@ai-ecoverse/pi-audio 1.4.0 → 1.5.0

package/README.md

@@ -21,6 +21,8 @@ addAudioCounters(total, usage);
 
 pi-audio does not apply that patch. Hosts such as slicc-agent record monkeypatches in `patches/patches.json`.
 
+`@ai-ecoverse/pi-audio/openai` registers OpenAI and OpenRouter speech adapters when `registerOpenAI()` is called. Batch transcription posts multipart audio to `/audio/transcriptions`. OpenRouter (`provider` `openrouter`, or a base URL on `openrouter.ai`) posts JSON `input_audio` instead. Speech posts JSON to `/audio/speech` and reads the raw audio response. OpenAI can return pcm, mp3, wav, opus, or flac. OpenRouter returns pcm or mp3. Realtime transcription uses a WebSocket at `/realtime?intent=transcription` for `gpt-realtime-whisper` and `gpt-live-transcribe`. The catalog key is type, provider, and id, so the batch model keeps the id `gpt-4o-transcribe`.
+
 `convertAudio` and `convertAudioSource` convert between `AudioFormat`s: PCM16, float32, G.711 μ-law and A-law, mono and stereo, sample rate, and PCM16 WAV. Opus, MP3 and FLAC pass through only when the format does not change. `encodeAudioBase64` and `decodeAudioBase64` are the wire helpers for JSON protocols. `AudioChunk.data` stays binary. The same converter runs in an AudioWorklet from `@ai-ecoverse/pi-audio/worklet` (`AUDIO_CONVERT_PROCESSOR`, `registerAudioConvertWorklet`). Adapters declare `inputFormats` and `outputFormats` and do not resample on their own.
 
 ## Development

package/dist/openai/catalog.js

@@ -0,0 +1,126 @@
+import {
+  registerVoiceModel
+} from "../models.js";
+import { registerSpeechApiProvider, registerTranscriptionApiProvider } from "../registry.js";
+import { listenOpenAI } from "./listen.js";
+import { speakOpenAI } from "./speak.js";
+import { transcribeOpenAI } from "./transcribe.js";
+const tokenCost = { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 };
+const openai = "https://api.openai.com/v1";
+const openrouter = "https://openrouter.ai/api/v1";
+const pcm24 = { encoding: "pcm16", sampleRate: 24e3, channels: 1 };
+const fileFormats = [
+  { encoding: "wav", sampleRate: 16e3, channels: 1 },
+  { encoding: "mp3", sampleRate: 16e3, channels: 1 },
+  { encoding: "flac", sampleRate: 16e3, channels: 1 }
+];
+const openAISpeechFormats = [
+  pcm24,
+  { encoding: "mp3", sampleRate: 24e3, channels: 1 },
+  { encoding: "wav", sampleRate: 24e3, channels: 1 },
+  { encoding: "opus", sampleRate: 24e3, channels: 1 },
+  { encoding: "flac", sampleRate: 24e3, channels: 1 }
+];
+const openRouterSpeechFormats = [
+  pcm24,
+  { encoding: "mp3", sampleRate: 24e3, channels: 1 }
+];
+const classicVoices = ["alloy", "ash", "coral", "echo", "fable", "onyx", "nova", "sage", "shimmer"];
+const miniVoices = [...classicVoices, "ballad", "verse", "marin", "cedar"];
+function voices(ids) {
+  return ids.map((id) => ({ id, name: id }));
+}
+function transcription(id, provider, baseUrl, perInputMinute, streaming, turnDetection, inputFormats) {
+  return {
+    id,
+    name: id,
+    api: streaming ? "openai-realtime-transcription" : "openai-transcriptions",
+    provider,
+    baseUrl,
+    type: "transcription",
+    streaming,
+    turnDetection,
+    inputFormats,
+    cost: tokenCost,
+    audioCost: { perInputMinute }
+  };
+}
+function speech(id, provider, baseUrl, outputFormats, voiceIds, audioCost) {
+  return {
+    id,
+    name: id,
+    api: "openai-speech",
+    provider,
+    baseUrl,
+    type: "speech",
+    textStreaming: false,
+    outputFormats,
+    voices: voices(voiceIds),
+    cloning: false,
+    cost: tokenCost,
+    audioCost
+  };
+}
+function openAICatalog() {
+  return [
+    transcription("whisper-1", "openai", openai, 6e-3, false, "none", fileFormats),
+    transcription("gpt-4o-transcribe", "openai", openai, 6e-3, false, "none", fileFormats),
+    transcription("gpt-4o-mini-transcribe", "openai", openai, 3e-3, false, "none", fileFormats),
+    transcription("gpt-realtime-whisper", "openai", openai, 0.017, true, "vad", [pcm24]),
+    transcription("gpt-live-transcribe", "openai", openai, 0.017, true, "none", [pcm24]),
+    transcription("openai/whisper-1", "openrouter", openrouter, 6e-3, false, "none", fileFormats),
+    transcription(
+      "openai/whisper-large-v3",
+      "openrouter",
+      openrouter,
+      6e-3,
+      false,
+      "none",
+      fileFormats
+    ),
+    speech("tts-1", "openai", openai, openAISpeechFormats, classicVoices, {
+      perThousandChars: 0.015
+    }),
+    speech("tts-1-hd", "openai", openai, openAISpeechFormats, classicVoices, {
+      perThousandChars: 0.03
+    }),
+    speech("gpt-4o-mini-tts", "openai", openai, openAISpeechFormats, miniVoices, {
+      perThousandChars: 15e-5,
+      perOutputMinute: 0.015
+    }),
+    speech(
+      "openai/gpt-4o-mini-tts-2025-12-15",
+      "openrouter",
+      openrouter,
+      openRouterSpeechFormats,
+      miniVoices,
+      { perThousandChars: 15e-5, perOutputMinute: 0.015 }
+    )
+  ];
+}
+function registerOpenAI() {
+  const models = openAICatalog();
+  const unregisterModels = models.map((model) => registerVoiceModel(model));
+  const unregisterBatch = registerTranscriptionApiProvider({
+    api: "openai-transcriptions",
+    transcribe: transcribeOpenAI
+  });
+  const unregisterRealtime = registerTranscriptionApiProvider({
+    api: "openai-realtime-transcription",
+    listen: listenOpenAI
+  });
+  const unregisterSpeech = registerSpeechApiProvider({ api: "openai-speech", speak: speakOpenAI });
+  let released = false;
+  return () => {
+    if (released) return;
+    released = true;
+    for (const unregister of unregisterModels) unregister();
+    unregisterBatch();
+    unregisterRealtime();
+    unregisterSpeech();
+  };
+}
+export {
+  registerOpenAI
+};
+//# sourceMappingURL=catalog.js.map

package/dist/openai/catalog.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../src/openai/catalog.ts"],
+  "sourcesContent": ["import type { AudioFormat } from '../audio.ts';\nimport {\n  registerVoiceModel,\n  type SpeechModel,\n  type TranscriptionModel,\n  type VoiceInfo,\n} from '../models.ts';\nimport { registerSpeechApiProvider, registerTranscriptionApiProvider } from '../registry.ts';\nimport { listenOpenAI } from './listen.ts';\nimport { speakOpenAI } from './speak.ts';\nimport { transcribeOpenAI } from './transcribe.ts';\n\nconst tokenCost = { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 };\nconst openai = 'https://api.openai.com/v1';\nconst openrouter = 'https://openrouter.ai/api/v1';\n\nconst pcm24: AudioFormat = { encoding: 'pcm16', sampleRate: 24000, channels: 1 };\nconst fileFormats: AudioFormat[] = [\n  { encoding: 'wav', sampleRate: 16000, channels: 1 },\n  { encoding: 'mp3', sampleRate: 16000, channels: 1 },\n  { encoding: 'flac', sampleRate: 16000, channels: 1 },\n];\nconst openAISpeechFormats: AudioFormat[] = [\n  pcm24,\n  { encoding: 'mp3', sampleRate: 24000, channels: 1 },\n  { encoding: 'wav', sampleRate: 24000, channels: 1 },\n  { encoding: 'opus', sampleRate: 24000, channels: 1 },\n  { encoding: 'flac', sampleRate: 24000, channels: 1 },\n];\nconst openRouterSpeechFormats: AudioFormat[] = [\n  pcm24,\n  { encoding: 'mp3', sampleRate: 24000, channels: 1 },\n];\n\nconst classicVoices = ['alloy', 'ash', 'coral', 'echo', 'fable', 'onyx', 'nova', 'sage', 'shimmer'];\nconst miniVoices = [...classicVoices, 'ballad', 'verse', 'marin', 'cedar'];\n\nfunction voices(ids: string[]): VoiceInfo[] {\n  return ids.map((id) => ({ id, name: id }));\n}\n\nfunction transcription(\n  id: string,\n  provider: string,\n  baseUrl: string,\n  perInputMinute: number,\n  streaming: boolean,\n  turnDetection: TranscriptionModel<string>['turnDetection'],\n  inputFormats: AudioFormat[]\n): TranscriptionModel<string> {\n  return {\n    id,\n    name: id,\n    api: streaming ? 'openai-realtime-transcription' : 'openai-transcriptions',\n    provider,\n    baseUrl,\n    type: 'transcription',\n    streaming,\n    turnDetection,\n    inputFormats,\n    cost: tokenCost,\n    audioCost: { perInputMinute },\n  };\n}\n\nfunction speech(\n  id: string,\n  provider: string,\n  baseUrl: string,\n  outputFormats: AudioFormat[],\n  voiceIds: string[],\n  audioCost: SpeechModel<string>['audioCost']\n): SpeechModel<string> {\n  return {\n    id,\n    name: id,\n    api: 'openai-speech',\n    provider,\n    baseUrl,\n    type: 'speech',\n    textStreaming: false,\n    outputFormats,\n    voices: voices(voiceIds),\n    cloning: false,\n    cost: tokenCost,\n    audioCost,\n  };\n}\n\nfunction openAICatalog(): Array<TranscriptionModel<string> | SpeechModel<string>> {\n  return [\n    transcription('whisper-1', 'openai', openai, 0.006, false, 'none', fileFormats),\n    transcription('gpt-4o-transcribe', 'openai', openai, 0.006, false, 'none', fileFormats),\n    transcription('gpt-4o-mini-transcribe', 'openai', openai, 0.003, false, 'none', fileFormats),\n    transcription('gpt-realtime-whisper', 'openai', openai, 0.017, true, 'vad', [pcm24]),\n    transcription('gpt-live-transcribe', 'openai', openai, 0.017, true, 'none', [pcm24]),\n    transcription('openai/whisper-1', 'openrouter', openrouter, 0.006, false, 'none', fileFormats),\n    transcription(\n      'openai/whisper-large-v3',\n      'openrouter',\n      openrouter,\n      0.006,\n      false,\n      'none',\n      fileFormats\n    ),\n    speech('tts-1', 'openai', openai, openAISpeechFormats, classicVoices, {\n      perThousandChars: 0.015,\n    }),\n    speech('tts-1-hd', 'openai', openai, openAISpeechFormats, classicVoices, {\n      perThousandChars: 0.03,\n    }),\n    speech('gpt-4o-mini-tts', 'openai', openai, openAISpeechFormats, miniVoices, {\n      perThousandChars: 0.00015,\n      perOutputMinute: 0.015,\n    }),\n    speech(\n      'openai/gpt-4o-mini-tts-2025-12-15',\n      'openrouter',\n      openrouter,\n      openRouterSpeechFormats,\n      miniVoices,\n      { perThousandChars: 0.00015, perOutputMinute: 0.015 }\n    ),\n  ];\n}\n\nexport function registerOpenAI(): () => void {\n  const models = openAICatalog();\n  const unregisterModels = models.map((model) => registerVoiceModel(model));\n  const unregisterBatch = registerTranscriptionApiProvider({\n    api: 'openai-transcriptions',\n    transcribe: transcribeOpenAI,\n  });\n  const unregisterRealtime = registerTranscriptionApiProvider({\n    api: 'openai-realtime-transcription',\n    listen: listenOpenAI,\n  });\n  const unregisterSpeech = registerSpeechApiProvider({ api: 'openai-speech', speak: speakOpenAI });\n  let released = false;\n  return () => {\n    if (released) return;\n    released = true;\n    for (const unregister of unregisterModels) unregister();\n    unregisterBatch();\n    unregisterRealtime();\n    unregisterSpeech();\n  };\n}\n"],
+  "mappings": "AACA;AAAA,EACE;AAAA,OAIK;AACP,SAAS,2BAA2B,wCAAwC;AAC5E,SAAS,oBAAoB;AAC7B,SAAS,mBAAmB;AAC5B,SAAS,wBAAwB;AAEjC,MAAM,YAAY,EAAE,OAAO,GAAG,QAAQ,GAAG,WAAW,GAAG,YAAY,EAAE;AACrE,MAAM,SAAS;AACf,MAAM,aAAa;AAEnB,MAAM,QAAqB,EAAE,UAAU,SAAS,YAAY,MAAO,UAAU,EAAE;AAC/E,MAAM,cAA6B;AAAA,EACjC,EAAE,UAAU,OAAO,YAAY,MAAO,UAAU,EAAE;AAAA,EAClD,EAAE,UAAU,OAAO,YAAY,MAAO,UAAU,EAAE;AAAA,EAClD,EAAE,UAAU,QAAQ,YAAY,MAAO,UAAU,EAAE;AACrD;AACA,MAAM,sBAAqC;AAAA,EACzC;AAAA,EACA,EAAE,UAAU,OAAO,YAAY,MAAO,UAAU,EAAE;AAAA,EAClD,EAAE,UAAU,OAAO,YAAY,MAAO,UAAU,EAAE;AAAA,EAClD,EAAE,UAAU,QAAQ,YAAY,MAAO,UAAU,EAAE;AAAA,EACnD,EAAE,UAAU,QAAQ,YAAY,MAAO,UAAU,EAAE;AACrD;AACA,MAAM,0BAAyC;AAAA,EAC7C;AAAA,EACA,EAAE,UAAU,OAAO,YAAY,MAAO,UAAU,EAAE;AACpD;AAEA,MAAM,gBAAgB,CAAC,SAAS,OAAO,SAAS,QAAQ,SAAS,QAAQ,QAAQ,QAAQ,SAAS;AAClG,MAAM,aAAa,CAAC,GAAG,eAAe,UAAU,SAAS,SAAS,OAAO;AAEzE,SAAS,OAAO,KAA4B;AAC1C,SAAO,IAAI,IAAI,CAAC,QAAQ,EAAE,IAAI,MAAM,GAAG,EAAE;AAC3C;AAEA,SAAS,cACP,IACA,UACA,SACA,gBACA,WACA,eACA,cAC4B;AAC5B,SAAO;AAAA,IACL;AAAA,IACA,MAAM;AAAA,IACN,KAAK,YAAY,kCAAkC;AAAA,IACnD;AAAA,IACA;AAAA,IACA,MAAM;AAAA,IACN;AAAA,IACA;AAAA,IACA;AAAA,IACA,MAAM;AAAA,IACN,WAAW,EAAE,eAAe;AAAA,EAC9B;AACF;AAEA,SAAS,OACP,IACA,UACA,SACA,eACA,UACA,WACqB;AACrB,SAAO;AAAA,IACL;AAAA,IACA,MAAM;AAAA,IACN,KAAK;AAAA,IACL;AAAA,IACA;AAAA,IACA,MAAM;AAAA,IACN,eAAe;AAAA,IACf;AAAA,IACA,QAAQ,OAAO,QAAQ;AAAA,IACvB,SAAS;AAAA,IACT,MAAM;AAAA,IACN;AAAA,EACF;AACF;AAEA,SAAS,gBAAyE;AAChF,SAAO;AAAA,IACL,cAAc,aAAa,UAAU,QAAQ,MAAO,OAAO,QAAQ,WAAW;AAAA,IAC9E,cAAc,qBAAqB,UAAU,QAAQ,MAAO,OAAO,QAAQ,WAAW;AAAA,IACtF,cAAc,0BAA0B,UAAU,QAAQ,MAAO,OAAO,QAAQ,WAAW;AAAA,IAC3F,cAAc,wBAAwB,UAAU,QAAQ,OAAO,MAAM,OAAO,CAAC,KAAK,CAAC;AAAA,IACnF,cAAc,uBAAuB,UAAU,QAAQ,OAAO,MAAM,QAAQ,CAAC,KAAK,CAAC;AAAA,IACnF,cAAc,oBAAoB,cAAc,YAAY,MAAO,OAAO,QAAQ,WAAW;AAAA,IAC7F;AAAA,MACE;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,IACF;AAAA,IACA,OAAO,SAAS,UAAU,QAAQ,qBAAqB,eAAe;AAAA,MACpE,kBAAkB;AAAA,IACpB,CAAC;AAAA,IACD,OAAO,YAAY,UAAU,QAAQ,qBAAqB,eAAe;AAAA,MACvE,kBAAkB;AAAA,IACpB,CAAC;AAAA,IACD,OAAO,mBAAmB,UAAU,QAAQ,qBAAqB,YAAY;AAAA,MAC3E,kBAAkB;AAAA,MAClB,iBAAiB;AAAA,IACnB,CAAC;AAAA,IACD;AAAA,MACE;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA,EAAE,kBAAkB,OAAS,iBAAiB,MAAM;AAAA,IACtD;AAAA,EACF;AACF;AAEO,SAAS,iBAA6B;AAC3C,QAAM,SAAS,cAAc;AAC7B,QAAM,mBAAmB,OAAO,IAAI,CAAC,UAAU,mBAAmB,KAAK,CAAC;AACxE,QAAM,kBAAkB,iCAAiC;AAAA,IACvD,KAAK;AAAA,IACL,YAAY;AAAA,EACd,CAAC;AACD,QAAM,qBAAqB,iCAAiC;AAAA,IAC1D,KAAK;AAAA,IACL,QAAQ;AAAA,EACV,CAAC;AACD,QAAM,mBAAmB,0BAA0B,EAAE,KAAK,iBAAiB,OAAO,YAAY,CAAC;AAC/F,MAAI,WAAW;AACf,SAAO,MAAM;AACX,QAAI,SAAU;AACd,eAAW;AACX,eAAW,cAAc,iBAAkB,YAAW;AACtD,oBAAgB;AAChB,uBAAmB;AACnB,qBAAiB;AAAA,EACnB;AACF;",
+  "names": []
+}

