM,kBAAkB;GAC/B,CAAC,KAAK,OAAO,QAAQ,OAAO,WAAW;AAKzC,QAAO,OACJ,UAAU,kBAAkB,MAAM,CAAC,CACnC,KAAK,OAAO,SAAS,MAAM,IAAI,MAAM,CAAC,EAAE,OAAO,WAAW;AAE7D,QAAO,OAAO,UAAU,MAAM;EAC9B,CACH"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@effect-uai/inworld",
-  "version": "0.6.0",
+  "version": "0.7.0",
   "description": "Inworld speech-to-text and text-to-speech provider for @effect-uai/core.",
   "keywords": [
     "ai",
@@ -61,7 +61,7 @@
     "effect": "4.0.0-beta.57",
     "typescript": "^6.0.3",
     "ws": "^8.18.0",
-    "@effect-uai/core": "0.6.0"
+    "@effect-uai/core": "0.7.0"
   },
   "peerDependencies": {
     "@effect-uai/core": ">=0.2.0 <1",

package/src/InworldSynthesizer.ts

@@ -76,9 +76,9 @@ export type Config = { readonly apiKey: Redacted.Redacted; readonly baseUrl?: st
 
 /**
  * Inworld accepts inline phoneme overrides as `/ipa/` tokens directly
- * in the text — no separate field. Only IPA is supported; entries with
- * other encodings are silently dropped (audio still renders with the
- * default pronunciation). Whole-word, case-insensitive replacement.
+ * in the text, no separate field. `CustomPronunciation.pronunciation`
+ * is always IPA, so every entry is inlined. Whole-word,
+ * case-insensitive replacement.
  */
 const escapeForRegex = (s: string): string => s.replace(/[.*+?^${}()|[\]\\]/g, "\\$&")
 
@@ -88,9 +88,7 @@ const applyPronunciations = (
 ): string => {
   if (pronunciations === undefined) return text
   return Arr.reduce(pronunciations, text, (acc, p) =>
-    p.encoding === "ipa"
-      ? acc.replace(new RegExp(`\\b${escapeForRegex(p.phrase)}\\b`, "i"), `/${p.pronunciation}/`)
-      : acc,
+    acc.replace(new RegExp(`\\b${escapeForRegex(p.phrase)}\\b`, "i"), `/${p.pronunciation}/`),
   )
 }
 

package/src/InworldTranscriber.ts

@@ -12,6 +12,7 @@ import { Context, Effect, Encoding, Layer, Match, Redacted, Schema, Stream } fro
 import { HttpClient, HttpClientRequest } from "effect/unstable/http"
 import * as AiError from "@effect-uai/core/AiError"
 import type { AudioSource } from "@effect-uai/core/Audio"
+import * as Capabilities from "@effect-uai/core/Capabilities"
 import type { TranscriptEvent, TranscriptResult, WordTimestamp } from "@effect-uai/core/Transcript"
 import {
   type CommonStreamTranscribeRequest,
@@ -74,15 +75,17 @@ const audioToBase64: (audio: AudioSource) => Effect.Effect<string, AiError.AiErr
 // Request body
 // ---------------------------------------------------------------------------
 
-const promptToTerms = (
-  prompt: string | { readonly terms: ReadonlyArray<string> } | undefined,
-): ReadonlyArray<string> | undefined =>
-  prompt === undefined ? undefined : typeof prompt === "string" ? [prompt] : prompt.terms
-
 const buildBody = (request: InworldTranscribeRequest) =>
   Effect.gen(function* () {
     const content = yield* audioToBase64(request.audio)
-    const prompts = promptToTerms(request.prompt)
+    // Inworld's `prompts` is a vocab-biasing term list — maps from
+    // `biasingTerms`. There's no free-form prompt field, so `prompt` warns.
+    yield* Capabilities.warnDroppedWhen(request.prompt, {
+      provider: "inworld",
+      capability: "prompt",
+      field: "prompt",
+      reason: "Inworld STT has no free-form prompt field; bias via `biasingTerms`.",
+    })
     return {
       transcribeConfig: {
         modelId: request.model,
@@ -92,7 +95,7 @@ const buildBody = (request: InworldTranscribeRequest) =>
         }),
         numberOfChannels: request.numberOfChannels ?? 1,
         ...(request.language !== undefined && { language: request.language }),
-        ...(prompts !== undefined && { prompts }),
+        ...(request.biasingTerms !== undefined && { prompts: request.biasingTerms }),
         ...(request.wordTimestamps === true && { includeWordTimestamps: true }),
       },
       audioData: { content },

package/src/realtimeStt.ts

@@ -22,6 +22,7 @@ import { Cause, Effect, Encoding, Match, Queue, Redacted, Schema, Stream } from
 import * as Socket from "effect/unstable/socket/Socket"
 import * as AiError from "@effect-uai/core/AiError"
 import type { AudioFormat } from "@effect-uai/core/Audio"
+import * as Capabilities from "@effect-uai/core/Capabilities"
 import * as JSONL from "@effect-uai/core/JSONL"
 import type { TranscriptEvent, WordTimestamp } from "@effect-uai/core/Transcript"
 import type { CommonStreamTranscribeRequest } from "@effect-uai/core/Transcriber"
@@ -67,14 +68,8 @@ const buildWsUrl = (cfg: Config) => {
   return `${wsBase}/stt/v1/transcribe:streamBidirectional`
 }
 
