@effect-uai/elevenlabs 0.6.0 → 0.8.0

package/dist/realtimeStt.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"realtimeStt.mjs","names":[],"sources":["../src/realtimeStt.ts"],"sourcesContent":["import { Cause, Effect, Encoding, Match, Queue, Redacted, Schema, Stream } from \"effect\"\nimport { HttpClient, HttpClientRequest } from \"effect/unstable/http\"\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 { httpStatusError, transportFailure } from \"./codec.js\"\nimport { type ElevenLabsRegion, resolveHost } from \"./region.js\"\n\nexport type Config = {\n  readonly apiKey: Redacted.Redacted\n  readonly baseUrl?: string\n  readonly region?: ElevenLabsRegion\n}\n\n// ---------------------------------------------------------------------------\n// AudioFormat → audio_format slug\n// ---------------------------------------------------------------------------\n\nconst unsupportedInputFormat = (format: AudioFormat) =>\n  new AiError.Unsupported({\n    provider: \"elevenlabs\",\n    capability: \"inputFormat\",\n    reason: `ElevenLabs realtime STT accepts pcm_<rate> or ulaw_8000 only. Got ${JSON.stringify(format)}.`,\n  })\n\nexport const inputFormatToSlug: (format: AudioFormat) => Effect.Effect<string, AiError.AiError> =\n  Match.type<AudioFormat>().pipe(\n    Match.when({ container: \"raw\", encoding: \"pcm_s16le\" }, (f) =>\n      Effect.succeed(`pcm_${f.sampleRate}`),\n    ),\n    Match.when({ container: \"raw\", encoding: \"pcm_mulaw\", sampleRate: 8000 }, () =>\n      Effect.succeed(\"ulaw_8000\"),\n    ),\n    Match.orElse((f) => Effect.fail(unsupportedInputFormat(f))),\n  )\n\n// ---------------------------------------------------------------------------\n// Single-use token — auth rides on a query param so any\n// `globalThis.WebSocket` works (browser, Node 22+, Bun, Deno).\n// ---------------------------------------------------------------------------\n\nconst TokenWire = Schema.Struct({ token: Schema.String })\nconst decodeToken = Schema.decodeUnknownEffect(TokenWire)\n\nexport const fetchSingleUseToken = (cfg: Config) =>\n  Effect.gen(function* () {\n    const client = yield* HttpClient.HttpClient\n    const url = `${resolveHost(cfg)}/single-use-token/realtime_scribe`\n    const httpRequest = HttpClientRequest.post(url).pipe(\n      HttpClientRequest.setHeader(\"xi-api-key\", Redacted.value(cfg.apiKey)),\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* decodeToken(json).pipe(\n      Effect.mapError(\n        (cause) =>\n          new AiError.GenerationFailed({\n            provider: \"elevenlabs\",\n            raw: { message: \"single-use-token response missing `token` field\", cause, json },\n          }),\n      ),\n    )\n    return wire.token\n  })\n\n// ---------------------------------------------------------------------------\n// Wire schemas (server → client)\n// ---------------------------------------------------------------------------\n\nconst RealtimeWord = Schema.Struct({\n  text: Schema.String,\n  start: Schema.Number,\n  end: Schema.Number,\n  type: Schema.optional(Schema.String),\n  speaker_id: Schema.optional(Schema.NullOr(Schema.String)),\n  logprob: Schema.optional(Schema.Number),\n})\n\nconst ServerMessage = Schema.Union([\n  Schema.Struct({\n    message_type: Schema.Literal(\"session_started\"),\n    session_id: Schema.optional(Schema.String),\n  }),\n  Schema.Struct({\n    message_type: Schema.Literal(\"partial_transcript\"),\n    text: Schema.String,\n  }),\n  Schema.Struct({\n    message_type: Schema.Literal(\"committed_transcript\"),\n    text: Schema.String,\n  }),\n  Schema.Struct({\n    message_type: Schema.Literal(\"committed_transcript_with_timestamps\"),\n    text: Schema.String,\n    language_code: Schema.optional(Schema.NullOr(Schema.String)),\n    words: Schema.optional(Schema.NullOr(Schema.Array(RealtimeWord))),\n  }),\n  Schema.Struct({\n    message_type: Schema.Literals([\"error\", \"auth_error\", \"rate_limited\", \"quota_exceeded\"]),\n    error: Schema.optional(Schema.Unknown),\n  }),\n])\nconst decodeServerMessage = Schema.decodeUnknownEffect(ServerMessage)\n\nconst wireToWordTimestamp = (w: typeof RealtimeWord.Type): WordTimestamp => ({\n  text: w.text,\n  startSeconds: w.start,\n  endSeconds: w.end,\n  ...(w.speaker_id != null && { speakerId: w.speaker_id }),\n  ...(w.logprob !== undefined && { confidence: Math.exp(w.logprob) }),\n})\n\n/**\n * Map a server frame to a `TranscriptEvent`. The two `committed_*` variants\n * need disambiguation:\n *\n *   - When `include_timestamps=true` on the URL, the server emits BOTH\n *     `committed_transcript` and `committed_transcript_with_timestamps` for\n *     each segment — same text, the latter is a strict superset. The bare\n *     variant must be suppressed or callers see two finals per utterance.\n *   - When `include_timestamps=false`, the server emits ONLY\n *     `committed_transcript`. Suppressing it loses every final.\n *\n * So suppression must be conditional on `includeTimestamps`, which is what\n * `wordTimestamps` resolves to on the URL.\n */\nexport const wireToEvent =\n  (includeTimestamps: boolean) =>\n  (msg: typeof ServerMessage.Type): TranscriptEvent | undefined =>\n    Match.value(msg).pipe(\n      Match.when({ message_type: \"session_started\" }, () => undefined),\n      Match.when(\n        { message_type: \"partial_transcript\" },\n        (m): TranscriptEvent => ({ _tag: \"partial\", text: m.text }),\n      ),\n      Match.when({ message_type: \"committed_transcript\" }, (m): TranscriptEvent | undefined =>\n        includeTimestamps\n          ? undefined // superseded by the `_with_timestamps` variant\n          : { _tag: \"final\", text: m.text },\n      ),\n      Match.when({ message_type: \"committed_transcript_with_timestamps\" }, (m): TranscriptEvent => {\n        const words = m.words == null ? undefined : m.words.map(wireToWordTimestamp)\n        return {\n          _tag: \"final\",\n          text: m.text,\n          ...(m.language_code != null && { languageCode: m.language_code }),\n          ...(words !== undefined && words.length > 0 && { words }),\n        }\n      }),\n      Match.orElse(\n        (m): TranscriptEvent => ({\n          _tag: \"error\",\n          code: m.message_type,\n          message:\n            typeof m.error === \"string\"\n              ? m.error\n              : JSON.stringify(m.error ?? `ElevenLabs ${m.message_type}`),\n        }),\n      ),\n    )\n\n// ---------------------------------------------------------------------------\n// URL + frame builders\n// ---------------------------------------------------------------------------\n\nexport type RealtimeOptions = {\n  readonly model?: string\n  readonly languageCode?: string\n  readonly includeTimestamps?: boolean\n  readonly commitStrategy?: \"manual\" | \"vad\"\n}\n\nexport const buildWsUrl = (\n  cfg: Config,\n  token: string,\n  audioFormat: string,\n  opts: RealtimeOptions,\n) => {\n  const wsBase = resolveHost(cfg).replace(/^http/, \"ws\")\n  const params = new URLSearchParams({\n    token,\n    audio_format: audioFormat,\n    commit_strategy: opts.commitStrategy ?? \"vad\",\n    ...