package/dist/openai/index.js

@@ -0,0 +1,5 @@
+import { registerOpenAI } from "./catalog.js";
+export {
+  registerOpenAI
+};
+//# sourceMappingURL=index.js.map

package/dist/openai/index.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../src/openai/index.ts"],
+  "sourcesContent": ["export { registerOpenAI } from './catalog.ts';\n"],
+  "mappings": "AAAA,SAAS,sBAAsB;",
+  "names": []
+}

package/dist/openai/listen.js

@@ -0,0 +1,200 @@
+import { encodeAudioBase64 } from "../convert.js";
+import {
+  TranscriptionEventStream
+} from "../transcription.js";
+import { priceAudio } from "../usage.js";
+import {
+  asVoiceError,
+  isRecord,
+  numberField,
+  realtimeUrl,
+  requestHeaders,
+  stringField
+} from "./wire.js";
+function turnDetection(model) {
+  if (model.turnDetection === "vad") return { type: "server_vad" };
+  if (model.turnDetection === "semantic") return { type: "semantic_vad" };
+  return null;
+}
+function sessionBody(model, options, format) {
+  const transcription = { model: model.id };
+  if (options?.language) {
+    transcription.language = options.language;
+    transcription.languages = [options.language];
+  }
+  if (options?.vocabulary && options.vocabulary.length > 0)
+    transcription.keywords = options.vocabulary;
+  return {
+    type: "session.update",
+    session: {
+      type: "transcription",
+      audio: {
+        input: {
+          format: { type: "audio/pcm", rate: format.sampleRate },
+          transcription,
+          turn_detection: turnDetection(model)
+        }
+      }
+    }
+  };
+}
+function matches(chunk, format) {
+  return chunk.format.encoding === format.encoding && chunk.format.sampleRate === format.sampleRate && chunk.format.channels === format.channels;
+}
+async function messageText(data) {
+  if (typeof data === "string") return data;
+  if (data instanceof ArrayBuffer) return new TextDecoder().decode(data);
+  if (data instanceof Uint8Array) return new TextDecoder().decode(data);
+  if (typeof Blob !== "undefined" && data instanceof Blob) return data.text();
+  throw new Error("Unsupported realtime message");
+}
+function applyEvent(event, state, push) {
+  const type = stringField(event, "type");
+  if (type === "input_audio_buffer.speech_started") {
+    const at = numberField(event, "audio_start_ms");
+    push({ type: "speech_start", at: at === void 0 ? 0 : at });
+    return;
+  }
+  if (type === "conversation.item.input_audio_transcription.delta") {
+    const item = stringField(event, "item_id");
+    state.segmentId = item === void 0 ? "0" : item;
+    const delta = stringField(event, "delta");
+    state.partial += delta === void 0 ? "" : delta;
+    push({ type: "partial", segmentId: state.segmentId, text: state.partial });
+    return;
+  }
+  if (type === "conversation.item.input_audio_transcription.completed") {
+    const transcript = stringField(event, "transcript");
+    const text = transcript === void 0 ? state.partial : transcript;
+    state.turns.push(text);
+    state.partial = "";
+    push({ type: "final", segmentId: state.segmentId, text });
+    push({ type: "turn_end", text, at: 0 });
+    return;
+  }
+  if (type === "error") {
+    const detail = event.error;
+    const message = isRecord(detail) ? stringField(detail, "message") : void 0;
+    state.failed = true;
+    push({
+      type: "error",
+      reason: "error",
+      error: message === void 0 ? "Realtime transcription failed" : message
+    });
+  }
+}
+function listenOpenAI(model, audio, options) {
+  const stream = new TranscriptionEventStream();
+  void pumpListen(model, audio, options, stream);
+  return stream;
+}
+function openSocket(socket, fail) {
+  return new Promise((resolve) => {
+    let opened = false;
+    socket.addEventListener("open", () => {
+      opened = true;
+      resolve();
+    });
+    socket.addEventListener("error", () => {
+      fail("OpenAI realtime transcription socket failed");
+      if (!opened) resolve();
+    });
+    socket.addEventListener("close", () => {
+      if (!opened) {
+        fail("OpenAI realtime transcription socket closed");
+        resolve();
+      }
+    });
+  });
+}
+async function pumpListen(model, audio, options, stream) {
+  const state = {
+    turns: [],
+    partial: "",
+    segmentId: "0",
+    bytes: 0,
+    audioDone: false,
+    failed: false
+  };
+  let finished = false;
+  const push = (event) => {
+    if (finished) return;
+    if (event.type === "done" || event.type === "error") finished = true;
+    stream.push(event);
+  };
+  let socket;
+  let wake;
+  const wait = new Promise((resolve) => {
+    wake = resolve;
+  });
+  try {
+    if (options?.signal?.aborted) throw new DOMException("aborted", "AbortError");
+    const format = model.inputFormats[0];
+    if (!format) throw new Error("No input format");
+    requestHeaders(model, options);
+    const apiKey = options?.apiKey;
+    if (!apiKey) throw new Error(`No API key for provider: ${model.provider}`);
+    socket = new WebSocket(realtimeUrl(model.baseUrl), [
+      "realtime",
+      `openai-insecure-api-key.${apiKey}`
+    ]);
+    let incoming = Promise.resolve();
+    socket.addEventListener("message", (event) => {
+      incoming = incoming.then(() => onMessage(event.data, state, push, wake));
+    });
+    socket.addEventListener("close", () => wake?.());
+    await openSocket(socket, (message) => {
+      push({ type: "error", reason: "error", error: message });
+      wake?.();
+    });
+    if (finished) return;
+    socket.send(JSON.stringify(sessionBody(model, options, format)));
+    for await (const chunk of audio) {
+      if (finished) return;
+      if (options?.signal?.aborted) throw new DOMException("aborted", "AbortError");
+      if (!matches(chunk, format))
+        throw new Error(`Unsupported audio format ${chunk.format.encoding}`);
+      state.bytes += chunk.data.length;
+      socket.send(
+        JSON.stringify({ type: "input_audio_buffer.append", audio: encodeAudioBase64(chunk.data) })
+      );
+    }
+    state.audioDone = true;
+    if (model.turnDetection === "none") {
+      socket.send(JSON.stringify({ type: "input_audio_buffer.commit" }));
+    }
+    if (!finished && (model.turnDetection === "none" || state.turns.length === 0)) await wait;
+    if (finished) return;
+    const width = format.sampleRate * format.channels * 2;
+    const inputSeconds = width > 0 ? state.bytes / width : 0;
+    push({
+      type: "done",
+      transcript: {
+        text: state.turns.join(" "),
+        usage: priceAudio(model.audioCost, { inputSeconds })
+      }
+    });
+  } catch (error) {
+    const voice = asVoiceError(error);
+    push({ type: "error", reason: voice.reason, error: voice.error });
+  } finally {
+    socket?.close();
+  }
+}
+async function onMessage(data, state, push, wake) {
+  try {
+    const parsed = JSON.parse(await messageText(data));
+    if (!isRecord(parsed)) throw new Error("Realtime transcription event was not JSON");
+    applyEvent(parsed, state, push);
+    if (state.failed || state.audioDone && state.turns.length > 0) wake?.();
+  } catch (error) {
+    const voice = asVoiceError(error);
+    state.failed = true;
+    push({ type: "error", reason: voice.reason, error: voice.error });
+    wake?.();
+  }
+}
+export {
+  listenOpenAI
+};
+//# sourceMappingURL=listen.js.map