-const promptToTerms = (
-  prompt: string | { readonly terms: ReadonlyArray<string> } | undefined,
-): ReadonlyArray<string> | undefined =>
-  prompt === undefined ? undefined : typeof prompt === "string" ? [prompt] : prompt.terms
-
-const configFrame = (encoding: WireEncoding, request: CommonStreamTranscribeRequest) => {
-  const prompts = promptToTerms(request.prompt)
-  return JSON.stringify({
+const configFrame = (encoding: WireEncoding, request: CommonStreamTranscribeRequest) =>
+  JSON.stringify({
     transcribeConfig: {
       modelId: request.model,
       audioEncoding: encoding,
@@ -83,11 +78,11 @@ const configFrame = (encoding: WireEncoding, request: CommonStreamTranscribeRequ
       // Inworld's sample includes `language` even though docs mark it optional.
       // Default to en-US to match the sample's behavior; caller can override.
       language: request.language ?? "en-US",
-      ...(prompts !== undefined && { prompts }),
+      // `prompts` is a vocab-biasing term list — maps from `biasingTerms`.
+      ...(request.biasingTerms !== undefined && { prompts: request.biasingTerms }),
       ...(request.wordTimestamps === true && { includeWordTimestamps: true }),
     },
   })
-}
 
 const audioChunkFrame = (bytes: Uint8Array) =>
   JSON.stringify({ audioChunk: { content: Encoding.encodeBase64(bytes) } })
@@ -204,6 +199,12 @@ export const streamTranscription =
   ): Stream.Stream<TranscriptEvent, AiError.AiError | E, R> =>
     Stream.unwrap(
       Effect.gen(function* () {
+        yield* Capabilities.warnDroppedWhen(request.prompt, {
+          provider: "inworld",
+          capability: "prompt",
+          field: "prompt",
+          reason: "Inworld STT has no free-form prompt field; bias via `biasingTerms`.",
+        })
         const encoding = yield* inputFormatToWire(request.inputFormat)
         const socket = yield* Socket.makeWebSocket(buildWsUrl(cfg), {
           // Effect's Socket treats all close codes as errors by default —

package/src/realtimeTts.ts

@@ -24,6 +24,7 @@ import * as Socket from "effect/unstable/socket/Socket"
 import * as AiError from "@effect-uai/core/AiError"
 import type { AudioChunk, AudioFormat } from "@effect-uai/core/Audio"
 import * as JSONL from "@effect-uai/core/JSONL"
+import type { CustomPronunciation } from "@effect-uai/core/SpeechSynthesizer"
 import { audioConfigFor, decodeAudioContent, defaultFormat } from "./codec.js"
 import type { InworldDeliveryMode, InworldTtsModel, InworldVoiceId } from "./models.js"
 import { authedWsConstructor } from "./wsAuth.js"
@@ -43,6 +44,9 @@ export type StreamSynthesizeRequest = {
   readonly deliveryMode?: InworldDeliveryMode
   readonly applyTextNormalization?: "ON" | "OFF"
   readonly speed?: number
+  /** Carried from the Common request so the WS path can reject inline
+   *  pronunciations (no place to rewrite into a chunked text stream). */
+  readonly pronunciations?: ReadonlyArray<CustomPronunciation>
 }
 
 // ---------------------------------------------------------------------------
@@ -135,6 +139,19 @@ export const streamSynthesis =
   ): Stream.Stream<AudioChunk, AiError.AiError | E, R> =>
     Stream.unwrap(
       Effect.gen(function* () {
+        // Inline pronunciations are load-bearing (bucket 1) but there is no
+        // single text to rewrite `/ipa/` into on a chunked stream. Reject
+        // rather than silently drop; use sync `synthesize` for overrides.
+        if (request.pronunciations !== undefined && request.pronunciations.length > 0) {
+          return Stream.fail(
+            new AiError.Unsupported({
+              provider: "inworld",
+              capability: "pronunciations",
+              reason:
+                "Inworld applies inline /ipa/ pronunciations on sync `synthesize` only; the incremental WS path streams text in chunks with no place to rewrite. Use `synthesize` for pronunciation overrides.",
+            }),
+          )
+        }
         const create = yield* createFrame(request)
         const socket = yield* Socket.makeWebSocket(buildWsUrl(cfg), {
           // Effect's Socket treats all close codes as errors by default —