@ai-ecoverse/pi-audio 0.0.0 → 1.1.0

package/dist/faux/speech.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../src/faux/speech.ts"],
+  "sourcesContent": ["import type { ProviderRequestOptions } from '@earendil-works/pi-ai';\nimport type { AudioFormat, CharAlignment } from '../audio.ts';\nimport { blankUsage } from '../blank-usage.ts';\nimport { VoiceRequestError } from '../errors.ts';\nimport { registerVoiceModel, type SpeechModel } from '../models.ts';\nimport { registerSpeechApiProvider } from '../registry.ts';\nimport { SpeechEventStream, type SpeechInput } from '../speech.ts';\nimport { encodeChars, errorMessage, modelKey, pcm, shareApi, zeroCost } from './shared.ts';\n\nexport interface FauxSpeechOptions {\n  api?: string;\n  provider?: string;\n  id?: string;\n  name?: string;\n  alignment?: boolean;\n  chunkChars?: number;\n  textStreaming?: boolean;\n}\n\nexport interface FauxSpeech {\n  model: SpeechModel<string>;\n  calls: SpeechInput[];\n  setAlignment(alignment: boolean): void;\n  unregister(): void;\n}\n\ninterface SpeechState {\n  model: SpeechModel<string>;\n  alignment: boolean;\n  chunkChars: number;\n  calls: SpeechInput[];\n}\n\nconst states = new Map<string, SpeechState>();\nconst apis = new Map<string, () => void>();\n\nexport function registerFauxSpeech(options: FauxSpeechOptions = {}): FauxSpeech {\n  const model: SpeechModel<string> = {\n    id: options.id ?? 'faux-speech',\n    name: options.name ?? 'Faux speech',\n    api: options.api ?? 'faux-speech',\n    provider: options.provider ?? 'faux',\n    baseUrl: 'http://localhost:0',\n    type: 'speech',\n    textStreaming: options.textStreaming ?? true,\n    outputFormats: [pcm()],\n    voices: [{ id: 'faux', name: 'Faux', languages: ['en'] }],\n    cloning: false,\n    cost: zeroCost,\n  };\n  const state: SpeechState = {\n    model,\n    alignment: options.alignment ?? false,\n    chunkChars: Math.max(1, options.chunkChars ?? 4),\n    calls: [],\n  };\n  states.set(modelKey(model), state);\n  const unregisterModel = registerVoiceModel(model);\n  const release = shareApi(\n    apis,\n    model.api,\n    () => registerSpeechApiProvider({ api: model.api, speak: speakFaux }),\n    () => [...states.values()].some((entry) => entry.model.api === model.api)\n  );\n  return {\n    model,\n    calls: state.calls,\n    setAlignment(alignment) {\n      state.alignment = alignment;\n    },\n    unregister() {\n      if (states.get(modelKey(model)) === state) states.delete(modelKey(model));\n      unregisterModel();\n      release();\n    },\n  };\n}\n\nfunction speakFaux(\n  model: SpeechModel<string>,\n  input: SpeechInput,\n  options?: ProviderRequestOptions\n): SpeechEventStream {\n  const stream = new SpeechEventStream();\n  const state = states.get(modelKey(model));\n  if (!state) {\n    stream.push({\n      type: 'error',\n      reason: 'error',\n      error: `Unknown faux speech model ${model.provider}/${model.id}`,\n    });\n    return stream;\n  }\n  state.calls.push(input);\n  void pumpSpeech(state, input, options, stream);\n  return stream;\n}\n\nasync function readText(input: SpeechInput, signal: AbortSignal | undefined): Promise<string> {\n  if (typeof input.text === 'string') return input.text;\n  let text = '';\n  for await (const part of input.text) {\n    if (signal?.aborted) throw new VoiceRequestError('aborted', 'aborted');\n    text += part;\n  }\n  return text;\n}\n\nfunction alignmentFor(start: number, length: number): CharAlignment {\n  const startMs = [];\n  const durationMs = [];\n  for (let offset = 0; offset < length; offset++) {\n    startMs.push((start + offset) * 20);\n    durationMs.push(20);\n  }\n  return { startMs, durationMs };\n}\n\nasync function pumpSpeech(\n  state: SpeechState,\n  input: SpeechInput,\n  options: ProviderRequestOptions | undefined,\n  stream: SpeechEventStream\n) {\n  try {\n    if (options?.signal?.aborted) throw new VoiceRequestError('aborted', 'aborted');\n    const text = await readText(input, options?.signal);\n    const format: AudioFormat | undefined = input.format ?? state.model.outputFormats[0];\n    if (!format) throw new Error('No output format');\n    stream.push({ type: 'start', format });\n    for (let start = 0; start < text.length; start += state.chunkChars) {\n      const slice = text.slice(start, start + state.chunkChars);\n      stream.push({\n        type: 'audio',\n        chunk: { data: encodeChars(slice), format, timestamp: start * 20 },\n        span: { itemId: state.model.id, start, end: start + slice.length },\n        ...(state.alignment ? { alignment: alignmentFor(start, slice.length) } : {}),\n      });\n    }\n    stream.push({ type: 'done', usage: blankUsage() });\n  } catch (error) {\n    const reason = error instanceof VoiceRequestError ? error.reason : 'error';\n    stream.push({ type: 'error', reason, error: errorMessage(error) });\n  }\n}\n"],
+  "mappings": "AAEA,SAAS,kBAAkB;AAC3B,SAAS,yBAAyB;AAClC,SAAS,0BAA4C;AACrD,SAAS,iCAAiC;AAC1C,SAAS,yBAA2C;AACpD,SAAS,aAAa,cAAc,UAAU,KAAK,UAAU,gBAAgB;AA0B7E,MAAM,SAAS,oBAAI,IAAyB;AAC5C,MAAM,OAAO,oBAAI,IAAwB;AAElC,SAAS,mBAAmB,UAA6B,CAAC,GAAe;AAC9E,QAAM,QAA6B;AAAA,IACjC,IAAI,QAAQ,MAAM;AAAA,IAClB,MAAM,QAAQ,QAAQ;AAAA,IACtB,KAAK,QAAQ,OAAO;AAAA,IACpB,UAAU,QAAQ,YAAY;AAAA,IAC9B,SAAS;AAAA,IACT,MAAM;AAAA,IACN,eAAe,QAAQ,iBAAiB;AAAA,IACxC,eAAe,CAAC,IAAI,CAAC;AAAA,IACrB,QAAQ,CAAC,EAAE,IAAI,QAAQ,MAAM,QAAQ,WAAW,CAAC,IAAI,EAAE,CAAC;AAAA,IACxD,SAAS;AAAA,IACT,MAAM;AAAA,EACR;AACA,QAAM,QAAqB;AAAA,IACzB;AAAA,IACA,WAAW,QAAQ,aAAa;AAAA,IAChC,YAAY,KAAK,IAAI,GAAG,QAAQ,cAAc,CAAC;AAAA,IAC/C,OAAO,CAAC;AAAA,EACV;AACA,SAAO,IAAI,SAAS,KAAK,GAAG,KAAK;AACjC,QAAM,kBAAkB,mBAAmB,KAAK;AAChD,QAAM,UAAU;AAAA,IACd;AAAA,IACA,MAAM;AAAA,IACN,MAAM,0BAA0B,EAAE,KAAK,MAAM,KAAK,OAAO,UAAU,CAAC;AAAA,IACpE,MAAM,CAAC,GAAG,OAAO,OAAO,CAAC,EAAE,KAAK,CAAC,UAAU,MAAM,MAAM,QAAQ,MAAM,GAAG;AAAA,EAC1E;AACA,SAAO;AAAA,IACL;AAAA,IACA,OAAO,MAAM;AAAA,IACb,aAAa,WAAW;AACtB,YAAM,YAAY;AAAA,IACpB;AAAA,IACA,aAAa;AACX,UAAI,OAAO,IAAI,SAAS,KAAK,CAAC,MAAM,MAAO,QAAO,OAAO,SAAS,KAAK,CAAC;AACxE,sBAAgB;AAChB,cAAQ;AAAA,IACV;AAAA,EACF;AACF;AAEA,SAAS,UACP,OACA,OACA,SACmB;AACnB,QAAM,SAAS,IAAI,kBAAkB;AACrC,QAAM,QAAQ,OAAO,IAAI,SAAS,KAAK,CAAC;AACxC,MAAI,CAAC,OAAO;AACV,WAAO,KAAK;AAAA,MACV,MAAM;AAAA,MACN,QAAQ;AAAA,MACR,OAAO,6BAA6B,MAAM,QAAQ,IAAI,MAAM,EAAE;AAAA,IAChE,CAAC;AACD,WAAO;AAAA,EACT;AACA,QAAM,MAAM,KAAK,KAAK;AACtB,OAAK,WAAW,OAAO,OAAO,SAAS,MAAM;AAC7C,SAAO;AACT;AAEA,eAAe,SAAS,OAAoB,QAAkD;AAC5F,MAAI,OAAO,MAAM,SAAS,SAAU,QAAO,MAAM;AACjD,MAAI,OAAO;AACX,mBAAiB,QAAQ,MAAM,MAAM;AACnC,QAAI,QAAQ,QAAS,OAAM,IAAI,kBAAkB,WAAW,SAAS;AACrE,YAAQ;AAAA,EACV;AACA,SAAO;AACT;AAEA,SAAS,aAAa,OAAe,QAA+B;AAClE,QAAM,UAAU,CAAC;AACjB,QAAM,aAAa,CAAC;AACpB,WAAS,SAAS,GAAG,SAAS,QAAQ,UAAU;AAC9C,YAAQ,MAAM,QAAQ,UAAU,EAAE;AAClC,eAAW,KAAK,EAAE;AAAA,EACpB;AACA,SAAO,EAAE,SAAS,WAAW;AAC/B;AAEA,eAAe,WACb,OACA,OACA,SACA,QACA;AACA,MAAI;AACF,QAAI,SAAS,QAAQ,QAAS,OAAM,IAAI,kBAAkB,WAAW,SAAS;AAC9E,UAAM,OAAO,MAAM,SAAS,OAAO,SAAS,MAAM;AAClD,UAAM,SAAkC,MAAM,UAAU,MAAM,MAAM,cAAc,CAAC;AACnF,QAAI,CAAC,OAAQ,OAAM,IAAI,MAAM,kBAAkB;AAC/C,WAAO,KAAK,EAAE,MAAM,SAAS,OAAO,CAAC;AACrC,aAAS,QAAQ,GAAG,QAAQ,KAAK,QAAQ,SAAS,MAAM,YAAY;AAClE,YAAM,QAAQ,KAAK,MAAM,OAAO,QAAQ,MAAM,UAAU;AACxD,aAAO,KAAK;AAAA,QACV,MAAM;AAAA,QACN,OAAO,EAAE,MAAM,YAAY,KAAK,GAAG,QAAQ,WAAW,QAAQ,GAAG;AAAA,QACjE,MAAM,EAAE,QAAQ,MAAM,MAAM,IAAI,OAAO,KAAK,QAAQ,MAAM,OAAO;AAAA,QACjE,GAAI,MAAM,YAAY,EAAE,WAAW,aAAa,OAAO,MAAM,MAAM,EAAE,IAAI,CAAC;AAAA,MAC5E,CAAC;AAAA,IACH;AACA,WAAO,KAAK,EAAE,MAAM,QAAQ,OAAO,WAAW,EAAE,CAAC;AAAA,EACnD,SAAS,OAAO;AACd,UAAM,SAAS,iBAAiB,oBAAoB,MAAM,SAAS;AACnE,WAAO,KAAK,EAAE,MAAM,SAAS,QAAQ,OAAO,aAAa,KAAK,EAAE,CAAC;AAAA,EACnE;AACF;",
+  "names": []
+}

package/dist/faux/transcription.js

@@ -0,0 +1,152 @@
+import { blankUsage } from "../blank-usage.js";
+import { VoiceRequestError } from "../errors.js";
+import { registerVoiceModel } from "../models.js";
+import { registerTranscriptionApiProvider } from "../registry.js";
+import {
+  TranscriptionEventStream
+} from "../transcription.js";
+import { decodeBytes, errorMessage, modelKey, pcm, shareApi, zeroCost } from "./shared.js";
+const states = /* @__PURE__ */ new Map();
+const apis = /* @__PURE__ */ new Map();
+function requireState(model) {
+  const state = states.get(modelKey(model));
+  if (!state) throw new Error(`Unknown faux transcription model ${model.provider}/${model.id}`);
+  return state;