package/dist/openai/listen.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../src/openai/listen.ts"],
+  "sourcesContent": ["import type { AudioChunk, AudioFormat } from '../audio.ts';\nimport { encodeAudioBase64 } from '../convert.ts';\nimport type { TranscriptionModel } from '../models.ts';\nimport {\n  type ListenOptions,\n  type TranscriptionEvent,\n  TranscriptionEventStream,\n} from '../transcription.ts';\nimport { priceAudio } from '../usage.ts';\nimport {\n  asVoiceError,\n  isRecord,\n  numberField,\n  realtimeUrl,\n  requestHeaders,\n  stringField,\n} from './wire.ts';\n\ninterface TurnState {\n  turns: string[];\n  partial: string;\n  segmentId: string;\n  bytes: number;\n  audioDone: boolean;\n  failed: boolean;\n}\n\nfunction turnDetection(model: TranscriptionModel<string>): Record<string, unknown> | null {\n  if (model.turnDetection === 'vad') return { type: 'server_vad' };\n  if (model.turnDetection === 'semantic') return { type: 'semantic_vad' };\n  return null;\n}\n\nfunction sessionBody(\n  model: TranscriptionModel<string>,\n  options: ListenOptions | undefined,\n  format: AudioFormat\n) {\n  const transcription: Record<string, unknown> = { model: model.id };\n  if (options?.language) {\n    transcription.language = options.language;\n    transcription.languages = [options.language];\n  }\n  if (options?.vocabulary && options.vocabulary.length > 0)\n    transcription.keywords = options.vocabulary;\n  return {\n    type: 'session.update',\n    session: {\n      type: 'transcription',\n      audio: {\n        input: {\n          format: { type: 'audio/pcm', rate: format.sampleRate },\n          transcription,\n          turn_detection: turnDetection(model),\n        },\n      },\n    },\n  };\n}\n\nfunction matches(chunk: AudioChunk, format: AudioFormat): boolean {\n  return (\n    chunk.format.encoding === format.encoding &&\n    chunk.format.sampleRate === format.sampleRate &&\n    chunk.format.channels === format.channels\n  );\n}\n\nasync function messageText(data: unknown): Promise<string> {\n  if (typeof data === 'string') return data;\n  if (data instanceof ArrayBuffer) return new TextDecoder().decode(data);\n  if (data instanceof Uint8Array) return new TextDecoder().decode(data);\n  if (typeof Blob !== 'undefined' && data instanceof Blob) return data.text();\n  throw new Error('Unsupported realtime message');\n}\n\nfunction applyEvent(\n  event: Record<string, unknown>,\n  state: TurnState,\n  push: (event: TranscriptionEvent) => void\n) {\n  const type = stringField(event, 'type');\n  if (type === 'input_audio_buffer.speech_started') {\n    const at = numberField(event, 'audio_start_ms');\n    push({ type: 'speech_start', at: at === undefined ? 0 : at });\n    return;\n  }\n  if (type === 'conversation.item.input_audio_transcription.delta') {\n    const item = stringField(event, 'item_id');\n    state.segmentId = item === undefined ? '0' : item;\n    const delta = stringField(event, 'delta');\n    state.partial += delta === undefined ? '' : delta;\n    push({ type: 'partial', segmentId: state.segmentId, text: state.partial });\n    return;\n  }\n  if (type === 'conversation.item.input_audio_transcription.completed') {\n    const transcript = stringField(event, 'transcript');\n    const text = transcript === undefined ? state.partial : transcript;\n    state.turns.push(text);\n    state.partial = '';\n    push({ type: 'final', segmentId: state.segmentId, text });\n    push({ type: 'turn_end', text, at: 0 });\n    return;\n  }\n  if (type === 'error') {\n    const detail = event.error;\n    const message = isRecord(detail) ? stringField(detail, 'message') : undefined;\n    state.failed = true;\n    push({\n      type: 'error',\n      reason: 'error',\n      error: message === undefined ? 'Realtime transcription failed' : message,\n    });\n  }\n}\n\nexport function listenOpenAI(\n  model: TranscriptionModel<string>,\n  audio: AsyncIterable<AudioChunk>,\n  options?: ListenOptions\n): TranscriptionEventStream {\n  const stream = new TranscriptionEventStream();\n  void pumpListen(model, audio, options, stream);\n  return stream;\n}\n\nfunction openSocket(socket: WebSocket, fail: (message: string) => void): Promise<void> {\n  return new Promise((resolve) => {\n    let opened = false;\n    socket.addEventListener('open', () => {\n      opened = true;\n      resolve();\n    });\n    socket.addEventListener('error', () => {\n      fail('OpenAI realtime transcription socket failed');\n      if (!opened) resolve();\n    });\n    socket.addEventListener('close', () => {\n      if (!opened) {\n        fail('OpenAI realtime transcription socket closed');\n        resolve();\n      }\n    });\n  });\n}\n\nasync function pumpListen(\n  model: TranscriptionModel<string>,\n  audio: AsyncIterable<AudioChunk>,\n  options: ListenOptions | undefined,\n  stream: TranscriptionEventStream\n) {\n  const state: TurnState = {\n    turns: [],\n    partial: '',\n    segmentId: '0',\n    bytes: 0,\n    audioDone: false,\n    failed: false,\n  };\n  let finished = false;\n  const push = (event: TranscriptionEvent) => {\n    if (finished) return;\n    if (event.type === 'done' || event.type === 'error') finished = true;\n    stream.push(event);\n  };\n  let socket: WebSocket | undefined;\n  let wake: (() => void) | undefined;\n  const wait = new Promise<void>((resolve) => {\n    wake = resolve;\n  });\n  try {\n    if (options?.signal?.aborted) throw new DOMException('aborted', 'AbortError');\n    const format = model.inputFormats[0];\n    if (!format) throw new Error('No input format');\n    requestHeaders(model, options);\n    const apiKey = options?.apiKey;\n    if (!apiKey) throw new Error(`No API key for provider: ${model.provider}`);\n    socket = new WebSocket(realtimeUrl(model.baseUrl), [\n      'realtime',\n      `openai-insecure-api-key.${apiKey}`,\n    ]);\n    let incoming = Promise.resolve();\n    socket.addEventListener('message', (event) => {\n      incoming = incoming.then(() => onMessage(event.data, state, push, wake));\n    });\n    socket.addEventListener('close', () => wake?.());\n    await openSocket(socket, (message) => {\n      push({ type: 'error', reason: 'error', error: message });\n      wake?.();\n    });\n    if (finished) return;\n    socket.send(JSON.stringify(sessionBody(model, options, format)));\n    for await (const chunk of audio) {\n      if (finished) return;\n      if (options?.signal?.aborted) throw new DOMException('aborted', 'AbortError');\n      if (!matches(chunk, format))\n        throw new Error(`Unsupported audio format ${chunk.format.encoding}`);\n      state.bytes += chunk.data.length;\n      socket.send(\n        JSON.stringify({ type: 'input_audio_buffer.append', audio: encodeAudioBase64(chunk.data) })\n      );\n    }\n    state.audioDone = true;\n    if (model.turnDetection === 'none') {\n      socket.send(JSON.stringify({ type: 'input_audio_buffer.commit' }));\n    }\n    if (!finished && (model.turnDetection === 'none' || state.turns.length === 0)) await wait;\n    if (finished) return;\n    const width = format.sampleRate * format.channels * 2;\n    const inputSeconds = width > 0 ? state.bytes / width : 0;\n    push({\n      type: 'done',\n      transcript: {\n        text: state.turns.join(' '),\n        usage: priceAudio(model.audioCost, { inputSeconds }),\n      },\n    });\n  } catch (error) {\n    const voice = asVoiceError(error);\n    push({ type: 'error', reason: voice.reason, error: voice.error });\n  } finally {\n    socket?.close();\n  }\n}\n\nasync function onMessage(\n  data: unknown,\n  state: TurnState,\n  push: (event: TranscriptionEvent) => void,\n  wake: (() => void) | undefined\n) {\n  try {\n    const parsed = JSON.parse(await messageText(data)) as unknown;\n    if (!isRecord(parsed)) throw new Error('Realtime transcription event was not JSON');\n    applyEvent(parsed, state, push);\n    if (state.failed || (state.audioDone && state.turns.length > 0)) wake?.();\n  } catch (error) {\n    const voice = asVoiceError(error);\n    state.failed = true;\n    push({ type: 'error', reason: voice.reason, error: voice.error });\n    wake?.();\n  }\n}\n"],
+  "mappings": "AACA,SAAS,yBAAyB;AAElC;AAAA,EAGE;AAAA,OACK;AACP,SAAS,kBAAkB;AAC3B;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;AAWP,SAAS,cAAc,OAAmE;AACxF,MAAI,MAAM,kBAAkB,MAAO,QAAO,EAAE,MAAM,aAAa;AAC/D,MAAI,MAAM,kBAAkB,WAAY,QAAO,EAAE,MAAM,eAAe;AACtE,SAAO;AACT;AAEA,SAAS,YACP,OACA,SACA,QACA;AACA,QAAM,gBAAyC,EAAE,OAAO,MAAM,GAAG;AACjE,MAAI,SAAS,UAAU;AACrB,kBAAc,WAAW,QAAQ;AACjC,kBAAc,YAAY,CAAC,QAAQ,QAAQ;AAAA,EAC7C;AACA,MAAI,SAAS,cAAc,QAAQ,WAAW,SAAS;AACrD,kBAAc,WAAW,QAAQ;AACnC,SAAO;AAAA,IACL,MAAM;AAAA,IACN,SAAS;AAAA,MACP,MAAM;AAAA,MACN,OAAO;AAAA,QACL,OAAO;AAAA,UACL,QAAQ,EAAE,MAAM,aAAa,MAAM,OAAO,WAAW;AAAA,UACrD;AAAA,UACA,gBAAgB,cAAc,KAAK;AAAA,QACrC;AAAA,MACF;AAAA,IACF;AAAA,EACF;AACF;AAEA,SAAS,QAAQ,OAAmB,QAA8B;AAChE,SACE,MAAM,OAAO,aAAa,OAAO,YACjC,MAAM,OAAO,eAAe,OAAO,cACnC,MAAM,OAAO,aAAa,OAAO;AAErC;AAEA,eAAe,YAAY,MAAgC;AACzD,MAAI,OAAO,SAAS,SAAU,QAAO;AACrC,MAAI,gBAAgB,YAAa,QAAO,IAAI,YAAY,EAAE,OAAO,IAAI;AACrE,MAAI,gBAAgB,WAAY,QAAO,IAAI,YAAY,EAAE,OAAO,IAAI;AACpE,MAAI,OAAO,SAAS,eAAe,gBAAgB,KAAM,QAAO,KAAK,KAAK;AAC1E,QAAM,IAAI,MAAM,8BAA8B;AAChD;AAEA,SAAS,WACP,OACA,OACA,MACA;AACA,QAAM,OAAO,YAAY,OAAO,MAAM;AACtC,MAAI,SAAS,qCAAqC;AAChD,UAAM,KAAK,YAAY,OAAO,gBAAgB;AAC9C,SAAK,EAAE,MAAM,gBAAgB,IAAI,OAAO,SAAY,IAAI,GAAG,CAAC;AAC5D;AAAA,EACF;AACA,MAAI,SAAS,qDAAqD;AAChE,UAAM,OAAO,YAAY,OAAO,SAAS;AACzC,UAAM,YAAY,SAAS,SAAY,MAAM;AAC7C,UAAM,QAAQ,YAAY,OAAO,OAAO;AACxC,UAAM,WAAW,UAAU,SAAY,KAAK;AAC5C,SAAK,EAAE,MAAM,WAAW,WAAW,MAAM,WAAW,MAAM,MAAM,QAAQ,CAAC;AACzE;AAAA,EACF;AACA,MAAI,SAAS,yDAAyD;AACpE,UAAM,aAAa,YAAY,OAAO,YAAY;AAClD,UAAM,OAAO,eAAe,SAAY,MAAM,UAAU;AACxD,UAAM,MAAM,KAAK,IAAI;AACrB,UAAM,UAAU;AAChB,SAAK,EAAE,MAAM,SAAS,WAAW,MAAM,WAAW,KAAK,CAAC;AACxD,SAAK,EAAE,MAAM,YAAY,MAAM,IAAI,EAAE,CAAC;AACtC;AAAA,EACF;AACA,MAAI,SAAS,SAAS;AACpB,UAAM,SAAS,MAAM;AACrB,UAAM,UAAU,SAAS,MAAM,IAAI,YAAY,QAAQ,SAAS,IAAI;AACpE,UAAM,SAAS;AACf,SAAK;AAAA,MACH,MAAM;AAAA,MACN,QAAQ;AAAA,MACR,OAAO,YAAY,SAAY,kCAAkC;AAAA,IACnE,CAAC;AAAA,EACH;AACF;AAEO,SAAS,aACd,OACA,OACA,SAC0B;AAC1B,QAAM,SAAS,IAAI,yBAAyB;AAC5C,OAAK,WAAW,OAAO,OAAO,SAAS,MAAM;AAC7C,SAAO;AACT;AAEA,SAAS,WAAW,QAAmB,MAAgD;AACrF,SAAO,IAAI,QAAQ,CAAC,YAAY;AAC9B,QAAI,SAAS;AACb,WAAO,iBAAiB,QAAQ,MAAM;AACpC,eAAS;AACT,cAAQ;AAAA,IACV,CAAC;AACD,WAAO,iBAAiB,SAAS,MAAM;AACrC,WAAK,6CAA6C;AAClD,UAAI,CAAC,OAAQ,SAAQ;AAAA,IACvB,CAAC;AACD,WAAO,iBAAiB,SAAS,MAAM;AACrC,UAAI,CAAC,QAAQ;AACX,aAAK,6CAA6C;AAClD,gBAAQ;AAAA,MACV;AAAA,IACF,CAAC;AAAA,EACH,CAAC;AACH;AAEA,eAAe,WACb,OACA,OACA,SACA,QACA;AACA,QAAM,QAAmB;AAAA,IACvB,OAAO,CAAC;AAAA,IACR,SAAS;AAAA,IACT,WAAW;AAAA,IACX,OAAO;AAAA,IACP,WAAW;AAAA,IACX,QAAQ;AAAA,EACV;AACA,MAAI,WAAW;AACf,QAAM,OAAO,CAAC,UAA8B;AAC1C,QAAI,SAAU;AACd,QAAI,MAAM,SAAS,UAAU,MAAM,SAAS,QAAS,YAAW;AAChE,WAAO,KAAK,KAAK;AAAA,EACnB;AACA,MAAI;AACJ,MAAI;AACJ,QAAM,OAAO,IAAI,QAAc,CAAC,YAAY;AAC1C,WAAO;AAAA,EACT,CAAC;AACD,MAAI;AACF,QAAI,SAAS,QAAQ,QAAS,OAAM,IAAI,aAAa,WAAW,YAAY;AAC5E,UAAM,SAAS,MAAM,aAAa,CAAC;AACnC,QAAI,CAAC,OAAQ,OAAM,IAAI,MAAM,iBAAiB;AAC9C,mBAAe,OAAO,OAAO;AAC7B,UAAM,SAAS,SAAS;AACxB,QAAI,CAAC,OAAQ,OAAM,IAAI,MAAM,4BAA4B,MAAM,QAAQ,EAAE;AACzE,aAAS,IAAI,UAAU,YAAY,MAAM,OAAO,GAAG;AAAA,MACjD;AAAA,MACA,2BAA2B,MAAM;AAAA,IACnC,CAAC;AACD,QAAI,WAAW,QAAQ,QAAQ;AAC/B,WAAO,iBAAiB,WAAW,CAAC,UAAU;AAC5C,iBAAW,SAAS,KAAK,MAAM,UAAU,MAAM,MAAM,OAAO,MAAM,IAAI,CAAC;AAAA,IACzE,CAAC;AACD,WAAO,iBAAiB,SAAS,MAAM,OAAO,CAAC;AAC/C,UAAM,WAAW,QAAQ,CAAC,YAAY;AACpC,WAAK,EAAE,MAAM,SAAS,QAAQ,SAAS,OAAO,QAAQ,CAAC;AACvD,aAAO;AAAA,IACT,CAAC;AACD,QAAI,SAAU;AACd,WAAO,KAAK,KAAK,UAAU,YAAY,OAAO,SAAS,MAAM,CAAC,CAAC;AAC/D,qBAAiB,SAAS,OAAO;AAC/B,UAAI,SAAU;AACd,UAAI,SAAS,QAAQ,QAAS,OAAM,IAAI,aAAa,WAAW,YAAY;AAC5E,UAAI,CAAC,QAAQ,OAAO,MAAM;AACxB,cAAM,IAAI,MAAM,4BAA4B,MAAM,OAAO,QAAQ,EAAE;AACrE,YAAM,SAAS,MAAM,KAAK;AAC1B,aAAO;AAAA,QACL,KAAK,UAAU,EAAE,MAAM,6BAA6B,OAAO,kBAAkB,MAAM,IAAI,EAAE,CAAC;AAAA,MAC5F;AAAA,IACF;AACA,UAAM,YAAY;AAClB,QAAI,MAAM,kBAAkB,QAAQ;AAClC,aAAO,KAAK,KAAK,UAAU,EAAE,MAAM,4BAA4B,CAAC,CAAC;AAAA,IACnE;AACA,QAAI,CAAC,aAAa,MAAM,kBAAkB,UAAU,MAAM,MAAM,WAAW,GAAI,OAAM;AACrF,QAAI,SAAU;AACd,UAAM,QAAQ,OAAO,aAAa,OAAO,WAAW;AACpD,UAAM,eAAe,QAAQ,IAAI,MAAM,QAAQ,QAAQ;AACvD,SAAK;AAAA,MACH,MAAM;AAAA,MACN,YAAY;AAAA,QACV,MAAM,MAAM,MAAM,KAAK,GAAG;AAAA,QAC1B,OAAO,WAAW,MAAM,WAAW,EAAE,aAAa,CAAC;AAAA,MACrD;AAAA,IACF,CAAC;AAAA,EACH,SAAS,OAAO;AACd,UAAM,QAAQ,aAAa,KAAK;AAChC,SAAK,EAAE,MAAM,SAAS,QAAQ,MAAM,QAAQ,OAAO,MAAM,MAAM,CAAC;AAAA,EAClE,UAAE;AACA,YAAQ,MAAM;AAAA,EAChB;AACF;AAEA,eAAe,UACb,MACA,OACA,MACA,MACA;AACA,MAAI;AACF,UAAM,SAAS,KAAK,MAAM,MAAM,YAAY,IAAI,CAAC;AACjD,QAAI,CAAC,SAAS,MAAM,EAAG,OAAM,IAAI,MAAM,2CAA2C;AAClF,eAAW,QAAQ,OAAO,IAAI;AAC9B,QAAI,MAAM,UAAW,MAAM,aAAa,MAAM,MAAM,SAAS,EAAI,QAAO;AAAA,EAC1E,SAAS,OAAO;AACd,UAAM,QAAQ,aAAa,KAAK;AAChC,UAAM,SAAS;AACf,SAAK,EAAE,MAAM,SAAS,QAAQ,MAAM,QAAQ,OAAO,MAAM,MAAM,CAAC;AAChE,WAAO;AAAA,EACT;AACF;",
+  "names": []
+}