(opts.model !== undefined && { model_id: opts.model }),\n    ...(opts.languageCode !== undefined && { language_code: opts.languageCode }),\n    ...(opts.includeTimestamps === true && { include_timestamps: \"true\" }),\n  })\n  return `${wsBase}/speech-to-text/realtime?${params.toString()}`\n}\n\nexport const encodeAudioFrame = (bytes: Uint8Array, sampleRate: number) =>\n  JSON.stringify({\n    message_type: \"input_audio_chunk\",\n    audio_base_64: Encoding.encodeBase64(bytes),\n    sample_rate: sampleRate,\n  })\n\n// ---------------------------------------------------------------------------\n// Parse one server message and emit to the queue if it maps to a\n// TranscriptEvent. Invalid JSON / unknown shapes are silently dropped —\n// the connection stays open and we keep streaming subsequent messages.\n// ---------------------------------------------------------------------------\n\nconst handleServerMessage =\n  (\n    queue: Queue.Queue<TranscriptEvent, Cause.Done>,\n    mapEvent: (msg: typeof ServerMessage.Type) => TranscriptEvent | undefined,\n  ) =>\n  (raw: string) =>\n    Effect.gen(function* () {\n      const json = yield* JSONL.parseSafe(raw)\n      if (json === undefined) return\n      const decoded = yield* decodeServerMessage(json).pipe(Effect.option)\n      if (decoded._tag === \"None\") return\n      const event = mapEvent(decoded.value)\n      if (event !== undefined) yield* Queue.offer(queue, event)\n    })\n\n// ---------------------------------------------------------------------------\n// Stream<Uint8Array> → Stream<TranscriptEvent>\n//\n// Requires `HttpClient` (single-use token fetch) and\n// `Socket.WebSocketConstructor` (any `globalThis.WebSocket` works) in\n// context — `Socket.layerWebSocketConstructorGlobal` covers every\n// modern runtime.\n// ---------------------------------------------------------------------------\n\nexport const streamTranscription =\n  (cfg: Config) =>\n  <E, R>(\n    audioIn: Stream.Stream<Uint8Array, E, R>,\n    request: CommonStreamTranscribeRequest,\n  ): Stream.Stream<\n    TranscriptEvent,\n    AiError.AiError | E,\n    R | HttpClient.HttpClient | Socket.WebSocketConstructor\n  > =>\n    Stream.unwrap(\n      Effect.gen(function* () {\n        const slug = yield* inputFormatToSlug(request.inputFormat)\n        const token = yield* fetchSingleUseToken(cfg)\n        const includeTimestamps = request.wordTimestamps === true\n        const url = buildWsUrl(cfg, token, slug, {\n          ...(request.model !== undefined && { model: request.model }),\n          ...(request.language !== undefined && { languageCode: request.language }),\n          includeTimestamps,\n        })\n        const socket = yield* Socket.makeWebSocket(url, {\n          // Effect's Socket treats all close codes as errors by default — that\n          // surfaces a clean server-side close as a stream failure. Whitelist\n          // standard clean-close codes (1000 / 1001 / 1005).\n          closeCodeIsError: (code) => code !== 1000 && code !== 1001 && code !== 1005,\n        })\n        const queue = yield* Queue.bounded<TranscriptEvent, Cause.Done>(64)\n        const sampleRate = request.inputFormat.sampleRate\n\n        const write = yield* socket.writer\n        yield* Stream.runForEach(audioIn, (bytes) =>\n          write(encodeAudioFrame(bytes, sampleRate)),\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, wireToEvent(includeTimestamps)))\n          .pipe(Effect.ensuring(Queue.end(queue)), Effect.forkScoped)\n\n        return Stream.fromQueue(queue)\n      }),\n    )\n"],"mappings":";;;;;;;;AAqBA,MAAM,0BAA0B,WAC9B,IAAI,QAAQ,YAAY;CACtB,UAAU;CACV,YAAY;CACZ,QAAQ,qEAAqE,KAAK,UAAU,OAAO,CAAC;CACrG,CAAC;AAEJ,MAAa,oBACX,MAAM,MAAmB,CAAC,KACxB,MAAM,KAAK;CAAE,WAAW;CAAO,UAAU;CAAa,GAAG,MACvD,OAAO,QAAQ,OAAO,EAAE,aAAa,CACtC,EACD,MAAM,KAAK;CAAE,WAAW;CAAO,UAAU;CAAa,YAAY;CAAM,QACtE,OAAO,QAAQ,YAAY,CAC5B,EACD,MAAM,QAAQ,MAAM,OAAO,KAAK,uBAAuB,EAAE,CAAC,CAAC,CAC5D;AAOH,MAAM,YAAY,OAAO,OAAO,EAAE,OAAO,OAAO,QAAQ,CAAC;AACzD,MAAM,cAAc,OAAO,oBAAoB,UAAU;AAEzD,MAAa,uBAAuB,QAClC,OAAO,IAAI,aAAa;CACtB,MAAM,SAAS,OAAO,WAAW;CACjC,MAAM,MAAM,GAAG,YAAY,IAAI,CAAC;CAChC,MAAM,cAAc,kBAAkB,KAAK,IAAI,CAAC,KAC9C,kBAAkB,UAAU,cAAc,SAAS,MAAM,IAAI,OAAO,CAAC,CACtE;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;AAUzE,SAAO,OATa,YAAY,KAAK,CAAC,KACpC,OAAO,UACJ,UACC,IAAI,QAAQ,iBAAiB;EAC3B,UAAU;EACV,KAAK;GAAE,SAAS;GAAmD;GAAO;GAAM;EACjF,CAAC,CACL,CACF,EACW;EACZ;AAMJ,MAAM,eAAe,OAAO,OAAO;CACjC,MAAM,OAAO;CACb,OAAO,OAAO;CACd,KAAK,OAAO;CACZ,MAAM,OAAO,SAAS,OAAO,OAAO;CACpC,YAAY,OAAO,SAAS,OAAO,OAAO,OAAO,OAAO,CAAC;CACzD,SAAS,OAAO,SAAS,OAAO,OAAO;CACxC,CAAC;AAEF,MAAM,gBAAgB,OAAO,MAAM;CACjC,OAAO,OAAO;EACZ,cAAc,OAAO,QAAQ,kBAAkB;EAC/C,YAAY,OAAO,SAAS,OAAO,OAAO;EAC3C,CAAC;CACF,OAAO,OAAO;EACZ,cAAc,OAAO,QAAQ,qBAAqB;EAClD,MAAM,OAAO;EACd,CAAC;CACF,OAAO,OAAO;EACZ,cAAc,OAAO,QAAQ,uBAAuB;EACpD,MAAM,OAAO;EACd,CAAC;CACF,OAAO,OAAO;EACZ,cAAc,OAAO,QAAQ,uCAAuC;EACpE,MAAM,OAAO;EACb,eAAe,OAAO,SAAS,OAAO,OAAO,OAAO,OAAO,CAAC;EAC5D,OAAO,OAAO,SAAS,OAAO,OAAO,OAAO,MAAM,aAAa,CAAC,CAAC;EAClE,CAAC;CACF,OAAO,OAAO;EACZ,cAAc,OAAO,SAAS;GAAC;GAAS;GAAc;GAAgB;GAAiB,CAAC;EACxF,OAAO,OAAO,SAAS,OAAO,QAAQ;EACvC,CAAC;CACH,CAAC;AACF,MAAM,sBAAsB,OAAO,oBAAoB,cAAc;AAErE,MAAM,uBAAuB,OAAgD;CAC3E,MAAM,EAAE;CACR,cAAc,EAAE;CAChB,YAAY,EAAE;CACd,GAAI,EAAE,cAAc,QAAQ,EAAE,WAAW,EAAE,YAAY;CACvD,GAAI,EAAE,YAAY,KAAA,KAAa,EAAE,YAAY,KAAK,IAAI,EAAE,QAAQ,EAAE;CACnE;;;;;;;;;;;;;;;AAgBD,MAAa,eACV,uBACA,QACC,MAAM,MAAM,IAAI,CAAC,KACf,MAAM,KAAK,EAAE,cAAc,mBAAmB,QAAQ,KAAA,EAAU,EAChE,MAAM,KACJ,EAAE,cAAc,sBAAsB,GACrC,OAAwB;CAAE,MAAM;CAAW,MAAM,EAAE;CAAM,EAC3D,EACD,MAAM,KAAK,EAAE,cAAc,wBAAwB,GAAG,MACpD,oBACI,KAAA,IACA;CAAE,MAAM;CAAS,MAAM,EAAE;CAAM,CACpC,EACD,MAAM,KAAK,EAAE,cAAc,wCAAwC,GAAG,MAAuB;CAC3F,MAAM,QAAQ,EAAE,SAAS,OAAO,KAAA,IAAY,EAAE,MAAM,IAAI,oBAAoB;AAC5E,QAAO;EACL,MAAM;EACN,MAAM,EAAE;EACR,GAAI,EAAE,iBAAiB,QAAQ,EAAE,cAAc,EAAE,eAAe;EAChE,GAAI,UAAU,KAAA,KAAa,MAAM,SAAS,KAAK,EAAE,OAAO;EACzD;EACD,EACF,MAAM,QACH,OAAwB;CACvB,MAAM;CACN,MAAM,EAAE;CACR,SACE,OAAO,EAAE,UAAU,WACf,EAAE,QACF,KAAK,UAAU,EAAE,SAAS,cAAc,EAAE,eAAe;CAChE,EACF,CACF;AAaL,MAAa,cACX,KACA,OACA,aACA,SACG;AAUH,QAAO,GATQ,YAAY,IAAI,CAAC,QAAQ,SAAS,KASjC,CAAC,2BAA2B,IARzB,gBAAgB;EACjC;EACA,cAAc;EACd,iBAAiB,KAAK,kBAAkB;EACxC,GAAI,KAAK,UAAU,KAAA,KAAa,EAAE,UAAU,KAAK,OAAO;EACxD,GAAI,KAAK,iBAAiB,KAAA,KAAa,EAAE,eAAe,KAAK,cAAc;EAC3E,GAAI,KAAK,sBAAsB,QAAQ,EAAE,oBAAoB,QAAQ;EACtE,CACiD,CAAC,UAAU;;AAG/D,MAAa,oBAAoB,OAAmB,eAClD,KAAK,UAAU;CACb,cAAc;CACd,eAAe,SAAS,aAAa,MAAM;CAC3C,aAAa;CACd,CAAC;AAQJ,MAAM,uBAEF,OACA,cAED,QACC,OAAO,IAAI,aAAa;CACtB,MAAM,OAAO,OAAO,MAAM,UAAU,IAAI;AACxC,KAAI,SAAS,KAAA,EAAW;CACxB,MAAM,UAAU,OAAO,oBAAoB,KAAK,CAAC,KAAK,OAAO,OAAO;AACpE,KAAI,QAAQ,SAAS,OAAQ;CAC7B,MAAM,QAAQ,SAAS,QAAQ,MAAM;AACrC,KAAI,UAAU,KAAA,EAAW,QAAO,MAAM,MAAM,OAAO,MAAM;EACzD;AAWN,MAAa,uBACV,SAEC,SACA,YAMA,OAAO,OACL,OAAO,IAAI,aAAa;CACtB,MAAM,OAAO,OAAO,kBAAkB,QAAQ,YAAY;CAC1D,MAAM,QAAQ,OAAO,oBAAoB,IAAI;CAC7C,MAAM,oBAAoB,QAAQ,mBAAmB;CACrD,MAAM,MAAM,WAAW,KAAK,OAAO,MAAM;EACvC,GAAI,QAAQ,UAAU,KAAA,KAAa,EAAE,OAAO,QAAQ,OAAO;EAC3D,GAAI,QAAQ,aAAa,KAAA,KAAa,EAAE,cAAc,QAAQ,UAAU;EACxE;EACD,CAAC;CACF,MAAM,SAAS,OAAO,OAAO,cAAc,KAAK,EAI9C,mBAAmB,SAAS,SAAS,OAAQ,SAAS,QAAQ,SAAS,MACxE,CAAC;CACF,MAAM,QAAQ,OAAO,MAAM,QAAqC,GAAG;CACnE,MAAM,aAAa,QAAQ,YAAY;CAEvC,MAAM,QAAQ,OAAO,OAAO;AAC5B,QAAO,OAAO,WAAW,UAAU,UACjC,MAAM,iBAAiB,OAAO,WAAW,CAAC,CAC3C,CAAC,KAAK,OAAO,QAAQ,OAAO,WAAW;AAKxC,QAAO,OACJ,UAAU,oBAAoB,OAAO,YAAY,kBAAkB,CAAC,CAAC,CACrE,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":["import { Cause, Effect, Encoding, Match, Queue, Redacted, Schema, Stream } from \"effect\"\nimport { HttpClient, HttpClientRequest } from \"effect/unstable/http\"\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 { httpStatusError, transportFailure } from \"./codec.js\"\nimport { type ElevenLabsRegion, resolveHost } from \"./region.js\"\n\nexport type Config = {\n  readonly apiKey: Redacted.Redacted\n  readonly baseUrl?: string\n  readonly region?: ElevenLabsRegion\n}\n\n// ---------------------------------------------------------------------------\n// AudioFormat → audio_format slug\n// ---------------------------------------------------------------------------\n\nconst unsupportedInputFormat = (format: AudioFormat) =>\n  new AiError.Unsupported({\n    provider: \"elevenlabs\",\n    capability: \"inputFormat\",\n    reason: `ElevenLabs realtime STT accepts pcm_<rate> or ulaw_8000 only. Got ${JSON.stringify(format)}.`,\n  })\n\nexport const inputFormatToSlug: (format: AudioFormat) => Effect.Effect<string, AiError.AiError> =\n  Match.type<AudioFormat>().pipe(\n    Match.when({ container: \"raw\", encoding: \"pcm_s16le\" }, (f) =>\n      Effect.succeed(`pcm_${f.sampleRate}`),\n    ),\n    Match.when({ container: \"raw\", encoding: \"pcm_mulaw\", sampleRate: 8000 }, () =>\n      Effect.succeed(\"ulaw_8000\"),\n    ),\n    Match.orElse((f) => Effect.fail(unsupportedInputFormat(f))),\n  )\n\n// ---------------------------------------------------------------------------\n// Single-use token — auth rides on a query param so any\n// `globalThis.WebSocket` works (browser, Node 22+, Bun, Deno).\n// ---------------------------------------------------------------------------\n\nconst TokenWire = Schema.Struct({ token: Schema.String })\nconst decodeToken = Schema.decodeUnknownEffect(TokenWire)\n\nexport const fetchSingleUseToken = (cfg: Config) =>\n  Effect.gen(function* () {\n    const client = yield* HttpClient.HttpClient\n    const url = `${resolveHost(cfg)}/single-use-token/realtime_scribe`\n    const httpRequest = HttpClientRequest.post(url).pipe(\n      HttpClientRequest.setHeader(\"xi-api-key\", Redacted.value(cfg.apiKey)),\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* decodeToken(json).pipe(\n      Effect.mapError(\n        (cause) =>\n          new AiError.GenerationFailed({\n            provider: \"elevenlabs\",\n            raw: { message: \"single-use-token response missing `token` field\", cause, json },\n          }),\n      ),\n    )\n    return wire.token\n  })\n\n// ---------------------------------------------------------------------------\n// Wire schemas (server → client)\n// ---------------------------------------------------------------------------\n\nconst RealtimeWord = Schema.Struct({\n  text: Schema.String,\n  start: Schema.Number,\n  end: Schema.Number,\n  type: Schema.optional(Schema.String),\n  speaker_id: Schema.optional(Schema.NullOr(Schema.String)),\n  logprob: Schema.optional(Schema.Number),\n})\n\nconst ServerMessage = Schema.Union([\n  Schema.Struct({\n    message_type: Schema.Literal(\"session_started\"),\n    session_id: Schema.optional(Schema.String),\n  }),\n  Schema.Struct({\n    message_type: Schema.Literal(\"partial_transcript\"),\n    text: Schema.String,\n  }),\n  Schema.Struct({\n    message_type: Schema.Literal(\"committed_transcript\"),\n    text: Schema.String,\n  }),\n  Schema.Struct({\n    message_type: Schema.Literal(\"committed_transcript_with_timestamps\"),\n    text: Schema.String,\n    language_code: Schema.optional(Schema.NullOr(Schema.String)),\n    words: Schema.optional(Schema.NullOr(Schema.Array(RealtimeWord))),\n  }),\n  Schema.Struct({\n    message_type: Schema.Literals([\"error\", \"auth_error\", \"rate_limited\", \"quota_exceeded\"]),\n    error: Schema.optional(Schema.Unknown),\n  }),\n])\nconst decodeServerMessage = Schema.decodeUnknownEffect(ServerMessage)\n\nconst wireToWordTimestamp = (w: typeof RealtimeWord.Type): WordTimestamp => ({\n  text: w.text,\n  startSeconds: w.start,\n  endSeconds: w.end,\n  ...(w.speaker_id != null && { speakerId: w.speaker_id }),\n  ...(w.logprob !== undefined && { confidence: Math.exp(w.logprob) }),\n})\n\n/**\n * Map a server frame to a `TranscriptEvent`. The two `committed_*` variants\n * need disambiguation:\n *\n *   - When `include_timestamps=true` on the URL, the server emits BOTH\n *     `committed_transcript` and `committed_transcript_with_timestamps` for\n *     each segment — same text, the latter is a strict superset. The bare\n *     variant must be suppressed or callers see two finals per utterance.\n *   - When `include_timestamps=false`, the server emits ONLY\n *     `committed_transcript`. Suppressing it loses every final.\n *\n * So suppression must be conditional on `includeTimestamps`, which is what\n * `wordTimestamps` resolves to on the URL.\n */\nexport const wireToEvent =\n  (includeTimestamps: boolean) =>\n  (msg: typeof ServerMessage.Type): TranscriptEvent | undefined =>\n    Match.value(msg).pipe(\n      Match.when({ message_type: \"session_started\" }, () => undefined),\n      Match.when(\n        { message_type: \"partial_transcript\" },\n        (m): TranscriptEvent => ({ _tag: \"partial\", text: m.text }),\n      ),\n      Match.when({ message_type: \"committed_transcript\" }, (m): TranscriptEvent | undefined =>\n        includeTimestamps\n          ? undefined // superseded by the `_with_timestamps` variant\n          : { _tag: \"final\", text: m.text },\n      ),\n      Match.when({ message_type: \"committed_transcript_with_timestamps\" }, (m): TranscriptEvent => {\n        const words = m.words == null ? undefined : m.words.map(wireToWordTimestamp)\n        return {\n          _tag: \"final\",\n          text: m.text,\n          ...(m.language_code != null && { languageCode: m.language_code }),\n          ...(words !