+}
+function registerFauxTranscription(options = {}) {
+  const model = {
+    id: options.id ?? "faux-transcription",
+    name: options.name ?? "Faux transcription",
+    api: options.api ?? "faux-transcription",
+    provider: options.provider ?? "faux",
+    baseUrl: "http://localhost:0",
+    type: "transcription",
+    streaming: true,
+    turnDetection: options.turnDetection ?? "vad",
+    inputFormats: [pcm(16e3)],
+    languages: ["en"],
+    diarization: false,
+    cost: zeroCost
+  };
+  const state = { model, cues: options.cues, calls: [] };
+  states.set(modelKey(model), state);
+  const unregisterModel = registerVoiceModel(model);
+  const release = shareApi(
+    apis,
+    model.api,
+    () => registerTranscriptionApiProvider({
+      api: model.api,
+      transcribe: transcribeFaux,
+      listen: listenFaux
+    }),
+    () => [...states.values()].some((entry) => entry.model.api === model.api)
+  );
+  return {
+    model,
+    calls: state.calls,
+    setCues(cues) {
+      state.cues = cues;
+    },
+    unregister() {
+      if (states.get(modelKey(model)) === state) states.delete(modelKey(model));
+      unregisterModel();
+      release();
+    }
+  };
+}
+async function transcribeFaux(model, input, options) {
+  if (options?.signal?.aborted) throw new VoiceRequestError("aborted", "aborted");
+  const state = requireState(model);
+  state.calls.push({ kind: "transcribe", input, chunks: [] });
+  const text = "url" in input ? input.url : decodeBytes(input.data);
+  return { text, usage: blankUsage() };
+}
+function listenFaux(model, audio, options) {
+  const stream = new TranscriptionEventStream();
+  let state;
+  try {
+    state = requireState(model);
+  } catch (error) {
+    stream.push({ type: "error", reason: "error", error: errorMessage(error) });
+    return stream;
+  }
+  void pumpListen(state, audio, options, stream);
+  return stream;
+}
+function createPump(stream) {
+  const pump = {
+    stream,
+    finished: false,
+    push(event) {
+      if (pump.finished) return;
+      if (event.type === "turn_end") pump.turnText = event.text;
+      stream.push(event);
+      if (event.type === "done" || event.type === "error") pump.finished = true;
+    }
+  };
+  return pump;
+}
+function cuesAt(cues, at) {
+  return cues.filter((cue) => cue.at === at).flatMap((cue) => cue.events);
+}
+function endCues(cues, count) {
+  return cues.filter((cue) => cue.at === "end" || typeof cue.at === "number" && cue.at > count).flatMap((cue) => cue.events);
+}
+function finishScript(pump, cues, text, at, count) {
+  if (pump.finished) return;
+  if (cues) for (const event of endCues(cues, count)) pump.push(event);
+  else {
+    pump.push({ type: "final", segmentId: "0", text });