package/dist/openai/speak.js

@@ -0,0 +1,91 @@
+import { VoiceRequestError } from "../errors.js";
+import { SpeechEventStream } from "../speech.js";
+import { priceAudio } from "../usage.js";
+import {
+  applyPayload,
+  asVoiceError,
+  endpoint,
+  isOpenRouter,
+  isRecord,
+  voiceFetch
+} from "./wire.js";
+function sameFormat(left, right) {
+  return left.encoding === right.encoding && left.sampleRate === right.sampleRate && left.channels === right.channels;
+}
+function resolveSpeechFormat(model, requested) {
+  const fallback = model.outputFormats[0];
+  if (!fallback) throw new Error("No output format");
+  if (!requested) return fallback;
+  const match = model.outputFormats.find((format) => sameFormat(format, requested));
+  if (!match) throw new Error(`Unsupported speech format ${requested.encoding}`);
+  return match;
+}
+function speechResponseFormat(encoding, openRouter) {
+  if (encoding === "pcm16") return "pcm";
+  if (encoding === "mp3") return "mp3";
+  if (openRouter) throw new Error(`OpenRouter speech cannot return ${encoding}`);
+  if (encoding === "wav" || encoding === "opus" || encoding === "flac") return encoding;
+  throw new Error(`OpenAI speech cannot return ${encoding}`);
+}
+function outputSeconds(format, byteLength) {
+  if (format.encoding !== "pcm16") return void 0;
+  const width = format.sampleRate * format.channels * 2;
+  return width > 0 ? byteLength / width : 0;
+}
+async function readText(text, signal) {
+  if (typeof text === "string") return text;
+  let combined = "";
+  for await (const part of text) {
+    if (signal?.aborted) throw new VoiceRequestError("aborted", "aborted");
+    combined += part;
+  }
+  return combined;
+}
+function speakOpenAI(model, input, options) {
+  const stream = new SpeechEventStream();
+  void pumpSpeech(model, input, options, stream);
+  return stream;
+}
+async function pumpSpeech(model, input, options, stream) {
+  try {
+    if (options?.signal?.aborted) throw new DOMException("aborted", "AbortError");
+    const format = resolveSpeechFormat(model, input.format);
+    const openRouter = isOpenRouter(model);
+    const responseFormat = speechResponseFormat(format.encoding, openRouter);
+    const text = await readText(input.text, options?.signal);
+    const payload = {
+      model: model.id,
+      input: text,
+      voice: input.voice,
+      response_format: responseFormat
+    };
+    if (input.instructions !== void 0) payload.instructions = input.instructions;
+    const body = await applyPayload(payload, model, options);
+    if (!isRecord(body)) throw new Error("Speech payload was not JSON");
+    const response = await voiceFetch(
+      model,
+      options,
+      endpoint(model.baseUrl, "audio/speech"),
+      JSON.stringify(body),
+      "application/json"
+    );
+    const data = new Uint8Array(await response.arrayBuffer());
+    const seconds = outputSeconds(format, data.length);
+    const meters = { characters: text.length };
+    if (seconds !== void 0) meters.outputSeconds = seconds;
+    stream.push({ type: "start", format });
+    stream.push({
+      type: "audio",
+      chunk: { data, format, timestamp: 0 },
+      span: { itemId: model.id, start: 0, end: text.length }
+    });
+    stream.push({ type: "done", usage: priceAudio(model.audioCost, meters) });
+  } catch (error) {
+    const voice = asVoiceError(error);
+    stream.push({ type: "error", reason: voice.reason, error: voice.error });
+  }
+}
+export {
+  speakOpenAI
+};
+//# sourceMappingURL=speak.js.map