== undefined && words.length > 0 && { words }),\n        }\n      }),\n      Match.orElse(\n        (m): TranscriptEvent => ({\n          _tag: \"error\",\n          code: m.message_type,\n          message:\n            typeof m.error === \"string\"\n              ? m.error\n              : JSON.stringify(m.error ?? `ElevenLabs ${m.message_type}`),\n        }),\n      ),\n    )\n\n// ---------------------------------------------------------------------------\n// URL + frame builders\n// ---------------------------------------------------------------------------\n\nexport type RealtimeOptions = {\n  readonly model?: string\n  readonly languageCode?: string\n  readonly includeTimestamps?: boolean\n  readonly commitStrategy?: \"manual\" | \"vad\"\n  readonly keyterms?: ReadonlyArray<string>\n}\n\nexport const buildWsUrl = (\n  cfg: Config,\n  token: string,\n  audioFormat: string,\n  opts: RealtimeOptions,\n) => {\n  const wsBase = resolveHost(cfg).replace(/^http/, \"ws\")\n  const params = new URLSearchParams({\n    token,\n    audio_format: audioFormat,\n    commit_strategy: opts.commitStrategy ?? \"vad\",\n    ...(opts.model !== undefined && { model_id: opts.model }),\n    ...(opts.languageCode !== undefined && { language_code: opts.languageCode }),\n    ...(opts.includeTimestamps === true && { include_timestamps: \"true\" }),\n  })\n  // Vocab biasing — repeated `keyterms` param (≤50 on realtime).\n  for (const term of opts.keyterms ?? []) params.append(\"keyterms\", term)\n  return `${wsBase}/speech-to-text/realtime?${params.toString()}`\n}\n\nexport const encodeAudioFrame = (bytes: Uint8Array, sampleRate: number) =>\n  JSON.stringify({\n    message_type: \"input_audio_chunk\",\n    audio_base_64: Encoding.encodeBase64(bytes),\n    sample_rate: sampleRate,\n  })\n\n// ---------------------------------------------------------------------------\n// Parse one server message and emit to the queue if it maps to a\n// TranscriptEvent. Invalid JSON / unknown shapes are silently dropped —\n// the connection stays open and we keep streaming subsequent messages.\n// ---------------------------------------------------------------------------\n\nconst handleServerMessage =\n  (\n    queue: Queue.Queue<TranscriptEvent, Cause.Done>,\n    mapEvent: (msg: typeof ServerMessage.Type) => TranscriptEvent | undefined,\n  ) =>\n  (raw: string) =>\n    Effect.gen(function* () {\n      const json = yield* JSONL.parseSafe(raw)\n      if (json === undefined) return\n      const decoded = yield* decodeServerMessage(json).pipe(Effect.option)\n      if (decoded._tag === \"None\") return\n      const event = mapEvent(decoded.value)\n      if (event !== undefined) yield* Queue.offer(queue, event)\n    })\n\n// ---------------------------------------------------------------------------\n// Stream<Uint8Array> → Stream<TranscriptEvent>\n//\n// Requires `HttpClient` (single-use token fetch) and\n// `Socket.WebSocketConstructor` (any `globalThis.WebSocket` works) in\n// context — `Socket.layerWebSocketConstructorGlobal` covers every\n// modern runtime.\n// ---------------------------------------------------------------------------\n\nexport const streamTranscription =\n  (cfg: Config) =>\n  <E, R>(\n    audioIn: Stream.Stream<Uint8Array, E, R>,\n    request: CommonStreamTranscribeRequest,\n  ): Stream.Stream<\n    TranscriptEvent,\n    AiError.AiError | E,\n    R | HttpClient.HttpClient | Socket.WebSocketConstructor\n  > =>\n    Stream.unwrap(\n      Effect.gen(function* () {\n        const slug = yield* inputFormatToSlug(request.inputFormat)\n        const token = yield* fetchSingleUseToken(cfg)\n        const includeTimestamps = request.wordTimestamps === true\n        yield* Capabilities.warnDroppedWhen(request.prompt, {\n          provider: \"elevenlabs\",\n          capability: \"prompt\",\n          field: \"prompt\",\n          reason: \"ElevenLabs Scribe has no free-form prompt field; bias via `biasingTerms`.\",\n        })\n        const url = buildWsUrl(cfg, token, slug, {\n          ...(request.model !== undefined && { model: request.model }),\n          ...(request.language !== undefined && { languageCode: request.language }),\n          ...(request.biasingTerms !== undefined && { keyterms: request.biasingTerms }),\n          includeTimestamps,\n        })\n        const socket = yield* Socket.makeWebSocket(url, {\n          // Effect's Socket treats all close codes as errors by default — that\n          // surfaces a clean server-side close as a stream failure. Whitelist\n          // standard clean-close codes (1000 / 1001 / 1005).\n          closeCodeIsError: (code) => code !== 1000 && code !== 1001 && code !== 1005,\n        })\n        const queue = yield* Queue.bounded<TranscriptEvent, Cause.Done>(64)\n        const sampleRate = request.inputFormat.sampleRate\n\n        const write = yield* socket.writer\n        yield* Stream.runForEach(audioIn, (bytes) =>\n          write(encodeAudioFrame(bytes, sampleRate)),\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, wireToEvent(includeTimestamps)))\n          .pipe(Effect.ensuring(Queue.end(queue)), Effect.forkScoped)\n\n        return Stream.fromQueue(queue)\n      }),\n    )\n"],"mappings":";;;;;;;;;AAsBA,MAAM,0BAA0B,WAC9B,IAAI,QAAQ,YAAY;CACtB,UAAU;CACV,YAAY;CACZ,QAAQ,qEAAqE,KAAK,UAAU,OAAO,CAAC;CACrG,CAAC;AAEJ,MAAa,oBACX,MAAM,MAAmB,CAAC,KACxB,MAAM,KAAK;CAAE,WAAW;CAAO,UAAU;CAAa,GAAG,MACvD,OAAO,QAAQ,OAAO,EAAE,aAAa,CACtC,EACD,MAAM,KAAK;CAAE,WAAW;CAAO,UAAU;CAAa,YAAY;CAAM,QACtE,OAAO,QAAQ,YAAY,CAC5B,EACD,MAAM,QAAQ,MAAM,OAAO,KAAK,uBAAuB,EAAE,CAAC,CAAC,CAC5D;AAOH,MAAM,YAAY,OAAO,OAAO,EAAE,OAAO,OAAO,QAAQ,CAAC;AACzD,MAAM,cAAc,OAAO,oBAAoB,UAAU;AAEzD,MAAa,uBAAuB,QAClC,OAAO,IAAI,aAAa;CACtB,MAAM,SAAS,OAAO,WAAW;CACjC,MAAM,MAAM,GAAG,YAAY,IAAI,CAAC;CAChC,MAAM,cAAc,kBAAkB,KAAK,IAAI,CAAC,KAC9C,kBAAkB,UAAU,cAAc,SAAS,MAAM,IAAI,OAAO,CAAC,CACtE;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;AAUzE,SAAO,OATa,YAAY,KAAK,CAAC,KACpC,OAAO,UACJ,UACC,IAAI,QAAQ,iBAAiB;EAC3B,UAAU;EACV,KAAK;GAAE,SAAS;GAAmD;GAAO;GAAM;EACjF,CAAC,CACL,CACF,EACW;EACZ;AAMJ,MAAM,eAAe,OAAO,OAAO;CACjC,MAAM,OAAO;CACb,OAAO,OAAO;CACd,KAAK,OAAO;CACZ,MAAM,OAAO,SAAS,OAAO,OAAO;CACpC,YAAY,OAAO,SAAS,OAAO,OAAO,OAAO,OAAO,CAAC;CACzD,SAAS,OAAO,SAAS,OAAO,OAAO;CACxC,CAAC;AAEF,MAAM,gBAAgB,OAAO,MAAM;CACjC,OAAO,OAAO;EACZ,cAAc,OAAO,QAAQ,kBAAkB;EAC/C,YAAY,OAAO,SAAS,OAAO,OAAO;EAC3C,CAAC;CACF,OAAO,OAAO;EACZ,cAAc,OAAO,QAAQ,qBAAqB;EAClD,MAAM,OAAO;EACd,CAAC;CACF,OAAO,OAAO;EACZ,cAAc,OAAO,QAAQ,uBAAuB;EACpD,MAAM,OAAO;EACd,CAAC;CACF,OAAO,OAAO;EACZ,cAAc,OAAO,QAAQ,uCAAuC;EACpE,MAAM,OAAO;EACb,eAAe,OAAO,SAAS,OAAO,OAAO,OAAO,OAAO,CAAC;EAC5D,OAAO,OAAO,SAAS,OAAO,OAAO,OAAO,MAAM,aAAa,CAAC,CAAC;EAClE,CAAC;CACF,OAAO,OAAO;EACZ,cAAc,OAAO,SAAS;GAAC;GAAS;GAAc;GAAgB;GAAiB,CAAC;EACxF,OAAO,OAAO,SAAS,OAAO,QAAQ;EACvC,CAAC;CACH,CAAC;AACF,MAAM,sBAAsB,OAAO,oBAAoB,cAAc;AAErE,MAAM,uBAAuB,OAAgD;CAC3E,MAAM,EAAE;CACR,cAAc,EAAE;CAChB,YAAY,EAAE;CACd,GAAI,EAAE,cAAc,QAAQ,EAAE,WAAW,EAAE,YAAY;CACvD,GAAI,EAAE,YAAY,KAAA,KAAa,EAAE,YAAY,KAAK,IAAI,EAAE,QAAQ,EAAE;CACnE;;;;;;;;;;;;;;;AAgBD,MAAa,eACV,uBACA,QACC,MAAM,MAAM,IAAI,CAAC,KACf,MAAM,KAAK,EAAE,cAAc,mBAAmB,QAAQ,KAAA,EAAU,EAChE,MAAM,KACJ,EAAE,cAAc,sBAAsB,GACrC,OAAwB;CAAE,MAAM;CAAW,MAAM,EAAE;CAAM,EAC3D,EACD,MAAM,KAAK,EAAE,cAAc,wBAAwB,GAAG,MACpD,oBACI,KAAA,IACA;CAAE,MAAM;CAAS,MAAM,EAAE;CAAM,CACpC,EACD,MAAM,KAAK,EAAE,cAAc,wCAAwC,GAAG,MAAuB;CAC3F,MAAM,QAAQ,EAAE,SAAS,OAAO,KAAA,IAAY,EAAE,MAAM,IAAI,oBAAoB;AAC5E,QAAO;EACL,MAAM;EACN,MAAM,EAAE;EACR,GAAI,EAAE,iBAAiB,QAAQ,EAAE,cAAc,EAAE,eAAe;EAChE,GAAI,UAAU,KAAA,KAAa,MAAM,SAAS,KAAK,EAAE,OAAO;EACzD;EACD,EACF,MAAM,QACH,OAAwB;CACvB,MAAM;CACN,MAAM,EAAE;CACR,SACE,OAAO,EAAE,UAAU,WACf,EAAE,QACF,KAAK,UAAU,EAAE,SAAS,cAAc,EAAE,eAAe;CAChE,EACF,CACF;AAcL,MAAa,cACX,KACA,OACA,aACA,SACG;CACH,MAAM,SAAS,YAAY,IAAI,CAAC,QAAQ,SAAS,KAAK;CACtD,MAAM,SAAS,IAAI,gBAAgB;EACjC;EACA,cAAc;EACd,iBAAiB,KAAK,kBAAkB;EACxC,GAAI,KAAK,UAAU,KAAA,KAAa,EAAE,UAAU,KAAK,OAAO;EACxD,GAAI,KAAK,iBAAiB,KAAA,KAAa,EAAE,eAAe,KAAK,cAAc;EAC3E,GAAI,KAAK,sBAAsB,QAAQ,EAAE,oBAAoB,QAAQ;EACtE,CAAC;AAEF,MAAK,MAAM,QAAQ,KAAK,YAAY,EAAE,CAAE,QAAO,OAAO,YAAY,KAAK;AACvE,QAAO,GAAG,OAAO,2BAA2B,OAAO,UAAU;;AAG/D,MAAa,oBAAoB,OAAmB,eAClD,KAAK,UAAU;CACb,cAAc;CACd,eAAe,SAAS,aAAa,MAAM;CAC3C,aAAa;CACd,CAAC;AAQJ,MAAM,uBAEF,OACA,cAED,QACC,OAAO,IAAI,aAAa;CACtB,MAAM,OAAO,OAAO,MAAM,UAAU,IAAI;AACxC,KAAI,SAAS,KAAA,EAAW;CACxB,MAAM,UAAU,OAAO,oBAAoB,KAAK,CAAC,KAAK,OAAO,OAAO;AACpE,KAAI,QAAQ,SAAS,OAAQ;CAC7B,MAAM,QAAQ,SAAS,QAAQ,MAAM;AACrC,KAAI,UAAU,KAAA,EAAW,QAAO,MAAM,MAAM,OAAO,MAAM;EACzD;AAWN,MAAa,uBACV,SAEC,SACA,YAMA,OAAO,OACL,OAAO,IAAI,aAAa;CACtB,MAAM,OAAO,OAAO,kBAAkB,QAAQ,YAAY;CAC1D,MAAM,QAAQ,OAAO,oBAAoB,IAAI;CAC7C,MAAM,oBAAoB,QAAQ,mBAAmB;AACrD,QAAO,aAAa,gBAAgB,QAAQ,QAAQ;EAClD,UAAU;EACV,YAAY;EACZ,OAAO;EACP,QAAQ;EACT,CAAC;CACF,MAAM,MAAM,WAAW,KAAK,OAAO,MAAM;EACvC,GAAI,QAAQ,UAAU,KAAA,KAAa,EAAE,OAAO,QAAQ,OAAO;EAC3D,GAAI,QAAQ,aAAa,KAAA,KAAa,EAAE,cAAc,QAAQ,UAAU;EACxE,GAAI,QAAQ,iBAAiB,KAAA,KAAa,EAAE,UAAU,QAAQ,cAAc;EAC5E;EACD,CAAC;CACF,MAAM,SAAS,OAAO,OAAO,cAAc,KAAK,EAI9C,mBAAmB,SAAS,SAAS,OAAQ,SAAS,QAAQ,SAAS,MACxE,CAAC;CACF,MAAM,QAAQ,OAAO,MAAM,QAAqC,GAAG;CACnE,MAAM,aAAa,QAAQ,YAAY;CAEvC,MAAM,QAAQ,OAAO,OAAO;AAC5B,QAAO,OAAO,WAAW,UAAU,UACjC,MAAM,iBAAiB,OAAO,WAAW,CAAC,CAC3C,CAAC,KAAK,OAAO,QAAQ,OAAO,WAAW;AAKxC,QAAO,OACJ,UAAU,oBAAoB,OAAO,YAAY,kBAAkB,CAAC,CAAC,CACrE,KAAK,OAAO,SAAS,MAAM,IAAI,MAAM,CAAC,EAAE,OAAO,WAAW;AAE7D,QAAO,OAAO,UAAU,MAAM;EAC9B,CACH"}

package/dist/realtimeTts.d.mts

@@ -1,9 +1,10 @@
-import { VoiceSettings } from "./codec.mjs";
 import { ElevenLabsTtsModel, ElevenLabsVoiceId } from "./models.mjs";
 import { ElevenLabsRegion } from "./region.mjs";
+import { PronunciationDictionaryLocator, VoiceSettings } from "./codec.mjs";
 import { Redacted, Stream } from "effect";
-import * as Socket from "effect/unstable/socket/Socket";
 import * as AiError from "@effect-uai/core/AiError";
+import * as Socket from "effect/unstable/socket/Socket";
+import { CustomPronunciation } from "@effect-uai/core/SpeechSynthesizer";
 import { AudioChunk, AudioFormat } from "@effect-uai/core/Audio";
 
 //#region src/realtimeTts.d.ts
@@ -24,7 +25,11 @@ type StreamSynthesizeRequest = {
   readonly outputFormat?: AudioFormat;
   readonly languageCode?: string;
   readonly voiceSettings?: VoiceSettings;
-  readonly autoMode?: boolean;
+  readonly autoMode?: boolean; /** Pre-provisioned pronunciation dictionaries, sent on the BOS frame. */
+  readonly pronunciationDictionaryLocators?: ReadonlyArray<PronunciationDictionaryLocator>;
+  /** Carried from the Common request so the WS path can reject inline
+   *  pronunciations (ElevenLabs has no stateless inline IPA path). */
+  readonly pronunciations?: ReadonlyArray<CustomPronunciation>;
 };
 declare const buildWsUrl: (cfg: Config, request: StreamSynthesizeRequest, outputFormat: string) => string;
 declare const streamSynthesis: (cfg: Config) => <E, R>(textIn: Stream.Stream<string, E, R>, request: StreamSynthesizeRequest) => Stream.Stream<AudioChunk, AiError.AiError | E, R | Socket.WebSocketConstructor>;

package/dist/realtimeTts.d.mts.map

@@ -1 +1 @@
-{"version":3,"file":"realtimeTts.d.mts","names":[],"sources":["../src/realtimeTts.ts"],"mappings":";;;;;;;;;KAcY,MAAA;EAAA,SACD,MAAA,EAAQ,QAAA,CAAS,QAAA;EAAA,SACjB,OAAA;EAAA,SACA,MAAA,GAAS,gBAAA;AAAA;;;;;;;KAWR,uBAAA;EAAA,SACD,KAAA,GAAQ,kBAAA;EAAA,SACR,OAAA,EAAS,iBAAA;EAAA,SACT,YAAA,GAAe,WAAA;EAAA,SACf,YAAA;EAAA,SACA,aAAA,GAAgB,aAAA;EAAA,SAChB,QAAA;AAAA;AAAA,cAOE,UAAA,GAAc,GAAA,EAAK,MAAA,EAAQ,OAAA,EAAS,uBAAA,EAAyB,YAAA;AAAA,cAsE7D,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,GAAI,MAAA,CAAO,oBAAA"}
+{"version":3,"file":"realtimeTts.d.mts","names":[],"sources":["../src/realtimeTts.ts"],"mappings":";;;;;;;;;;KAkBY,MAAA;EAAA,SACD,MAAA,EAAQ,QAAA,CAAS,QAAA;EAAA,SACjB,OAAA;EAAA,SACA,MAAA,GAAS,gBAAA;AAAA;;;;;;;KAWR,uBAAA;EAAA,SACD,KAAA,GAAQ,kBAAA;EAAA,SACR,OAAA,EAAS,iBAAA;EAAA,SACT,YAAA,GAAe,WAAA;EAAA,SACf,YAAA;EAAA,SACA,aAAA,GAAgB,aAAA;EAAA,SAChB,QAAA,YAHe;EAAA,SAKf,+BAAA,GAAkC,aAAA,CAAc,8BAAA;EAAA;;EAAA,SAGhD,cAAA,GAAiB,aAAA,CAAc,mBAAA;AAAA;AAAA,cAO7B,UAAA,GAAc,GAAA,EAAK,MAAA,EAAQ,OAAA,EAAS,uBAAA,EAAyB,YAAA;AAAA,cAuE7D,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,GAAI,MAAA,CAAO,oBAAA"}

package/dist/realtimeTts.mjs

@@ -1,8 +1,8 @@
-import { defaultFormat, formatToOutputSlug, wireVoiceSettings } from "./codec.mjs";
+import { defaultFormat, formatToOutputSlug, rejectInlinePronunciations, wirePronunciationLocators, wireVoiceSettings } from "./codec.mjs";
 import { resolveHost } from "./region.mjs";
 import { Effect, Encoding, Queue, Redacted, Result, Schema, Stream } from "effect";
-import * as Socket from "effect/unstable/socket/Socket";
 import * as AiError from "@effect-uai/core/AiError";
+import * as Socket from "effect/unstable/socket/Socket";
 import * as JSONL from "@effect-uai/core/JSONL";
 //#region src/realtimeTts.ts
 const buildWsUrl = (cfg, request, outputFormat) => {
@@ -20,7 +20,8 @@ const bosFrame = (cfg, request) => {
 	return JSON.stringify({
 		text: " ",
 		"xi-api-key": Redacted.value(cfg.apiKey),
-		...vs !== void 0 && { voice_settings: vs }
+		...vs !== void 0 && { voice_settings: vs },
+		...wirePronunciationLocators(request.pronunciationDictionaryLocators)
 	});
 };
 const textFrame = (text) => JSON.stringify({ text: text.endsWith(" ") ? text : `${text} ` });
@@ -60,6 +61,7 @@ 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* () {
+	yield* rejectInlinePronunciations(request.pronunciations);
 	const slug = yield* formatToOutputSlug(request.outputFormat ?? defaultFormat);
 	const socket = yield* Socket.makeWebSocket(buildWsUrl(cfg, request, slug), { closeCodeIsError: (code) => code !== 1e3 && code !== 1001 && code !== 1005 });
 	const queue = yield* Queue.bounded(64);