+    pump.push({ type: "turn_end", text, at });
+  }
+  if (!pump.finished) {
+    pump.push({
+      type: "done",
+      transcript: { text: pump.turnText ?? text, usage: blankUsage() }
+    });
+  }
+}
+async function pumpListen(state, audio, options, stream) {
+  const pump = createPump(stream);
+  const call = { kind: "listen", chunks: [] };
+  state.calls.push(call);
+  let text = "";
+  let at = 0;
+  let count = 0;
+  try {
+    if (options?.signal?.aborted) {
+      pump.push({ type: "error", reason: "aborted", error: "aborted" });
+      return;
+    }
+    for (const event of cuesAt(state.cues ?? [], 0)) pump.push(event);
+    if (!pump.finished) {
+      for await (const chunk of audio) {
+        if (pump.finished || options?.signal?.aborted) break;
+        call.chunks.push(chunk);
+        count += 1;
+        at = chunk.timestamp;
+        text += decodeBytes(chunk.data);
+        deliverChunk(pump, state.cues, chunk, count, text);
+      }
+    }
+    if (options?.signal?.aborted && !pump.finished) {
+      pump.push({ type: "error", reason: "aborted", error: "aborted" });
+      return;
+    }
+    finishScript(pump, state.cues, text, at, count);
+  } catch (error) {
+    pump.push({ type: "error", reason: "error", error: errorMessage(error) });
+  }
+}
+function deliverChunk(pump, cues, chunk, count, text) {
+  if (cues) {
+    for (const event of cuesAt(cues, count)) pump.push(event);
+    return;
+  }
+  if (count === 1) pump.push({ type: "speech_start", at: chunk.timestamp });
+  pump.push({ type: "partial", segmentId: "0", text });
+}
+export {
+  registerFauxTranscription
+};
+//# sourceMappingURL=transcription.js.map

package/dist/faux/transcription.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../src/faux/transcription.ts"],
+  "sourcesContent": ["import type { ProviderRequestOptions } from '@earendil-works/pi-ai';\nimport type { AudioChunk, AudioSource } from '../audio.ts';\nimport { blankUsage } from '../blank-usage.ts';\nimport { VoiceRequestError } from '../errors.ts';\nimport { registerVoiceModel, type TranscriptionModel } from '../models.ts';\nimport { registerTranscriptionApiProvider } from '../registry.ts';\nimport {\n  type AudioInput,\n  type ListenOptions,\n  type TranscriptionEvent,\n  TranscriptionEventStream,\n} from '../transcription.ts';\nimport { decodeBytes, errorMessage, modelKey, pcm, shareApi, zeroCost } from './shared.ts';\n\nexport interface FauxTranscriptionCue {\n  at: number | 'end';\n  events: TranscriptionEvent[];\n}\n\nexport interface FauxTranscriptionOptions {\n  api?: string;\n  provider?: string;\n  id?: string;\n  name?: string;\n  turnDetection?: TranscriptionModel<string>['turnDetection'];\n  cues?: FauxTranscriptionCue[];\n}\n\nexport interface FauxTranscriptionCall {\n  