package/dist/openai/speak.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../src/openai/speak.ts"],
+  "sourcesContent": ["import type { ProviderRequestOptions } from '@earendil-works/pi-ai';\nimport type { AudioFormat } from '../audio.ts';\nimport { VoiceRequestError } from '../errors.ts';\nimport type { SpeechModel } from '../models.ts';\nimport { SpeechEventStream, type SpeechInput } from '../speech.ts';\nimport { type AudioCounters, priceAudio } from '../usage.ts';\nimport {\n  applyPayload,\n  asVoiceError,\n  endpoint,\n  isOpenRouter,\n  isRecord,\n  voiceFetch,\n} from './wire.ts';\n\nfunction sameFormat(left: AudioFormat, right: AudioFormat): boolean {\n  return (\n    left.encoding === right.encoding &&\n    left.sampleRate === right.sampleRate &&\n    left.channels === right.channels\n  );\n}\n\nfunction resolveSpeechFormat(\n  model: SpeechModel<string>,\n  requested: AudioFormat | undefined\n): AudioFormat {\n  const fallback = model.outputFormats[0];\n  if (!fallback) throw new Error('No output format');\n  if (!requested) return fallback;\n  const match = model.outputFormats.find((format) => sameFormat(format, requested));\n  if (!match) throw new Error(`Unsupported speech format ${requested.encoding}`);\n  return match;\n}\n\nfunction speechResponseFormat(encoding: AudioFormat['encoding'], openRouter: boolean): string {\n  if (encoding === 'pcm16') return 'pcm';\n  if (encoding === 'mp3') return 'mp3';\n  if (openRouter) throw new Error(`OpenRouter speech cannot return ${encoding}`);\n  if (encoding === 'wav' || encoding === 'opus' || encoding === 'flac') return encoding;\n  throw new Error(`OpenAI speech cannot return ${encoding}`);\n}\n\nfunction outputSeconds(format: AudioFormat, byteLength: number): number | undefined {\n  if (format.encoding !== 'pcm16') return undefined;\n  const width = format.sampleRate * format.channels * 2;\n  return width > 0 ? byteLength / width : 0;\n}\n\nasync function readText(\n  text: SpeechInput['text'],\n  signal: AbortSignal | undefined\n): Promise<string> {\n  if (typeof text === 'string') return text;\n  let combined = '';\n  for await (const part of text) {\n    if (signal?.aborted) throw new VoiceRequestError('aborted', 'aborted');\n    combined += part;\n  }\n  return combined;\n}\n\nexport function speakOpenAI(\n  model: SpeechModel<string>,\n  input: SpeechInput,\n  options?: ProviderRequestOptions\n): SpeechEventStream {\n  const stream = new SpeechEventStream();\n  void pumpSpeech(model, input, options, stream);\n  return stream;\n}\n\nasync function pumpSpeech(\n  model: SpeechModel<string>,\n  input: SpeechInput,\n  options: ProviderRequestOptions | undefined,\n  stream: SpeechEventStream\n) {\n  try {\n    if (options?.signal?.aborted) throw new DOMException('aborted', 'AbortError');\n    const format = resolveSpeechFormat(model, input.format);\n    const openRouter = isOpenRouter(model);\n    const responseFormat = speechResponseFormat(format.encoding, openRouter);\n    const text = await readText(input.text, options?.signal);\n    const payload: Record<string, unknown> = {\n      model: model.id,\n      input: text,\n      voice: input.voice,\n      response_format: responseFormat,\n    };\n    if (input.instructions !== undefined) payload.instructions = input.instructions;\n    const body = await applyPayload(payload, model, options);\n    if (!isRecord(body)) throw new Error('Speech payload was not JSON');\n    const response = await voiceFetch(\n      model,\n      options,\n      endpoint(model.baseUrl, 'audio/speech'),\n      JSON.stringify(body),\n      'application/json'\n    );\n    const data = new Uint8Array(await response.arrayBuffer());\n    const seconds = outputSeconds(format, data.length);\n    const meters: AudioCounters = { characters: text.length };\n    if (seconds !== undefined) meters.outputSeconds = seconds;\n    stream.push({ type: 'start', format });\n    stream.push({\n      type: 'audio',\n      chunk: { data, format, timestamp: 0 },\n      span: { itemId: model.id, start: 0, end: text.length },\n    });\n    stream.push({ type: 'done', usage: priceAudio(model.audioCost, meters) });\n  } catch (error) {\n    const voice = asVoiceError(error);\n    stream.push({ type: 'error', reason: voice.reason, error: voice.error });\n  }\n}\n"],
+  "mappings": "AAEA,SAAS,yBAAyB;AAElC,SAAS,yBAA2C;AACpD,SAA6B,kBAAkB;AAC/C;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;AAEP,SAAS,WAAW,MAAmB,OAA6B;AAClE,SACE,KAAK,aAAa,MAAM,YACxB,KAAK,eAAe,MAAM,cAC1B,KAAK,aAAa,MAAM;AAE5B;AAEA,SAAS,oBACP,OACA,WACa;AACb,QAAM,WAAW,MAAM,cAAc,CAAC;AACtC,MAAI,CAAC,SAAU,OAAM,IAAI,MAAM,kBAAkB;AACjD,MAAI,CAAC,UAAW,QAAO;AACvB,QAAM,QAAQ,MAAM,cAAc,KAAK,CAAC,WAAW,WAAW,QAAQ,SAAS,CAAC;AAChF,MAAI,CAAC,MAAO,OAAM,IAAI,MAAM,6BAA6B,UAAU,QAAQ,EAAE;AAC7E,SAAO;AACT;AAEA,SAAS,qBAAqB,UAAmC,YAA6B;AAC5F,MAAI,aAAa,QAAS,QAAO;AACjC,MAAI,aAAa,MAAO,QAAO;AAC/B,MAAI,WAAY,OAAM,IAAI,MAAM,mCAAmC,QAAQ,EAAE;AAC7E,MAAI,aAAa,SAAS,aAAa,UAAU,aAAa,OAAQ,QAAO;AAC7E,QAAM,IAAI,MAAM,+BAA+B,QAAQ,EAAE;AAC3D;AAEA,SAAS,cAAc,QAAqB,YAAwC;AAClF,MAAI,OAAO,aAAa,QAAS,QAAO;AACxC,QAAM,QAAQ,OAAO,aAAa,OAAO,WAAW;AACpD,SAAO,QAAQ,IAAI,aAAa,QAAQ;AAC1C;AAEA,eAAe,SACb,MACA,QACiB;AACjB,MAAI,OAAO,SAAS,SAAU,QAAO;AACrC,MAAI,WAAW;AACf,mBAAiB,QAAQ,MAAM;AAC7B,QAAI,QAAQ,QAAS,OAAM,IAAI,kBAAkB,WAAW,SAAS;AACrE,gBAAY;AAAA,EACd;AACA,SAAO;AACT;AAEO,SAAS,YACd,OACA,OACA,SACmB;AACnB,QAAM,SAAS,IAAI,kBAAkB;AACrC,OAAK,WAAW,OAAO,OAAO,SAAS,MAAM;AAC7C,SAAO;AACT;AAEA,eAAe,WACb,OACA,OACA,SACA,QACA;AACA,MAAI;AACF,QAAI,SAAS,QAAQ,QAAS,OAAM,IAAI,aAAa,WAAW,YAAY;AAC5E,UAAM,SAAS,oBAAoB,OAAO,MAAM,MAAM;AACtD,UAAM,aAAa,aAAa,KAAK;AACrC,UAAM,iBAAiB,qBAAqB,OAAO,UAAU,UAAU;AACvE,UAAM,OAAO,MAAM,SAAS,MAAM,MAAM,SAAS,MAAM;AACvD,UAAM,UAAmC;AAAA,MACvC,OAAO,MAAM;AAAA,MACb,OAAO;AAAA,MACP,OAAO,MAAM;AAAA,MACb,iBAAiB;AAAA,IACnB;AACA,QAAI,MAAM,iBAAiB,OAAW,SAAQ,eAAe,MAAM;AACnE,UAAM,OAAO,MAAM,aAAa,SAAS,OAAO,OAAO;AACvD,QAAI,CAAC,SAAS,IAAI,EAAG,OAAM,IAAI,MAAM,6BAA6B;AAClE,UAAM,WAAW,MAAM;AAAA,MACrB;AAAA,MACA;AAAA,MACA,SAAS,MAAM,SAAS,cAAc;AAAA,MACtC,KAAK,UAAU,IAAI;AAAA,MACnB;AAAA,IACF;AACA,UAAM,OAAO,IAAI,WAAW,MAAM,SAAS,YAAY,CAAC;AACxD,UAAM,UAAU,cAAc,QAAQ,KAAK,MAAM;AACjD,UAAM,SAAwB,EAAE,YAAY,KAAK,OAAO;AACxD,QAAI,YAAY,OAAW,QAAO,gBAAgB;AAClD,WAAO,KAAK,EAAE,MAAM,SAAS,OAAO,CAAC;AACrC,WAAO,KAAK;AAAA,MACV,MAAM;AAAA,MACN,OAAO,EAAE,MAAM,QAAQ,WAAW,EAAE;AAAA,MACpC,MAAM,EAAE,QAAQ,MAAM,IAAI,OAAO,GAAG,KAAK,KAAK,OAAO;AAAA,IACvD,CAAC;AACD,WAAO,KAAK,EAAE,MAAM,QAAQ,OAAO,WAAW,MAAM,WAAW,MAAM,EAAE,CAAC;AAAA,EAC1E,SAAS,OAAO;AACd,UAAM,QAAQ,aAAa,KAAK;AAChC,WAAO,KAAK,EAAE,MAAM,SAAS,QAAQ,MAAM,QAAQ,OAAO,MAAM,MAAM,CAAC;AAAA,EACzE;AACF;",
+  "names": []
+}

package/dist/openai/transcribe.js

@@ -0,0 +1,112 @@
+import { VoiceRequestError } from "../errors.js";
+import { priceAudio } from "../usage.js";
+import {
+  applyPayload,
+  asVoiceError,
+  audioFile,
+  audioPayload,
+  encodeMultipart,
+  endpoint,
+  isOpenRouter,
+  isRecord,
+  multipartType,
+  numberField,
+  readAudio,
+  readJson,
+  requestSignal,
+  stringField,
+  voiceFetch
+} from "./wire.js";
+function durationSeconds(record) {
+  const usage = record.usage;
+  if (isRecord(usage)) {
+    const seconds = numberField(usage, "seconds");
+    if (seconds !== void 0) return seconds;
+  }
+  const duration = numberField(record, "duration");
+  return duration === void 0 ? 0 : duration;
+}
+function fieldText(segment) {
+  const text = stringField(segment, "text");
+  return text === void 0 ? "" : text;
+}
+function fieldNumber(segment, key) {
+  const value = numberField(segment, key);
+  return value === void 0 ? 0 : value;
+}
+function segmentFrom(segment) {
+  const speaker = stringField(segment, "speaker");
+  const mapped = {
+    text: fieldText(segment),
+    start: fieldNumber(segment, "start"),
+    end: fieldNumber(segment, "end")
+  };
+  if (speaker !== void 0) mapped.speaker = speaker;
+  return mapped;
+}
+function transcriptFrom(body, model, seconds) {
+  if (!isRecord(body)) throw new Error("Transcription response had no text");
+  const text = stringField(body, "text");
+  if (text === void 0) throw new Error("Transcription response had no text");
+  const language = stringField(body, "language");
+  const segments = Array.isArray(body.segments) ? body.segments.filter(isRecord).map(segmentFrom) : void 0;
+  return {
+    text,
+    ...language !== void 0 ? { language } : {},
+    ...segments !== void 0 ? { segments } : {},
+    usage: priceAudio(model.audioCost, { inputSeconds: seconds })
+  };
+}
+function fileFrom(loaded) {
+  const named = audioFile(loaded.mimeType);
+  return { data: loaded.data, mimeType: loaded.mimeType, ...named };
+}
+async function transcribeOpenAI(model, input, options) {
+  try {
+    if (options?.signal?.aborted) throw new DOMException("aborted", "AbortError");
+    const fetchImpl = options?.fetch ?? globalThis.fetch;
+    const loaded = await readAudio(input, fetchImpl, requestSignal(options));
+    const file = fileFrom(loaded);
+    const payload = audioPayload(file, isOpenRouter(model), model.id);
+    const body = await applyPayload(payload, model, options);
+    if (!isRecord(body)) throw new Error("Transcription payload was not JSON");
+    const url = endpoint(model.baseUrl, "audio/transcriptions");
+    const response = await postTranscription(model, options, url, body);
+    const json = await readJson(response);
+    const seconds = isRecord(json) ? durationSeconds(json) : 0;
+    return transcriptFrom(json, model, seconds);
+  } catch (error) {
+    const voice = asVoiceError(error);
+    throw new VoiceRequestError(voice.reason, voice.error);
+  }
+}
+async function postTranscription(model, options, url, body) {
+  const inline = body.input_audio;
+  if (isRecord(inline)) {
+    return voiceFetch(model, options, url, JSON.stringify(body), "application/json");
+  }
+  const file = body.file;
+  if (!isRecord(file) || !(file.data instanceof Uint8Array) || typeof file.mimeType !== "string") {
+    throw new Error("Transcription payload was not JSON");
+  }
+  const named = audioFile(file.mimeType);
+  const fields = {};
+  const modelId = stringField(body, "model");
+  fields.model = modelId === void 0 ? model.id : modelId;
+  const encoded = encodeMultipart(
+    {
+      data: file.data,
+      mimeType: file.mimeType,
+      filename: typeof file.filename === "string" ? file.filename : named.filename,
+      format: named.format
+    },
+    fields
+  );
+  const copy = new ArrayBuffer(encoded.byteLength);
+  new Uint8Array(copy).set(encoded);
+  return voiceFetch(model, options, url, copy, multipartType());
+}
+export {
+  transcribeOpenAI
+};
+//# sourceMappingURL=transcribe.js.map