package/dist/realtimeTts.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"realtimeTts.mjs","names":[],"sources":["../src/realtimeTts.ts"],"sourcesContent":["import { Cause, Effect, Encoding, Queue, Redacted, Result, 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 {\n  defaultFormat,\n  formatToOutputSlug,\n  type VoiceSettings,\n  wireVoiceSettings,\n} from \"./codec.js\"\nimport type { ElevenLabsTtsModel, ElevenLabsVoiceId } from \"./models.js\"\nimport { type ElevenLabsRegion, resolveHost } from \"./region.js\"\n\nexport type Config = {\n  readonly apiKey: Redacted.Redacted\n  readonly baseUrl?: string\n  readonly region?: ElevenLabsRegion\n}\n\nexport type { VoiceSettings } from \"./codec.js\"\n\n/**\n * Incremental-text-in request for `/stream-input`. `voiceSettings`\n * applies only on the BOS frame — mid-stream voice changes are\n * rejected. `autoMode: true` (default) lets the model pick flush\n * boundaries, which is what you want for LLM-token streams.\n */\nexport type StreamSynthesizeRequest = {\n  readonly model?: ElevenLabsTtsModel\n  readonly voiceId: ElevenLabsVoiceId\n  readonly outputFormat?: AudioFormat\n  readonly languageCode?: string\n  readonly voiceSettings?: VoiceSettings\n  readonly autoMode?: boolean\n}\n\n// ---------------------------------------------------------------------------\n// URL + frame builders\n// ---------------------------------------------------------------------------\n\nexport const buildWsUrl = (cfg: Config, request: StreamSynthesizeRequest, outputFormat: string) => {\n  const wsBase = resolveHost(cfg).replace(/^http/, \"ws\")\n  const params = new URLSearchParams({\n    output_format: outputFormat,\n    auto_mode: String(request.autoMode ?? true),\n    ...(request.model !== undefined && { model_id: request.model }),\n    ...(request.languageCode !== undefined && { language_code: request.languageCode }),\n  })\n  return `${wsBase}/text-to-speech/${request.voiceId}/stream-input?${params.toString()}`\n}\n\nconst bosFrame = (cfg: Config, request: StreamSynthesizeRequest) => {\n  const vs = wireVoiceSettings(request.voiceSettings)\n  return JSON.stringify({\n    text: \" \",\n    \"xi-api-key\": Redacted.value(cfg.apiKey),\n    ...(vs !== undefined && { voice_settings: vs }),\n  })\n}\n\nconst textFrame = (text: string) => JSON.stringify({ text: text.endsWith(\" \") ? text : `${text} ` })\nconst eosFrame = JSON.stringify({ text: \"\" })\n\n// ---------------------------------------------------------------------------\n// Wire schema (server → client) + helpers\n// ---------------------------------------------------------------------------\n\nconst ServerFrame = Schema.Struct({\n  audio: Schema.optional(Schema.NullOr(Schema.String)),\n  isFinal: Schema.optional(Schema.NullOr(Schema.Boolean)),\n  error: Schema.optional(Schema.Unknown),\n  message: Schema.optional(Schema.String),\n})\nconst decodeServerFrame = Schema.decodeUnknownEffect(ServerFrame)\n\nconst decodeAudio = (b64: string): Effect.Effect<Uint8Array, AiError.AiError> =>\n  Result.match(Encoding.decodeBase64(b64), {\n    onSuccess: Effect.succeed,\n    onFailure: (cause) =>\n      Effect.fail(\n        new AiError.GenerationFailed({\n          provider: \"elevenlabs\",\n          raw: { message: \"failed to decode audio frame\", cause },\n        }),\n      ),\n  })\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(\"[elevenlabs-tts] server error frame\", {\n        error: frame.error,\n        message: frame.message,\n      })\n      return\n    }\n    if (frame.audio == null || frame.audio === \"\") return\n    const bytes = yield* decodeAudio(frame.audio).pipe(Effect.option)\n    if (bytes._tag === \"Some\") yield* Queue.offer(queue, { bytes: bytes.value })\n  })\n\n// ---------------------------------------------------------------------------\n// Stream<string> → Stream<AudioChunk>. Requires `Socket.WebSocketConstructor`.\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 | Socket.WebSocketConstructor> =>\n    Stream.unwrap(\n      Effect.gen(function* () {\n        const slug = yield* formatToOutputSlug(request.outputFormat ?? defaultFormat)\n        // ElevenLabs closes `/stream-input` with code 1000 after delivering the\n        // final audio chunk. Effect's default treats all close codes as errors,\n        // which would surface as a stream failure right after the last audio\n        // arrives. Whitelist standard clean-close codes.\n        const socket = yield* Socket.makeWebSocket(buildWsUrl(cfg, request, slug), {\n          closeCodeIsError: (code) => code !== 1000 && code !== 1001 && code !== 1005,\n        })\n        const queue = yield* Queue.bounded<AudioChunk, Cause.Done>(64)\n        const write = yield* socket.writer\n\n        // Writer fiber: BOS → drain text stream → EOS. Socket-side failures\n        // end this fiber; the reader still surfaces a clean stream end via\n        // `Queue.end` (in the ensuring below) when the upstream WS closes.\n        yield* Effect.gen(function* () {\n          yield* write(bosFrame(cfg, request))\n          yield* Stream.runForEach(textIn, (text) =>\n            text.length === 0 ? Effect.void : write(textFrame(text)),\n          )\n          yield* write(eosFrame)\n        }).pipe(Effect.ignore, Effect.forkScoped)\n\n        // Reader fiber. `ensuring(Queue.end)` flushes pending chunks then\n        // fails the next take with `Done`, which `Stream.fromQueue` treats\n        // as a clean end. (`Queue.shutdown` would CLEAR queued items and\n        // 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":";;;;;;;AAyCA,MAAa,cAAc,KAAa,SAAkC,iBAAyB;CACjG,MAAM,SAAS,YAAY,IAAI,CAAC,QAAQ,SAAS,KAAK;CACtD,MAAM,SAAS,IAAI,gBAAgB;EACjC,eAAe;EACf,WAAW,OAAO,QAAQ,YAAY,KAAK;EAC3C,GAAI,QAAQ,UAAU,KAAA,KAAa,EAAE,UAAU,QAAQ,OAAO;EAC9D,GAAI,QAAQ,iBAAiB,KAAA,KAAa,EAAE,eAAe,QAAQ,cAAc;EAClF,CAAC;AACF,QAAO,GAAG,OAAO,kBAAkB,QAAQ,QAAQ,gBAAgB,OAAO,UAAU;;AAGtF,MAAM,YAAY,KAAa,YAAqC;CAClE,MAAM,KAAK,kBAAkB,QAAQ,cAAc;AACnD,QAAO,KAAK,UAAU;EACpB,MAAM;EACN,cAAc,SAAS,MAAM,IAAI,OAAO;EACxC,GAAI,OAAO,KAAA,KAAa,EAAE,gBAAgB,IAAI;EAC/C,CAAC;;AAGJ,MAAM,aAAa,SAAiB,KAAK,UAAU,EAAE,MAAM,KAAK,SAAS,IAAI,GAAG,OAAO,GAAG,KAAK,IAAI,CAAC;AACpG,MAAM,WAAW,KAAK,UAAU,EAAE,MAAM,IAAI,CAAC;AAM7C,MAAM,cAAc,OAAO,OAAO;CAChC,OAAO,OAAO,SAAS,OAAO,OAAO,OAAO,OAAO,CAAC;CACpD,SAAS,OAAO,SAAS,OAAO,OAAO,OAAO,QAAQ,CAAC;CACvD,OAAO,OAAO,SAAS,OAAO,QAAQ;CACtC,SAAS,OAAO,SAAS,OAAO,OAAO;CACxC,CAAC;AACF,MAAM,oBAAoB,OAAO,oBAAoB,YAAY;AAEjE,MAAM,eAAe,QACnB,OAAO,MAAM,SAAS,aAAa,IAAI,EAAE;CACvC,WAAW,OAAO;CAClB,YAAY,UACV,OAAO,KACL,IAAI,QAAQ,iBAAiB;EAC3B,UAAU;EACV,KAAK;GAAE,SAAS;GAAgC;GAAO;EACxD,CAAC,CACH;CACJ,CAAC;AAEJ,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,uCAAuC;GAC9D,OAAO,MAAM;GACb,SAAS,MAAM;GAChB,CAAC;AACF;;AAEF,KAAI,MAAM,SAAS,QAAQ,MAAM,UAAU,GAAI;CAC/C,MAAM,QAAQ,OAAO,YAAY,MAAM,MAAM,CAAC,KAAK,OAAO,OAAO;AACjE,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,OAAO,OAAO,mBAAmB,QAAQ,gBAAgB,cAAc;CAK7E,MAAM,SAAS,OAAO,OAAO,cAAc,WAAW,KAAK,SAAS,KAAK,EAAE,EACzE,mBAAmB,SAAS,SAAS,OAAQ,SAAS,QAAQ,SAAS,MACxE,CAAC;CACF,MAAM,QAAQ,OAAO,MAAM,QAAgC,GAAG;CAC9D,MAAM,QAAQ,OAAO,OAAO;AAK5B,QAAO,OAAO,IAAI,aAAa;AAC7B,SAAO,MAAM,SAAS,KAAK,QAAQ,CAAC;AACpC,SAAO,OAAO,WAAW,SAAS,SAChC,KAAK,WAAW,IAAI,OAAO,OAAO,MAAM,UAAU,KAAK,CAAC,CACzD;AACD,SAAO,MAAM,SAAS;GACtB,CAAC,KAAK,OAAO,QAAQ,OAAO,WAAW;AAMzC,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":["import { Cause, Effect, Encoding, Queue, Redacted, Result, 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 {\n  defaultFormat,\n  formatToOutputSlug,\n  type PronunciationDictionaryLocator,\n  rejectInlinePronunciations,\n  type VoiceSettings,\n  wirePronunciationLocators,\n  wireVoiceSettings,\n} from \"./codec.js\"\nimport type { ElevenLabsTtsModel, ElevenLabsVoiceId } from \"./models.js\"\nimport { type ElevenLabsRegion, resolveHost } from \"./region.js\"\n\nexport type Config = {\n  readonly apiKey: Redacted.Redacted\n  readonly baseUrl?: string\n  readonly region?: ElevenLabsRegion\n}\n\nexport type { VoiceSettings } from \"./codec.js\"\n\n/**\n * Incremental-text-in request for `/stream-input`. `voiceSettings`\n * applies only on the BOS frame — mid-stream voice changes are\n * rejected. `autoMode: true` (default) lets the model pick flush\n * boundaries, which is what you want for LLM-token streams.\n */\nexport type StreamSynthesizeRequest = {\n  readonly model?: ElevenLabsTtsModel\n  readonly voiceId: ElevenLabsVoiceId\n  readonly outputFormat?: AudioFormat\n  readonly languageCode?: string\n  readonly voiceSettings?: VoiceSettings\n  readonly autoMode?: boolean\n  /** Pre-provisioned pronunciation dictionaries, sent on the BOS frame. */\n  readonly pronunciationDictionaryLocators?: ReadonlyArray<PronunciationDictionaryLocator>\n  /** Carried from the Common request so the WS path can reject inline\n   *  pronunciations (ElevenLabs has no stateless inline IPA path). */\n  readonly pronunciations?: ReadonlyArray<CustomPronunciation>\n}\n\n// ---------------------------------------------------------------------------\n// URL + frame builders\n// ---------------------------------------------------------------------------\n\nexport const buildWsUrl = (cfg: Config, request: StreamSynthesizeRequest, outputFormat: string) => {\n  const wsBase = resolveHost(cfg).replace(/^http/, \"ws\")\n  const params = new URLSearchParams({\n    output_format: outputFormat,\n    auto_mode: String(request.autoMode ?? true),\n    ...(request.model !== undefined && { model_id: request.model }),\n    ...(request.languageCode !== undefined && { language_code: request.languageCode }),\n  })\n  return `${wsBase}/text-to-speech/${request.voiceId}/stream-input?${params.toString()}`\n}\n\nconst bosFrame = (cfg: Config, request: StreamSynthesizeRequest) => {\n  const vs = wireVoiceSettings(request.voiceSettings)\n  return JSON.stringify({\n    text: \" \",\n    \"xi-api-key\": Redacted.value(cfg.apiKey),\n    ...(vs !== undefined && { voice_settings: vs }),\n    ...wirePronunciationLocators(request.pronunciationDictionaryLocators),\n  })\n}\n\nconst textFrame = (text: string) => JSON.stringify({ text: text.endsWith(\" \") ? text : `${text} ` })\nconst eosFrame = JSON.stringify({ text: \"\" })\n\n// ---------------------------------------------------------------------------\n// Wire schema (server → client) + helpers\n// ---------------------------------------------------------------------------\n\nconst ServerFrame = Schema.Struct({\n  audio: Schema.optional(Schema.NullOr(Schema.String)),\n  isFinal: Schema.optional(Schema.NullOr(Schema.Boolean)),\n  error: Schema.optional(Schema.Unknown),\n  message: Schema.optional(Schema.String),\n})\nconst decodeServerFrame = Schema.decodeUnknownEffect(ServerFrame)\n\nconst decodeAudio = (b64: string): Effect.Effect<Uint8Array, AiError.AiError> =>\n  Result.match(Encoding.decodeBase64(b64), {\n    onSuccess: Effect.succeed,\n    onFailure: (cause) =>\n      Effect.fail(\n        new AiError.GenerationFailed({\n          provider: \"elevenlabs\",\n          raw: { message: \"failed to decode audio frame\", cause },\n        }),\n      ),\n  })\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(\"[elevenlabs-tts] server error frame\", {\n        error: frame.error,\n        message: frame.message,\n      })\n      return\n    }\n    if (frame.audio == null || frame.audio === \"\") return\n    const bytes = yield* decodeAudio(frame.audio).pipe(Effect.option)\n    if (bytes._tag === \"Some\") yield* Queue.offer(queue, { bytes: bytes.value })\n  })\n\n// ---------------------------------------------------------------------------\n// Stream<string> → Stream<AudioChunk>. Requires `Socket.WebSocketConstructor`.\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 | Socket.WebSocketConstructor> =>\n    Stream.unwrap(\n      Effect.gen(function* () {\n        yield* rejectInlinePronunciations(request.pronunciations)\n        const slug = yield* formatToOutputSlug(request.outputFormat ?? defaultFormat)\n        // ElevenLabs closes `/stream-input` with code 1000 after delivering the\n        // final audio chunk. Effect's default treats all close codes as errors,\n        // which would surface as a stream failure right after the last audio\n        // arrives. Whitelist standard clean-close codes.\n        const socket = yield* Socket.makeWebSocket(buildWsUrl(cfg, request, slug), {\n          closeCodeIsError: (code) => code !== 1000 && code !== 1001 && code !== 1005,\n        })\n        const queue = yield* Queue.bounded<AudioChunk, Cause.Done>(64)\n        const write = yield* socket.writer\n\n        // Writer fiber: BOS → drain text stream → EOS. Socket-side failures\n        // end this fiber; the reader still surfaces a clean stream end via\n        // `Queue.end` (in the ensuring below) when the upstream WS closes.\n        yield* Effect.gen(function* () {\n          yield* write(bosFrame(cfg, request))\n          yield* Stream.runForEach(textIn, (text) =>\n            text.length === 0 ? Effect.void : write(textFrame(text)),\n          )\n          yield* write(eosFrame)\n        }).pipe(Effect.ignore, Effect.forkScoped)\n\n        // Reader fiber. `ensuring(Queue.end)` flushes pending chunks then\n        // fails the next take with `Done`, which `Stream.fromQueue` treats\n        // as a clean end. (`Queue.shutdown` would CLEAR queued items and\n        // 