kind: 'transcribe' | 'listen';\n  input?: AudioInput;\n  chunks: AudioChunk[];\n}\n\nexport interface FauxTranscription {\n  model: TranscriptionModel<string>;\n  calls: FauxTranscriptionCall[];\n  setCues(cues: FauxTranscriptionCue[] | undefined): void;\n  unregister(): void;\n}\n\ninterface TranscriptionState {\n  model: TranscriptionModel<string>;\n  cues?: FauxTranscriptionCue[];\n  calls: FauxTranscriptionCall[];\n}\n\nconst states = new Map<string, TranscriptionState>();\nconst apis = new Map<string, () => void>();\n\nfunction requireState(model: TranscriptionModel<string>): TranscriptionState {\n  const state = states.get(modelKey(model));\n  if (!state) throw new Error(`Unknown faux transcription model ${model.provider}/${model.id}`);\n  return state;\n}\n\nexport function registerFauxTranscription(\n  options: FauxTranscriptionOptions = {}\n): FauxTranscription {\n  const model: TranscriptionModel<string> = {\n    id: options.id ?? 'faux-transcription',\n    name: options.name ?? 'Faux transcription',\n    api: options.api ?? 'faux-transcription',\n    provider: options.provider ?? 'faux',\n    baseUrl: 'http://localhost:0',\n    type: 'transcription',\n    streaming: true,\n    turnDetection: options.turnDetection ?? 'vad',\n    inputFormats: [pcm(16000)],\n    languages: ['en'],\n    diarization: false,\n    cost: zeroCost,\n  };\n  const state: TranscriptionState = { model, cues: options.cues, calls: [] };\n  states.set(modelKey(model), state);\n  const unregisterModel = registerVoiceModel(model);\n  const release = shareApi(\n    apis,\n    model.api,\n    () =>\n      registerTranscriptionApiProvider({\n        api: model.api,\n        transcribe: transcribeFaux,\n        listen: listenFaux,\n      }),\n    () => [...states.values()].some((entry) => entry.model.api === model.api)\n  );\n  return {\n    model,\n    calls: state.calls,\n    setCues(cues) {\n      state.cues = cues;\n    },\n    unregister() {\n      if (states.get(modelKey(model)) === state) states.delete(modelKey(model));\n      unregisterModel();\n      release();\n    },\n  };\n}\n\nasync function transcribeFaux(\n  model: TranscriptionModel<string>,\n  input: AudioInput,\n  options?: ProviderRequestOptions\n) {\n  if (options?.signal?.aborted) throw new VoiceRequestError('aborted', 'aborted');\n  const state = requireState(model);\n  state.calls.push({ kind: 'transcribe', input, chunks: [] });\n  const text = 'url' in input ? input.url : decodeBytes(input.data);\n  return { text, usage: blankUsage() };\n}\n\nfunction listenFaux(\n  model: TranscriptionModel<string>,\n  audio: AudioSource,\n  options?: ListenOptions\n): TranscriptionEventStream {\n  const stream = new TranscriptionEventStream();\n  let state: TranscriptionState;\n  try {\n    