package/dist/openai/transcribe.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../src/openai/transcribe.ts"],
+  "sourcesContent": ["import type { ProviderRequestOptions } from '@earendil-works/pi-ai';\nimport { VoiceRequestError } from '../errors.ts';\nimport type { TranscriptionModel } from '../models.ts';\nimport type { AudioInput, Transcript, TranscriptSegment } from '../transcription.ts';\nimport { priceAudio } from '../usage.ts';\nimport {\n  type AudioFile,\n  applyPayload,\n  asVoiceError,\n  audioFile,\n  audioPayload,\n  encodeMultipart,\n  endpoint,\n  isOpenRouter,\n  isRecord,\n  multipartType,\n  numberField,\n  readAudio,\n  readJson,\n  requestSignal,\n  stringField,\n  voiceFetch,\n} from './wire.ts';\n\nfunction durationSeconds(record: Record<string, unknown>): number {\n  const usage = record.usage;\n  if (isRecord(usage)) {\n    const seconds = numberField(usage, 'seconds');\n    if (seconds !== undefined) return seconds;\n  }\n  const duration = numberField(record, 'duration');\n  return duration === undefined ? 0 : duration;\n}\n\nfunction fieldText(segment: Record<string, unknown>): string {\n  const text = stringField(segment, 'text');\n  return text === undefined ? '' : text;\n}\n\nfunction fieldNumber(segment: Record<string, unknown>, key: string): number {\n  const value = numberField(segment, key);\n  return value === undefined ? 0 : value;\n}\n\nfunction segmentFrom(segment: Record<string, unknown>): TranscriptSegment {\n  const speaker = stringField(segment, 'speaker');\n  const mapped: TranscriptSegment = {\n    text: fieldText(segment),\n    start: fieldNumber(segment, 'start'),\n    end: fieldNumber(segment, 'end'),\n  };\n  if (speaker !== undefined) mapped.speaker = speaker;\n  return mapped;\n}\n\nfunction transcriptFrom(\n  body: unknown,\n  model: TranscriptionModel<string>,\n  seconds: number\n): Transcript {\n  if (!isRecord(body)) throw new Error('Transcription response had no text');\n  const text = stringField(body, 'text');\n  if (text === undefined) throw new Error('Transcription response had no text');\n  const language = stringField(body, 'language');\n  const segments = Array.isArray(body.segments)\n    ? body.segments.filter(isRecord).map(segmentFrom)\n    : undefined;\n  return {\n    text,\n    ...(language !== undefined ? { language } : {}),\n    ...(segments !== undefined ? { segments } : {}),\n    usage: priceAudio(model.audioCost, { inputSeconds: seconds }),\n  };\n}\n\nfunction fileFrom(loaded: { data: Uint8Array; mimeType: string }): AudioFile {\n  const named = audioFile(loaded.mimeType);\n  return { data: loaded.data, mimeType: loaded.mimeType, ...named };\n}\n\nexport async function transcribeOpenAI(\n  model: TranscriptionModel<string>,\n  input: AudioInput,\n  options?: ProviderRequestOptions\n): Promise<Transcript> {\n  try {\n    if (options?.signal?.aborted) throw new DOMException('aborted', 'AbortError');\n    const fetchImpl = options?.fetch ?? globalThis.fetch;\n    const loaded = await readAudio(input, fetchImpl, requestSignal(options));\n    const file = fileFrom(loaded);\n    const payload = audioPayload(file, isOpenRouter(model), model.id);\n    const body = await applyPayload(payload, model, options);\n    if (!isRecord(body)) throw new Error('Transcription payload was not JSON');\n    const url = endpoint(model.baseUrl, 'audio/transcriptions');\n    const response = await postTranscription(model, options, url, body);\n    const json = await readJson(response);\n    const seconds = isRecord(json) ? durationSeconds(json) : 0;\n    return transcriptFrom(json, model, seconds);\n  } catch (error) {\n    const voice = asVoiceError(error);\n    throw new VoiceRequestError(voice.reason, voice.error);\n  }\n}\n\nasync function postTranscription(\n  model: TranscriptionModel<string>,\n  options: ProviderRequestOptions | undefined,\n  url: string,\n  body: Record<string, unknown>\n): Promise<Response> {\n  const inline = body.input_audio;\n  if (isRecord(inline)) {\n    return voiceFetch(model, options, url, JSON.stringify(body), 'application/json');\n  }\n  const file = body.file;\n  if (!isRecord(file) || !(file.data instanceof Uint8Array) || typeof file.mimeType !== 'string') {\n    throw new Error('Transcription payload was not JSON');\n  }\n  const named = audioFile(file.mimeType);\n  const fields: Record<string, string> = {};\n  const modelId = stringField(body, 'model');\n  fields.model = modelId === undefined ? model.id : modelId;\n  const encoded = encodeMultipart(\n    {\n      data: file.data,\n      mimeType: file.mimeType,\n      filename: typeof file.filename === 'string' ? file.filename : named.filename,\n      format: named.format,\n    },\n    fields\n  );\n  const copy = new ArrayBuffer(encoded.byteLength);\n  new Uint8Array(copy).set(encoded);\n  return voiceFetch(model, options, url, copy, multipartType());\n}\n"],
+  "mappings": "AACA,SAAS,yBAAyB;AAGlC,SAAS,kBAAkB;AAC3B;AAAA,EAEE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;AAEP,SAAS,gBAAgB,QAAyC;AAChE,QAAM,QAAQ,OAAO;AACrB,MAAI,SAAS,KAAK,GAAG;AACnB,UAAM,UAAU,YAAY,OAAO,SAAS;AAC5C,QAAI,YAAY,OAAW,QAAO;AAAA,EACpC;AACA,QAAM,WAAW,YAAY,QAAQ,UAAU;AAC/C,SAAO,aAAa,SAAY,IAAI;AACtC;AAEA,SAAS,UAAU,SAA0C;AAC3D,QAAM,OAAO,YAAY,SAAS,MAAM;AACxC,SAAO,SAAS,SAAY,KAAK;AACnC;AAEA,SAAS,YAAY,SAAkC,KAAqB;AAC1E,QAAM,QAAQ,YAAY,SAAS,GAAG;AACtC,SAAO,UAAU,SAAY,IAAI;AACnC;AAEA,SAAS,YAAY,SAAqD;AACxE,QAAM,UAAU,YAAY,SAAS,SAAS;AAC9C,QAAM,SAA4B;AAAA,IAChC,MAAM,UAAU,OAAO;AAAA,IACvB,OAAO,YAAY,SAAS,OAAO;AAAA,IACnC,KAAK,YAAY,SAAS,KAAK;AAAA,EACjC;AACA,MAAI,YAAY,OAAW,QAAO,UAAU;AAC5C,SAAO;AACT;AAEA,SAAS,eACP,MACA,OACA,SACY;AACZ,MAAI,CAAC,SAAS,IAAI,EAAG,OAAM,IAAI,MAAM,oCAAoC;AACzE,QAAM,OAAO,YAAY,MAAM,MAAM;AACrC,MAAI,SAAS,OAAW,OAAM,IAAI,MAAM,oCAAoC;AAC5E,QAAM,WAAW,YAAY,MAAM,UAAU;AAC7C,QAAM,WAAW,MAAM,QAAQ,KAAK,QAAQ,IACxC,KAAK,SAAS,OAAO,QAAQ,EAAE,IAAI,WAAW,IAC9C;AACJ,SAAO;AAAA,IACL;AAAA,IACA,GAAI,aAAa,SAAY,EAAE,SAAS,IAAI,CAAC;AAAA,IAC7C,GAAI,aAAa,SAAY,EAAE,SAAS,IAAI,CAAC;AAAA,IAC7C,OAAO,WAAW,MAAM,WAAW,EAAE,cAAc,QAAQ,CAAC;AAAA,EAC9D;AACF;AAEA,SAAS,SAAS,QAA2D;AAC3E,QAAM,QAAQ,UAAU,OAAO,QAAQ;AACvC,SAAO,EAAE,MAAM,OAAO,MAAM,UAAU,OAAO,UAAU,GAAG,MAAM;AAClE;AAEA,eAAsB,iBACpB,OACA,OACA,SACqB;AACrB,MAAI;AACF,QAAI,SAAS,QAAQ,QAAS,OAAM,IAAI,aAAa,WAAW,YAAY;AAC5E,UAAM,YAAY,SAAS,SAAS,WAAW;AAC/C,UAAM,SAAS,MAAM,UAAU,OAAO,WAAW,cAAc,OAAO,CAAC;AACvE,UAAM,OAAO,SAAS,MAAM;AAC5B,UAAM,UAAU,aAAa,MAAM,aAAa,KAAK,GAAG,MAAM,EAAE;AAChE,UAAM,OAAO,MAAM,aAAa,SAAS,OAAO,OAAO;AACvD,QAAI,CAAC,SAAS,IAAI,EAAG,OAAM,IAAI,MAAM,oCAAoC;AACzE,UAAM,MAAM,SAAS,MAAM,SAAS,sBAAsB;AAC1D,UAAM,WAAW,MAAM,kBAAkB,OAAO,SAAS,KAAK,IAAI;AAClE,UAAM,OAAO,MAAM,SAAS,QAAQ;AACpC,UAAM,UAAU,SAAS,IAAI,IAAI,gBAAgB,IAAI,IAAI;AACzD,WAAO,eAAe,MAAM,OAAO,OAAO;AAAA,EAC5C,SAAS,OAAO;AACd,UAAM,QAAQ,aAAa,KAAK;AAChC,UAAM,IAAI,kBAAkB,MAAM,QAAQ,MAAM,KAAK;AAAA,EACvD;AACF;AAEA,eAAe,kBACb,OACA,SACA,KACA,MACmB;AACnB,QAAM,SAAS,KAAK;AACpB,MAAI,SAAS,MAAM,GAAG;AACpB,WAAO,WAAW,OAAO,SAAS,KAAK,KAAK,UAAU,IAAI,GAAG,kBAAkB;AAAA,EACjF;AACA,QAAM,OAAO,KAAK;AAClB,MAAI,CAAC,SAAS,IAAI,KAAK,EAAE,KAAK,gBAAgB,eAAe,OAAO,KAAK,aAAa,UAAU;AAC9F,UAAM,IAAI,MAAM,oCAAoC;AAAA,EACtD;AACA,QAAM,QAAQ,UAAU,KAAK,QAAQ;AACrC,QAAM,SAAiC,CAAC;AACxC,QAAM,UAAU,YAAY,MAAM,OAAO;AACzC,SAAO,QAAQ,YAAY,SAAY,MAAM,KAAK;AAClD,QAAM,UAAU;AAAA,IACd;AAAA,MACE,MAAM,KAAK;AAAA,MACX,UAAU,KAAK;AAAA,MACf,UAAU,OAAO,KAAK,aAAa,WAAW,KAAK,WAAW,MAAM;AAAA,MACpE,QAAQ,MAAM;AAAA,IAChB;AAAA,IACA;AAAA,EACF;AACA,QAAM,OAAO,IAAI,YAAY,QAAQ,UAAU;AAC/C,MAAI,WAAW,IAAI,EAAE,IAAI,OAAO;AAChC,SAAO,WAAW,OAAO,SAAS,KAAK,MAAM,cAAc,CAAC;AAC9D;",
+  "names": []
+}

package/dist/openai/wire.js

@@ -0,0 +1,201 @@
+import { encodeAudioBase64 } from "../convert.js";
+import { VoiceRequestError } from "../errors.js";
+const boundary = "pi-audio-boundary";
+const mimeFiles = {
+  "audio/wav": { format: "wav", filename: "audio.wav" },
+  "audio/wave": { format: "wav", filename: "audio.wav" },
+  "audio/x-wav": { format: "wav", filename: "audio.wav" },
+  "audio/mpeg": { format: "mp3", filename: "audio.mp3" },
+  "audio/mp3": { format: "mp3", filename: "audio.mp3" },
+  "audio/flac": { format: "flac", filename: "audio.flac" },
+  "audio/ogg": { format: "ogg", filename: "audio.ogg" },
+  "audio/opus": { format: "ogg", filename: "audio.ogg" },
+  "audio/webm": { format: "webm", filename: "audio.webm" },
+  "audio/mp4": { format: "m4a", filename: "audio.m4a" },
+  "audio/m4a": { format: "m4a", filename: "audio.m4a" },
+  "audio/aac": { format: "aac", filename: "audio.aac" },
+  "audio/pcm": { format: "pcm", filename: "audio.pcm" }
+};
+function isOpenRouter(model) {
+  return model.provider === "openrouter" || model.baseUrl.includes("openrouter.ai");
+}
+function endpoint(baseUrl, path) {
+  return `${baseUrl.replace(/\/$/, "")}/${path}`;
+}
+function realtimeUrl(baseUrl) {
+  const url = new URL(baseUrl);
+  url.protocol = url.protocol === "https:" ? "wss:" : "ws:";
+  url.pathname = `${url.pathname.replace(/\/$/, "")}/realtime`;
+  url.search = "intent=transcription";
+  return url.toString();
+}
+function audioFile(mimeType) {
+  const found = mimeFiles[mimeType.toLowerCase()];
+  if (!found) throw new Error(`Unsupported audio type ${mimeType}`);
+  return found;
+}
+function isRecord(value) {
+  return typeof value === "object" && value !== null;
+}
+function stringField(record, key) {
+  const value = record[key];
+  return typeof value === "string" ? value : void 0;
+}
+function numberField(record, key) {
+  const value = record[key];
+  return typeof value === "number" ? value : void 0;
+}
+function asVoiceError(error) {
+  if (error instanceof VoiceRequestError) return { reason: error.reason, error: error.message };
+  if (error instanceof Error && error.name === "AbortError")
+    return { reason: "aborted", error: "aborted" };
+  if (error instanceof Error) return { reason: "error", error: error.message };
+  return { reason: "error", error: String(error) };
+}
+function headerValue(headers, name) {
+  if (!headers) return void 0;
+  const expected = name.toLowerCase();
+  for (const [key, value] of Object.entries(headers)) {
+    if (key.toLowerCase() === expected && value !== null && value.trim() !== "") return value;
+  }
+  return void 0;
+}
+function hasAuth(headers) {
+  return headerValue(headers, "authorization") !== void 0 || headerValue(headers, "cf-aig-authorization") !== void 0;
+}
+function requestHeaders(model, options) {
+  const headers = { ...model.headers };
+  const extra = options?.headers;
+  if (extra) {
+    for (const [key, value] of Object.entries(extra)) {
+      if (value === null) delete headers[key];
+      else headers[key] = value;
+    }
+  }
+  if (!hasAuth(headers)) {
+    const apiKey = options?.apiKey;
+    if (!apiKey) throw new Error(`No API key for provider: ${model.provider}`);
+    headers.Authorization = `Bearer ${apiKey}`;
+  }
+  return headers;
+}
+function requestSignal(options) {
+  const signals = [];
+  if (options?.signal) signals.push(options.signal);
+  const timeout = options?.timeoutMs;
+  if (timeout !== void 0 && timeout > 0) signals.push(AbortSignal.timeout(timeout));
+  if (signals.length === 0) return void 0;
+  if (signals.length === 1) return signals[0];
+  return AbortSignal.any(signals);
+}
+function encodeMultipart(file, fields) {
+  const chunks = [];
+  const encoder = new TextEncoder();
+  const write = (value) => chunks.push(encoder.encode(value));
+  write(
+    `--${boundary}\r
+Content-Disposition: form-data; name="file"; filename="${file.filename}"\r
+Content-Type: ${file.mimeType}\r
+\r
+`
+  );
+  chunks.push(file.data);
+  write("\r\n");
+  for (const [name, value] of Object.entries(fields)) {
+    write(`--${boundary}\r
+Content-Disposition: form-data; name="${name}"\r
+\r
+${value}\r
+`);
+  }
+  write(`--${boundary}--\r
+`);
+  const body = new Uint8Array(chunks.reduce((sum, chunk) => sum + chunk.length, 0));
+  let offset = 0;
+  for (const chunk of chunks) {
+    body.set(chunk, offset);
+    offset += chunk.length;
+  }
+  return body;
+}
+function multipartType() {
+  return `multipart/form-data; boundary=${boundary}`;
+}
+async function readAudio(input, fetchImpl, signal) {
+  if (!("url" in input)) return { data: input.data, mimeType: input.mimeType };
+  const response = await fetchImpl(input.url, { signal });
+  if (!response.ok) throw new Error(`Audio URL failed: HTTP ${response.status}`);
+  const mime = response.headers.get("content-type");
+  const mimeType = mime === null || mime.trim() === "" ? "audio/wav" : mime.split(";")[0].trim();
+  return { data: new Uint8Array(await response.arrayBuffer()), mimeType };
+}
+function audioPayload(file, openRouter, modelId) {
+  if (openRouter) {
+    return {
+      model: modelId,
+      input_audio: { data: encodeAudioBase64(file.data), format: file.format }
+    };
+  }
+  return { model: modelId, file };
+}
+async function applyPayload(payload, model, options) {
+  const hook = options?.onPayload;
+  const replaced = await hook?.(payload, model);
+  return replaced === void 0 ? payload : replaced;
+}
+function headerRecord(headers) {
+  const record = {};
+  headers.forEach((value, key) => {
+    record[key] = value;
+  });
+  return record;
+}
+async function voiceFetch(model, options, url, body, contentType) {
+  const headers = requestHeaders(model, options);
+  headers["Content-Type"] = contentType;
+  const fetchImpl = options?.fetch ?? globalThis.fetch;
+  const response = await fetchImpl(url, {
+    method: "POST",
+    headers,
+    body,
+    signal: requestSignal(options)
+  });
+  await options?.onResponse?.(
+    { status: response.status, headers: headerRecord(response.headers) },
+    model
+  );
+  if (!response.ok) {
+    const text = await response.text();
+    throw new Error(text === "" ? `HTTP ${response.status}` : text);
+  }
+  return response;
+}
+async function readJson(response) {
+  const text = await response.text();
+  try {
+    return JSON.parse(text);
+  } catch {
+    throw new Error("Transcription response was not JSON");
+  }
+}
+export {
+  applyPayload,
+  asVoiceError,
+  audioFile,
+  audioPayload,
+  encodeMultipart,
+  endpoint,
+  hasAuth,
+  isOpenRouter,
+  isRecord,
+  multipartType,
+  numberField,
+  readAudio,
+  readJson,
+  realtimeUrl,
+  requestHeaders,
+  requestSignal,
+  stringField,
+  voiceFetch
+};
+//# sourceMappingURL=wire.js.map