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":";;;;;;;AAkDA,MAAa,cAAc,KAAa,SAAkC,iBAAyB;CACjG,MAAM,SAAS,YAAY,IAAI,CAAC,QAAQ,SAAS,KAAK;CACtD,MAAM,SAAS,IAAI,gBAAgB;EACjC,eAAe;EACf,WAAW,OAAO,QAAQ,YAAY,KAAK;EAC3C,GAAI,QAAQ,UAAU,KAAA,KAAa,EAAE,UAAU,QAAQ,OAAO;EAC9D,GAAI,QAAQ,iBAAiB,KAAA,KAAa,EAAE,eAAe,QAAQ,cAAc;EAClF,CAAC;AACF,QAAO,GAAG,OAAO,kBAAkB,QAAQ,QAAQ,gBAAgB,OAAO,UAAU;;AAGtF,MAAM,YAAY,KAAa,YAAqC;CAClE,MAAM,KAAK,kBAAkB,QAAQ,cAAc;AACnD,QAAO,KAAK,UAAU;EACpB,MAAM;EACN,cAAc,SAAS,MAAM,IAAI,OAAO;EACxC,GAAI,OAAO,KAAA,KAAa,EAAE,gBAAgB,IAAI;EAC9C,GAAG,0BAA0B,QAAQ,gCAAgC;EACtE,CAAC;;AAGJ,MAAM,aAAa,SAAiB,KAAK,UAAU,EAAE,MAAM,KAAK,SAAS,IAAI,GAAG,OAAO,GAAG,KAAK,IAAI,CAAC;AACpG,MAAM,WAAW,KAAK,UAAU,EAAE,MAAM,IAAI,CAAC;AAM7C,MAAM,cAAc,OAAO,OAAO;CAChC,OAAO,OAAO,SAAS,OAAO,OAAO,OAAO,OAAO,CAAC;CACpD,SAAS,OAAO,SAAS,OAAO,OAAO,OAAO,QAAQ,CAAC;CACvD,OAAO,OAAO,SAAS,OAAO,QAAQ;CACtC,SAAS,OAAO,SAAS,OAAO,OAAO;CACxC,CAAC;AACF,MAAM,oBAAoB,OAAO,oBAAoB,YAAY;AAEjE,MAAM,eAAe,QACnB,OAAO,MAAM,SAAS,aAAa,IAAI,EAAE;CACvC,WAAW,OAAO;CAClB,YAAY,UACV,OAAO,KACL,IAAI,QAAQ,iBAAiB;EAC3B,UAAU;EACV,KAAK;GAAE,SAAS;GAAgC;GAAO;EACxD,CAAC,CACH;CACJ,CAAC;AAEJ,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,uCAAuC;GAC9D,OAAO,MAAM;GACb,SAAS,MAAM;GAChB,CAAC;AACF;;AAEF,KAAI,MAAM,SAAS,QAAQ,MAAM,UAAU,GAAI;CAC/C,MAAM,QAAQ,OAAO,YAAY,MAAM,MAAM,CAAC,KAAK,OAAO,OAAO;AACjE,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;AACtB,QAAO,2BAA2B,QAAQ,eAAe;CACzD,MAAM,OAAO,OAAO,mBAAmB,QAAQ,gBAAgB,cAAc;CAK7E,MAAM,SAAS,OAAO,OAAO,cAAc,WAAW,KAAK,SAAS,KAAK,EAAE,EACzE,mBAAmB,SAAS,SAAS,OAAQ,SAAS,QAAQ,SAAS,MACxE,CAAC;CACF,MAAM,QAAQ,OAAO,MAAM,QAAgC,GAAG;CAC9D,MAAM,QAAQ,OAAO,OAAO;AAK5B,QAAO,OAAO,IAAI,aAAa;AAC7B,SAAO,MAAM,SAAS,KAAK,QAAQ,CAAC;AACpC,SAAO,OAAO,WAAW,SAAS,SAChC,KAAK,WAAW,IAAI,OAAO,OAAO,MAAM,UAAU,KAAK,CAAC,CACzD;AACD,SAAO,MAAM,SAAS;GACtB,CAAC,KAAK,OAAO,QAAQ,OAAO,WAAW;AAMzC,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

@@ -1,11 +1,13 @@
 {
   "name": "@effect-uai/elevenlabs",
-  "version": "0.6.0",
-  "description": "ElevenLabs speech-to-text and text-to-speech provider for @effect-uai/core.",
+  "version": "0.8.0",
+  "description": "ElevenLabs speech-to-text, text-to-speech, and music-generation provider for @effect-uai/core.",
   "keywords": [
     "ai",
     "effect",
     "elevenlabs",
+    "music",
+    "music-generation",
     "speech",
     "stt",
     "transcription",
@@ -36,6 +38,10 @@
       "types": "./dist/index.d.mts",
       "import": "./dist/index.mjs"
     },
+    "./ElevenLabsMusicGenerator": {
+      "types": "./dist/ElevenLabsMusicGenerator.d.mts",
+      "import": "./dist/ElevenLabsMusicGenerator.mjs"
+    },
     "./ElevenLabsSynthesizer": {
       "types": "./dist/ElevenLabsSynthesizer.d.mts",
       "import": "./dist/ElevenLabsSynthesizer.mjs"
@@ -51,7 +57,7 @@
   "devDependencies": {
     "effect": "4.0.0-beta.57",
     "typescript": "^6.0.3",
-    "@effect-uai/core": "0.6.0"
+    "@effect-uai/core": "0.8.0"
   },
   "peerDependencies": {
     "@effect-uai/core": ">=0.2.0 <1",

package/src/ElevenLabsMusicGenerator.test.ts

@@ -0,0 +1,116 @@
+import { Duration, Effect, Layer, Redacted, Stream } from "effect"
+import { FetchHttpClient } from "effect/unstable/http"
+import { describe, expect, expectTypeOf, it } from "vitest"
+import type * as AiError from "@effect-uai/core/AiError"
+import type { GenerateResult } from "@effect-uai/core/Music"
+import { promptsInput } from "@effect-uai/core/Music"
+import * as MusicGenerator from "@effect-uai/core/MusicGenerator"
+import * as ElevenLabsMusicGenerator from "./ElevenLabsMusicGenerator.js"
+
+const cfg: ElevenLabsMusicGenerator.Config = { apiKey: Redacted.make("test-key") }
+const live = Layer.provide(ElevenLabsMusicGenerator.layer(cfg), FetchHttpClient.layer)
+
+describe("ElevenLabsMusicGenerator capability guards (runtime)", () => {
+  it("streamGenerationFrom returns an Unsupported stream", async () => {
+    const program = ElevenLabsMusicGenerator.ElevenLabsMusicGenerator.use((s) =>
+      Stream.runDrain(
+        s.streamGenerationFrom(Stream.fromIterable([promptsInput([{ text: "x" }])]), {
+          model: "music_v1",
+          prompt: "",
+        }),
+      ),
+    )
+    const exit = await Effect.runPromiseExit(program.pipe(Effect.provide(live)))
+    expect(exit._tag).toBe("Failure")
+    if (exit._tag === "Failure") {
+      expect(JSON.stringify(exit.cause)).toContain("Unsupported")
+      expect(JSON.stringify(exit.cause)).toContain("streamGenerationFrom")
+    }
+  })
+})
+
+describe("ElevenLabsMusicGenerator Layer (compile-time)", () => {
+  it("leaves `MusicInteractiveSession` unsatisfied via the generic surface", () => {
+    const inputs = Stream.fromIterable([promptsInput([{ text: "techno" }])])
+    const events = inputs.pipe(
+      MusicGenerator.streamGenerationFrom({ model: "music_v1", prompt: "" }),
+    )
+    const provided = Stream.runDrain(events).pipe(Effect.provide(live))
+    expectTypeOf(provided).toEqualTypeOf<
+      Effect.Effect<void, AiError.AiError, MusicGenerator.MusicInteractiveSession>
+    >()
+  })
+
+  it("sync `generate` returns GenerateResult and requires no marker", () => {
+    const gen = MusicGenerator.generate({ model: "music_v1", prompt: "house" }).pipe(
+      Effect.provide(live),
+    )
+    expectTypeOf(gen).toEqualTypeOf<Effect.Effect<GenerateResult, AiError.AiError, never>>()
+  })
+
+  it("typed request accepts compositionPlan, forceInstrumental, signWithC2pa", () => {
+    const _req: ElevenLabsMusicGenerator.ElevenLabsMusicGenerateRequest = {
+      model: "music_v1",
+      prompt: "",
+      compositionPlan: {
+        positiveGlobalStyles: ["lo-fi", "warm"],
+        negativeGlobalStyles: ["distorted"],
+        sections: [
+          {
+            sectionName: "Intro",
+            positiveLocalStyles: ["piano"],
+            negativeLocalStyles: [],
+            duration: Duration.seconds(15),
+            lines: [],
+          },
+        ],
+      },
+      forceInstrumental: true,
+      signWithC2pa: true,
+    }
+    expect(_req).toBeDefined()
+  })
+})
+
+describe("ElevenLabsMusicGenerator codec validation (runtime, no HTTP)", () => {
+  it("rejects compositionPlan + non-empty prompt", async () => {
+    const program = ElevenLabsMusicGenerator.ElevenLabsMusicGenerator.use((s) =>
+      s.generate({
+        model: "music_v1",
+        prompt: "house",
+        compositionPlan: {
+          positiveGlobalStyles: [],
+          negativeGlobalStyles: [],
+          sections: [],
+        },
+      }),
+    )
+    const exit = await Effect.runPromiseExit(program.pipe(Effect.provide(live)))
+    expect(exit._tag).toBe("Failure")
+    if (exit._tag === "Failure") {
+      expect(JSON.stringify(exit.cause)).toContain("InvalidRequest")
+      expect(JSON.stringify(exit.cause)).toContain("mutually exclusive")
+    }
+  })
+
+  it("rejects compositionPlan + duration", async () => {
+    const program = ElevenLabsMusicGenerator.ElevenLabsMusicGenerator.use((s) =>
+      s.generate({
+        model: "music_v1",
+        prompt: "",
+        duration: Duration.seconds(30),
+        compositionPlan: {
+          positiveGlobalStyles: [],
+          negativeGlobalStyles: [],
+          sections: [],
+        },
+      }),
+    )
+    const exit = await Effect.runPromiseExit(program.pipe(Effect.provide(live)))
+    expect(exit._tag).toBe("Failure")
+    if (exit._tag === "Failure") {
+      expect(JSON.stringify(exit.cause)).toContain("InvalidRequest")
+      expect(JSON.stringify(exit.cause)).toContain("`duration` is ignored")
+    }
+  })
+})