package/dist/openai/wire.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../src/openai/wire.ts"],
+  "sourcesContent": ["import type { ProviderRequestOptions } from '@earendil-works/pi-ai';\nimport { encodeAudioBase64 } from '../convert.ts';\nimport { VoiceRequestError } from '../errors.ts';\n\nconst boundary = 'pi-audio-boundary';\n\nconst mimeFiles: Record<string, { format: string; filename: string }> = {\n  'audio/wav': { format: 'wav', filename: 'audio.wav' },\n  'audio/wave': { format: 'wav', filename: 'audio.wav' },\n  'audio/x-wav': { format: 'wav', filename: 'audio.wav' },\n  'audio/mpeg': { format: 'mp3', filename: 'audio.mp3' },\n  'audio/mp3': { format: 'mp3', filename: 'audio.mp3' },\n  'audio/flac': { format: 'flac', filename: 'audio.flac' },\n  'audio/ogg': { format: 'ogg', filename: 'audio.ogg' },\n  'audio/opus': { format: 'ogg', filename: 'audio.ogg' },\n  'audio/webm': { format: 'webm', filename: 'audio.webm' },\n  'audio/mp4': { format: 'm4a', filename: 'audio.m4a' },\n  'audio/m4a': { format: 'm4a', filename: 'audio.m4a' },\n  'audio/aac': { format: 'aac', filename: 'audio.aac' },\n  'audio/pcm': { format: 'pcm', filename: 'audio.pcm' },\n};\n\nexport interface AudioFile {\n  data: Uint8Array;\n  mimeType: string;\n  filename: string;\n  format: string;\n}\n\nexport interface JsonBody {\n  [key: string]: unknown;\n}\n\ninterface HeaderModel {\n  provider: string;\n  headers?: Record<string, string>;\n}\n\nexport function isOpenRouter(model: { provider: string; baseUrl: string }): boolean {\n  return model.provider === 'openrouter' || model.baseUrl.includes('openrouter.ai');\n}\n\nexport function endpoint(baseUrl: string, path: string): string {\n  return `${baseUrl.replace(/\\/$/, '')}/${path}`;\n}\n\nexport function realtimeUrl(baseUrl: string): string {\n  const url = new URL(baseUrl);\n  url.protocol = url.protocol === 'https:' ? 'wss:' : 'ws:';\n  url.pathname = `${url.pathname.replace(/\\/$/, '')}/realtime`;\n  url.search = 'intent=transcription';\n  return url.toString();\n}\n\nexport function audioFile(mimeType: string): { format: string; filename: string } {\n  const found = mimeFiles[mimeType.toLowerCase()];\n  if (!found) throw new Error(`Unsupported audio type ${mimeType}`);\n  return found;\n}\n\nexport function isRecord(value: unknown): value is Record<string, unknown> {\n  return typeof value === 'object' && value !== null;\n}\n\nexport function stringField(record: Record<string, unknown>, key: string): string | undefined {\n  const value = record[key];\n  return typeof value === 'string' ? value : undefined;\n}\n\nexport function numberField(record: Record<string, unknown>, key: string): number | undefined {\n  const value = record[key];\n  return typeof value === 'number' ? value : undefined;\n}\n\nexport function asVoiceError(error: unknown): { reason: 'aborted' | 'error'; error: string } {\n  if (error instanceof VoiceRequestError) return { reason: error.reason, error: error.message };\n  if (error instanceof Error && error.name === 'AbortError')\n    return { reason: 'aborted', error: 'aborted' };\n  if (error instanceof Error) return { reason: 'error', error: error.message };\n  return { reason: 'error', error: String(error) };\n}\n\nfunction headerValue(\n  headers: Record<string, string | null> | undefined,\n  name: string\n): string | undefined {\n  if (!headers) return undefined;\n  const expected = name.toLowerCase();\n  for (const [key, value] of Object.entries(headers)) {\n    if (key.toLowerCase() === expected && value !== null && value.trim() !== '') return value;\n  }\n  return undefined;\n}\n\nexport function hasAuth(headers: Record<string, string | null> | undefined): boolean {\n  return (\n    headerValue(headers, 'authorization') !== undefined ||\n    headerValue(headers, 'cf-aig-authorization') !== undefined\n  );\n}\n\nexport function requestHeaders(\n  model: HeaderModel,\n  options: ProviderRequestOptions | undefined\n): Record<string, string> {\n  const headers: Record<string, string> = { ...model.headers };\n  const extra = options?.headers;\n  if (extra) {\n    for (const [key, value] of Object.entries(extra)) {\n      if (value === null) delete headers[key];\n      else headers[key] = value;\n    }\n  }\n  if (!hasAuth(headers)) {\n    const apiKey = options?.apiKey;\n    if (!apiKey) throw new Error(`No API key for provider: ${model.provider}`);\n    headers.Authorization = `Bearer ${apiKey}`;\n  }\n  return headers;\n}\n\nexport function requestSignal(\n  options: ProviderRequestOptions | undefined\n): AbortSignal | undefined {\n  const signals: AbortSignal[] = [];\n  if (options?.signal) signals.push(options.signal);\n  const timeout = options?.timeoutMs;\n  if (timeout !== undefined && timeout > 0) signals.push(AbortSignal.timeout(timeout));\n  if (signals.length === 0) return undefined;\n  if (signals.length === 1) return signals[0];\n  return AbortSignal.any(signals);\n}\n\nexport function encodeMultipart(file: AudioFile, fields: Record<string, string>): Uint8Array {\n  const chunks: Uint8Array[] = [];\n  const encoder = new TextEncoder();\n  const write = (value: string) => chunks.push(encoder.encode(value));\n  write(\n    `--${boundary}\\r\\nContent-Disposition: form-data; name=\"file\"; filename=\"${file.filename}\"\\r\\nContent-Type: ${file.mimeType}\\r\\n\\r\\n`\n  );\n  chunks.push(file.data);\n  write('\\r\\n');\n  for (const [name, value] of Object.entries(fields)) {\n    write(`--${boundary}\\r\\nContent-Disposition: form-data; name=\"${name}\"\\r\\n\\r\\n${value}\\r\\n`);\n  }\n  write(`--${boundary}--\\r\\n`);\n  const body = new Uint8Array(chunks.reduce((sum, chunk) => sum + chunk.length, 0));\n  let offset = 0;\n  for (const chunk of chunks) {\n    body.set(chunk, offset);\n    offset += chunk.length;\n  }\n  return body;\n}\n\nexport function multipartType(): string {\n  return `multipart/form-data; boundary=${boundary}`;\n}\n\nexport async function readAudio(\n  input: { data: Uint8Array; mimeType: string } | { url: string },\n  fetchImpl: typeof fetch,\n  signal: AbortSignal | undefined\n): Promise<{ data: Uint8Array; mimeType: string }> {\n  if (!('url' in input)) return { data: input.data, mimeType: input.mimeType };\n  const response = await fetchImpl(input.url, { signal });\n  if (!response.ok) throw new Error(`Audio URL failed: HTTP ${response.status}`);\n  const mime = response.headers.get('content-type');\n  const mimeType = mime === null || mime.trim() === '' ? 'audio/wav' : mime.split(';')[0].trim();\n  return { data: new Uint8Array(await response.arrayBuffer()), mimeType };\n}\n\nexport function audioPayload(file: AudioFile, openRouter: boolean, modelId: string): JsonBody {\n  if (openRouter) {\n    return {\n      model: modelId,\n      input_audio: { data: encodeAudioBase64(file.data), format: file.format },\n    };\n  }\n  return { model: modelId, file };\n}\n\nexport async function applyPayload(\n  payload: unknown,\n  model: unknown,\n  options: ProviderRequestOptions | undefined\n): Promise<unknown> {\n  const hook = options?.onPayload as\n    | ((body: unknown, voiceModel: unknown) => unknown | undefined | Promise<unknown | undefined>)\n    | undefined;\n  const replaced = await hook?.(payload, model);\n  return replaced === undefined ? payload : replaced;\n}\n\nfunction headerRecord(headers: Headers): Record<string, string> {\n  const record: Record<string, string> = {};\n  headers.forEach((value, key) => {\n    record[key] = value;\n  });\n  return record;\n}\n\nexport async function voiceFetch(\n  model: HeaderModel & { id: string },\n  options: ProviderRequestOptions | undefined,\n  url: string,\n  body: BodyInit,\n  contentType: string\n): Promise<Response> {\n  const headers = requestHeaders(model, options);\n  headers['Content-Type'] = contentType;\n  const fetchImpl = options?.fetch ?? globalThis.fetch;\n  const response = await fetchImpl(url, {\n    method: 'POST',\n    headers,\n    body,\n    signal: requestSignal(options),\n  });\n  await options?.onResponse?.(\n    { status: response.status, headers: headerRecord(response.headers) },\n    model as never\n  );\n  if (!response.ok) {\n    const text = await response.text();\n    throw new Error(text === '' ? `HTTP ${response.status}` : text);\n  }\n  return response;\n}\n\nexport async function readJson(response: Response): Promise<unknown> {\n  const text = await response.text();\n  try {\n    return JSON.parse(text);\n  } catch {\n    throw new Error('Transcription response was not JSON');\n  }\n}\n"],
+  "mappings": "AACA,SAAS,yBAAyB;AAClC,SAAS,yBAAyB;AAElC,MAAM,WAAW;AAEjB,MAAM,YAAkE;AAAA,EACtE,aAAa,EAAE,QAAQ,OAAO,UAAU,YAAY;AAAA,EACpD,cAAc,EAAE,QAAQ,OAAO,UAAU,YAAY;AAAA,EACrD,eAAe,EAAE,QAAQ,OAAO,UAAU,YAAY;AAAA,EACtD,cAAc,EAAE,QAAQ,OAAO,UAAU,YAAY;AAAA,EACrD,aAAa,EAAE,QAAQ,OAAO,UAAU,YAAY;AAAA,EACpD,cAAc,EAAE,QAAQ,QAAQ,UAAU,aAAa;AAAA,EACvD,aAAa,EAAE,QAAQ,OAAO,UAAU,YAAY;AAAA,EACpD,cAAc,EAAE,QAAQ,OAAO,UAAU,YAAY;AAAA,EACrD,cAAc,EAAE,QAAQ,QAAQ,UAAU,aAAa;AAAA,EACvD,aAAa,EAAE,QAAQ,OAAO,UAAU,YAAY;AAAA,EACpD,aAAa,EAAE,QAAQ,OAAO,UAAU,YAAY;AAAA,EACpD,aAAa,EAAE,QAAQ,OAAO,UAAU,YAAY;AAAA,EACpD,aAAa,EAAE,QAAQ,OAAO,UAAU,YAAY;AACtD;AAkBO,SAAS,aAAa,OAAuD;AAClF,SAAO,MAAM,aAAa,gBAAgB,MAAM,QAAQ,SAAS,eAAe;AAClF;AAEO,SAAS,SAAS,SAAiB,MAAsB;AAC9D,SAAO,GAAG,QAAQ,QAAQ,OAAO,EAAE,CAAC,IAAI,IAAI;AAC9C;AAEO,SAAS,YAAY,SAAyB;AACnD,QAAM,MAAM,IAAI,IAAI,OAAO;AAC3B,MAAI,WAAW,IAAI,aAAa,WAAW,SAAS;AACpD,MAAI,WAAW,GAAG,IAAI,SAAS,QAAQ,OAAO,EAAE,CAAC;AACjD,MAAI,SAAS;AACb,SAAO,IAAI,SAAS;AACtB;AAEO,SAAS,UAAU,UAAwD;AAChF,QAAM,QAAQ,UAAU,SAAS,YAAY,CAAC;AAC9C,MAAI,CAAC,MAAO,OAAM,IAAI,MAAM,0BAA0B,QAAQ,EAAE;AAChE,SAAO;AACT;AAEO,SAAS,SAAS,OAAkD;AACzE,SAAO,OAAO,UAAU,YAAY,UAAU;AAChD;AAEO,SAAS,YAAY,QAAiC,KAAiC;AAC5F,QAAM,QAAQ,OAAO,GAAG;AACxB,SAAO,OAAO,UAAU,WAAW,QAAQ;AAC7C;AAEO,SAAS,YAAY,QAAiC,KAAiC;AAC5F,QAAM,QAAQ,OAAO,GAAG;AACxB,SAAO,OAAO,UAAU,WAAW,QAAQ;AAC7C;AAEO,SAAS,aAAa,OAAgE;AAC3F,MAAI,iBAAiB,kBAAmB,QAAO,EAAE,QAAQ,MAAM,QAAQ,OAAO,MAAM,QAAQ;AAC5F,MAAI,iBAAiB,SAAS,MAAM,SAAS;AAC3C,WAAO,EAAE,QAAQ,WAAW,OAAO,UAAU;AAC/C,MAAI,iBAAiB,MAAO,QAAO,EAAE,QAAQ,SAAS,OAAO,MAAM,QAAQ;AAC3E,SAAO,EAAE,QAAQ,SAAS,OAAO,OAAO,KAAK,EAAE;AACjD;AAEA,SAAS,YACP,SACA,MACoB;AACpB,MAAI,CAAC,QAAS,QAAO;AACrB,QAAM,WAAW,KAAK,YAAY;AAClC,aAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,OAAO,GAAG;AAClD,QAAI,IAAI,YAAY,MAAM,YAAY,UAAU,QAAQ,MAAM,KAAK,MAAM,GAAI,QAAO;AAAA,EACtF;AACA,SAAO;AACT;AAEO,SAAS,QAAQ,SAA6D;AACnF,SACE,YAAY,SAAS,eAAe,MAAM,UAC1C,YAAY,SAAS,sBAAsB,MAAM;AAErD;AAEO,SAAS,eACd,OACA,SACwB;AACxB,QAAM,UAAkC,EAAE,GAAG,MAAM,QAAQ;AAC3D,QAAM,QAAQ,SAAS;AACvB,MAAI,OAAO;AACT,eAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,KAAK,GAAG;AAChD,UAAI,UAAU,KAAM,QAAO,QAAQ,GAAG;AAAA,UACjC,SAAQ,GAAG,IAAI;AAAA,IACtB;AAAA,EACF;AACA,MAAI,CAAC,QAAQ,OAAO,GAAG;AACrB,UAAM,SAAS,SAAS;AACxB,QAAI,CAAC,OAAQ,OAAM,IAAI,MAAM,4BAA4B,MAAM,QAAQ,EAAE;AACzE,YAAQ,gBAAgB,UAAU,MAAM;AAAA,EAC1C;AACA,SAAO;AACT;AAEO,SAAS,cACd,SACyB;AACzB,QAAM,UAAyB,CAAC;AAChC,MAAI,SAAS,OAAQ,SAAQ,KAAK,QAAQ,MAAM;AAChD,QAAM,UAAU,SAAS;AACzB,MAAI,YAAY,UAAa,UAAU,EAAG,SAAQ,KAAK,YAAY,QAAQ,OAAO,CAAC;AACnF,MAAI,QAAQ,WAAW,EAAG,QAAO;AACjC,MAAI,QAAQ,WAAW,EAAG,QAAO,QAAQ,CAAC;AAC1C,SAAO,YAAY,IAAI,OAAO;AAChC;AAEO,SAAS,gBAAgB,MAAiB,QAA4C;AAC3F,QAAM,SAAuB,CAAC;AAC9B,QAAM,UAAU,IAAI,YAAY;AAChC,QAAM,QAAQ,CAAC,UAAkB,OAAO,KAAK,QAAQ,OAAO,KAAK,CAAC;AAClE;AAAA,IACE,KAAK,QAAQ;AAAA,yDAA8D,KAAK,QAAQ;AAAA,gBAAsB,KAAK,QAAQ;AAAA;AAAA;AAAA,EAC7H;AACA,SAAO,KAAK,KAAK,IAAI;AACrB,QAAM,MAAM;AACZ,aAAW,CAAC,MAAM,KAAK,KAAK,OAAO,QAAQ,MAAM,GAAG;AAClD,UAAM,KAAK,QAAQ;AAAA,wCAA6C,IAAI;AAAA;AAAA,EAAY,KAAK;AAAA,CAAM;AAAA,EAC7F;AACA,QAAM,KAAK,QAAQ;AAAA,CAAQ;AAC3B,QAAM,OAAO,IAAI,WAAW,OAAO,OAAO,CAAC,KAAK,UAAU,MAAM,MAAM,QAAQ,CAAC,CAAC;AAChF,MAAI,SAAS;AACb,aAAW,SAAS,QAAQ;AAC1B,SAAK,IAAI,OAAO,MAAM;AACtB,cAAU,MAAM;AAAA,EAClB;AACA,SAAO;AACT;AAEO,SAAS,gBAAwB;AACtC,SAAO,iCAAiC,QAAQ;AAClD;AAEA,eAAsB,UACpB,OACA,WACA,QACiD;AACjD,MAAI,EAAE,SAAS,OAAQ,QAAO,EAAE,MAAM,MAAM,MAAM,UAAU,MAAM,SAAS;AAC3E,QAAM,WAAW,MAAM,UAAU,MAAM,KAAK,EAAE,OAAO,CAAC;AACtD,MAAI,CAAC,SAAS,GAAI,OAAM,IAAI,MAAM,0BAA0B,SAAS,MAAM,EAAE;AAC7E,QAAM,OAAO,SAAS,QAAQ,IAAI,cAAc;AAChD,QAAM,WAAW,SAAS,QAAQ,KAAK,KAAK,MAAM,KAAK,cAAc,KAAK,MAAM,GAAG,EAAE,CAAC,EAAE,KAAK;AAC7F,SAAO,EAAE,MAAM,IAAI,WAAW,MAAM,SAAS,YAAY,CAAC,GAAG,SAAS;AACxE;AAEO,SAAS,aAAa,MAAiB,YAAqB,SAA2B;AAC5F,MAAI,YAAY;AACd,WAAO;AAAA,MACL,OAAO;AAAA,MACP,aAAa,EAAE,MAAM,kBAAkB,KAAK,IAAI,GAAG,QAAQ,KAAK,OAAO;AAAA,IACzE;AAAA,EACF;AACA,SAAO,EAAE,OAAO,SAAS,KAAK;AAChC;AAEA,eAAsB,aACpB,SACA,OACA,SACkB;AAClB,QAAM,OAAO,SAAS;AAGtB,QAAM,WAAW,MAAM,OAAO,SAAS,KAAK;AAC5C,SAAO,aAAa,SAAY,UAAU;AAC5C;AAEA,SAAS,aAAa,SAA0C;AAC9D,QAAM,SAAiC,CAAC;AACxC,UAAQ,QAAQ,CAAC,OAAO,QAAQ;AAC9B,WAAO,GAAG,IAAI;AAAA,EAChB,CAAC;AACD,SAAO;AACT;AAEA,eAAsB,WACpB,OACA,SACA,KACA,MACA,aACmB;AACnB,QAAM,UAAU,eAAe,OAAO,OAAO;AAC7C,UAAQ,cAAc,IAAI;AAC1B,QAAM,YAAY,SAAS,SAAS,WAAW;AAC/C,QAAM,WAAW,MAAM,UAAU,KAAK;AAAA,IACpC,QAAQ;AAAA,IACR;AAAA,IACA;AAAA,IACA,QAAQ,cAAc,OAAO;AAAA,EAC/B,CAAC;AACD,QAAM,SAAS;AAAA,IACb,EAAE,QAAQ,SAAS,QAAQ,SAAS,aAAa,SAAS,OAAO,EAAE;AAAA,IACnE;AAAA,EACF;AACA,MAAI,CAAC,SAAS,IAAI;AAChB,UAAM,OAAO,MAAM,SAAS,KAAK;AACjC,UAAM,IAAI,MAAM,SAAS,KAAK,QAAQ,SAAS,MAAM,KAAK,IAAI;AAAA,EAChE;AACA,SAAO;AACT;AAEA,eAAsB,SAAS,UAAsC;AACnE,QAAM,OAAO,MAAM,SAAS,KAAK;AACjC,MAAI;AACF,WAAO,KAAK,MAAM,IAAI;AAAA,EACxB,QAAQ;AACN,UAAM,IAAI,MAAM,qCAAqC;AAAA,EACvD;AACF;",
+  "names": []
+}

package/dist/types/openai/catalog.d.ts

@@ -0,0 +1 @@
+export declare function registerOpenAI(): () => void;

package/dist/types/openai/index.d.ts

@@ -0,0 +1 @@
+export { registerOpenAI } from './catalog.ts';

package/dist/types/openai/listen.d.ts

@@ -0,0 +1,4 @@
+import type { AudioChunk } from '../audio.ts';
+import type { TranscriptionModel } from '../models.ts';
+import { type ListenOptions, TranscriptionEventStream } from '../transcription.ts';
+export declare function listenOpenAI(model: TranscriptionModel<string>, audio: AsyncIterable<AudioChunk>, options?: ListenOptions): TranscriptionEventStream;

package/dist/types/openai/speak.d.ts

@@ -0,0 +1,4 @@
+import type { ProviderRequestOptions } from '@earendil-works/pi-ai';
+import type { SpeechModel } from '../models.ts';
+import { SpeechEventStream, type SpeechInput } from '../speech.ts';
+export declare function speakOpenAI(model: SpeechModel<string>, input: SpeechInput, options?: ProviderRequestOptions): SpeechEventStream;

package/dist/types/openai/transcribe.d.ts

@@ -0,0 +1,4 @@
+import type { ProviderRequestOptions } from '@earendil-works/pi-ai';
+import type { TranscriptionModel } from '../models.ts';
+import type { AudioInput, Transcript } from '../transcription.ts';
+export declare function transcribeOpenAI(model: TranscriptionModel<string>, input: AudioInput, options?: ProviderRequestOptions): Promise<Transcript>;

package/dist/types/openai/wire.d.ts

@@ -0,0 +1,52 @@
+import type { ProviderRequestOptions } from '@earendil-works/pi-ai';
+export interface AudioFile {
+    data: Uint8Array;
+    mimeType: string;
+    filename: string;
+    format: string;
+}
+export interface JsonBody {
+    [key: string]: unknown;
+}
+interface HeaderModel {
+    provider: string;
+    headers?: Record<string, string>;
+}
+export declare function isOpenRouter(model: {
+    provider: string;
+    baseUrl: string;
+}): boolean;
+export declare function endpoint(baseUrl: string, path: string): string;
+export declare function realtimeUrl(baseUrl: string): string;
+export declare function audioFile(mimeType: string): {
+    format: string;
+    filename: string;
+};
+export declare function isRecord(value: unknown): value is Record<string, unknown>;
+export declare function stringField(record: Record<string, unknown>, key: string): string | undefined;
+export declare function numberField(record: Record<string, unknown>, key: string): number | undefined;
+export declare function asVoiceError(error: unknown): {
+    reason: 'aborted' | 'error';
+    error: string;
+};
+export declare function hasAuth(headers: Record<string, string | null> | undefined): boolean;
+export declare function requestHeaders(model: HeaderModel, options: ProviderRequestOptions | undefined): Record<string, string>;
+export declare function requestSignal(options: ProviderRequestOptions | undefined): AbortSignal | undefined;
+export declare function encodeMultipart(file: AudioFile, fields: Record<string, string>): Uint8Array;
+export declare function multipartType(): string;
+export declare function readAudio(input: {
+    data: Uint8Array;
+    mimeType: string;
+} | {
+    url: string;
+}, fetchImpl: typeof fetch, signal: AbortSignal | undefined): Promise<{
+    data: Uint8Array;
+    mimeType: string;
+}>;
+export declare function audioPayload(file: AudioFile, openRouter: boolean, modelId: string): JsonBody;
+export declare function applyPayload(payload: unknown, model: unknown, options: ProviderRequestOptions | undefined): Promise<unknown>;
+export declare function voiceFetch(model: HeaderModel & {
+    id: string;
+}, options: ProviderRequestOptions | undefined, url: string, body: BodyInit, contentType: string): Promise<Response>;
+export declare function readJson(response: Response): Promise<unknown>;
+export {};

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ai-ecoverse/pi-audio",
-  "version": "1.4.0",
+  "version": "1.5.0",
   "description": "Voice bindings for pi: speech-to-text, text-to-speech and realtime speech for pi-coding-agent plugins and slicc-agent",
   "type": "module",
   "license": "Apache-2.0",
@@ -33,6 +33,10 @@
     "./worklet": {
       "types": "./dist/types/worklet/convert.d.ts",
       "default": "./dist/worklet/convert.js"
+    },
+    "./openai": {
+      "types": "./dist/types/openai/index.d.ts",
+      "default": "./dist/openai/index.js"
     }
   },
   "files": [