state = requireState(model);\n  } catch (error) {\n    stream.push({ type: 'error', reason: 'error', error: errorMessage(error) });\n    return stream;\n  }\n  void pumpListen(state, audio, options, stream);\n  return stream;\n}\n\ninterface Pump {\n  stream: TranscriptionEventStream;\n  finished: boolean;\n  turnText?: string;\n  push(event: TranscriptionEvent): void;\n}\n\nfunction createPump(stream: TranscriptionEventStream): Pump {\n  const pump: Pump = {\n    stream,\n    finished: false,\n    push(event) {\n      if (pump.finished) return;\n      if (event.type === 'turn_end') pump.turnText = event.text;\n      stream.push(event);\n      if (event.type === 'done' || event.type === 'error') pump.finished = true;\n    },\n  };\n  return pump;\n}\n\nfunction cuesAt(cues: FauxTranscriptionCue[], at: number | 'end'): TranscriptionEvent[] {\n  return cues.filter((cue) => cue.at === at).flatMap((cue) => cue.events);\n}\n\nfunction endCues(cues: FauxTranscriptionCue[], count: number): TranscriptionEvent[] {\n  return cues\n    .filter((cue) => cue.at === 'end' || (typeof cue.at === 'number' && cue.at > count))\n    .flatMap((cue) => cue.events);\n}\n\nfunction finishScript(\n  pump: Pump,\n  cues: FauxTranscriptionCue[] | undefined,\n  text: string,\n  at: number,\n  count: number\n) {\n  if (pump.finished) return;\n  if (cues) for (const event of endCues(cues, count)) pump.push(event);\n  else {\n    pump.push({ type: 'final', segmentId: '0', text });\n    pump.push({ type: 'turn_end', text, at });\n  }\n  if (!pump.finished) {\n    pump.push({\n      type: 'done',\n      transcript: { text: pump.turnText ?? text, usage: blankUsage() },\n    });\n  }\n}\n\nasync function pumpListen(\n  state: TranscriptionState,\n  audio: AudioSource,\n  options: ListenOptions | undefined,\n  stream: TranscriptionEventStream\n) {\n  const pump = createPump(stream);\n  const call: FauxTranscriptionCall = { kind: 'listen', chunks: [] };\n  state.calls.push(call);\n  let text = '';\n  let at = 0;\n  let count = 0;\n  try {\n    if (options?.signal?.aborted) {\n      pump.push({ type: 'error', reason: 'aborted', error: 'aborted' });\n      return;\n    }\n    for (const event of cuesAt(state.cues ?? [], 0)) pump.push(event);\n    if (!pump.finished) {\n      for await (const chunk of audio) {\n        if (pump.finished || options?.signal?.aborted) break;\n        call.chunks.push(chunk);\n        count += 1;\n        at = chunk.timestamp;\n        text += decodeBytes(chunk.data);\n        deliverChunk(pump, state.cues, chunk, count, text);\n      }\n    }\n    if (options?.signal?.aborted && !pump.finished) {\n      pump.push({ type: 'error', reason: 'aborted', error: 'aborted' });\n      return;\n    }\n    finishScript(pump, state.cues, text, at, count);\n  } catch (error) {\n    pump.push({ type: 'error', reason: 'error', error: errorMessage(error) });\n  }\n}\n\nfunction deliverChunk(\n  pump: Pump,\n  cues: FauxTranscriptionCue[] | undefined,\n  chunk: AudioChunk,\n  count: number,\n  text: string\n) {\n  if (cues) {\n    for (const event of cuesAt(cues, count)) pump.push(event);\n    return;\n  }\n  if (count === 1) pump.push({ type: 'speech_start', at: chunk.timestamp });\n  pump.push({ type: 'partial', segmentId: '0', text });\n}\n"],
+  "mappings": "AAEA,SAAS,kBAAkB;AAC3B,SAAS,yBAAyB;AAClC,SAAS,0BAAmD;AAC5D,SAAS,wCAAwC;AACjD;AAAA,EAIE;AAAA,OACK;AACP,SAAS,aAAa,cAAc,UAAU,KAAK,UAAU,gBAAgB;AAmC7E,MAAM,SAAS,oBAAI,IAAgC;AACnD,MAAM,OAAO,oBAAI,IAAwB;AAEzC,SAAS,aAAa,OAAuD;AAC3E,QAAM,QAAQ,OAAO,IAAI,SAAS,KAAK,CAAC;AACxC,MAAI,CAAC,MAAO,OAAM,IAAI,MAAM,oCAAoC,MAAM,QAAQ,IAAI,MAAM,EAAE,EAAE;AAC5F,SAAO;AACT;AAEO,SAAS,0BACd,UAAoC,CAAC,GAClB;AACnB,QAAM,QAAoC;AAAA,IACxC,IAAI,QAAQ,MAAM;AAAA,IAClB,MAAM,QAAQ,QAAQ;AAAA,IACtB,KAAK,QAAQ,OAAO;AAAA,IACpB,UAAU,QAAQ,YAAY;AAAA,IAC9B,SAAS;AAAA,IACT,MAAM;AAAA,IACN,WAAW;AAAA,IACX,eAAe,QAAQ,iBAAiB;AAAA,IACxC,cAAc,CAAC,IAAI,IAAK,CAAC;AAAA,IACzB,WAAW,CAAC,IAAI;AAAA,IAChB,aAAa;AAAA,IACb,MAAM;AAAA,EACR;AACA,QAAM,QAA4B,EAAE,OAAO,MAAM,QAAQ,MAAM,OAAO,CAAC,EAAE;AACzE,SAAO,IAAI,SAAS,KAAK,GAAG,KAAK;AACjC,QAAM,kBAAkB,mBAAmB,KAAK;AAChD,QAAM,UAAU;AAAA,IACd;AAAA,IACA,MAAM;AAAA,IACN,MACE,iCAAiC;AAAA,MAC/B,KAAK,MAAM;AAAA,MACX,YAAY;AAAA,MACZ,QAAQ;AAAA,IACV,CAAC;AAAA,IACH,MAAM,CAAC,GAAG,OAAO,OAAO,CAAC,EAAE,KAAK,CAAC,UAAU,MAAM,MAAM,QAAQ,MAAM,GAAG;AAAA,EAC1E;AACA,SAAO;AAAA,IACL;AAAA,IACA,OAAO,MAAM;AAAA,IACb,QAAQ,MAAM;AACZ,YAAM,OAAO;AAAA,IACf;AAAA,IACA,aAAa;AACX,UAAI,OAAO,IAAI,SAAS,KAAK,CAAC,MAAM,MAAO,QAAO,OAAO,SAAS,KAAK,CAAC;AACxE,sBAAgB;AAChB,cAAQ;AAAA,IACV;AAAA,EACF;AACF;AAEA,eAAe,eACb,OACA,OACA,SACA;AACA,MAAI,SAAS,QAAQ,QAAS,OAAM,IAAI,kBAAkB,WAAW,SAAS;AAC9E,QAAM,QAAQ,aAAa,KAAK;AAChC,QAAM,MAAM,KAAK,EAAE,MAAM,cAAc,OAAO,QAAQ,CAAC,EAAE,CAAC;AAC1D,QAAM,OAAO,SAAS,QAAQ,MAAM,MAAM,YAAY,MAAM,IAAI;AAChE,SAAO,EAAE,MAAM,OAAO,WAAW,EAAE;AACrC;AAEA,SAAS,WACP,OACA,OACA,SAC0B;AAC1B,QAAM,SAAS,IAAI,yBAAyB;AAC5C,MAAI;AACJ,MAAI;AACF,YAAQ,aAAa,KAAK;AAAA,EAC5B,SAAS,OAAO;AACd,WAAO,KAAK,EAAE,MAAM,SAAS,QAAQ,SAAS,OAAO,aAAa,KAAK,EAAE,CAAC;AAC1E,WAAO;AAAA,EACT;AACA,OAAK,WAAW,OAAO,OAAO,SAAS,MAAM;AAC7C,SAAO;AACT;AASA,SAAS,WAAW,QAAwC;AAC1D,QAAM,OAAa;AAAA,IACjB;AAAA,IACA,UAAU;AAAA,IACV,KAAK,OAAO;AACV,UAAI,KAAK,SAAU;AACnB,UAAI,MAAM,SAAS,WAAY,MAAK,WAAW,MAAM;AACrD,aAAO,KAAK,KAAK;AACjB,UAAI,MAAM,SAAS,UAAU,MAAM,SAAS,QAAS,MAAK,WAAW;AAAA,IACvE;AAAA,EACF;AACA,SAAO;AACT;AAEA,SAAS,OAAO,MAA8B,IAA0C;AACtF,SAAO,KAAK,OAAO,CAAC,QAAQ,IAAI,OAAO,EAAE,EAAE,QAAQ,CAAC,QAAQ,IAAI,MAAM;AACxE;AAEA,SAAS,QAAQ,MAA8B,OAAqC;AAClF,SAAO,KACJ,OAAO,CAAC,QAAQ,IAAI,OAAO,SAAU,OAAO,IAAI,OAAO,YAAY,IAAI,KAAK,KAAM,EAClF,QAAQ,CAAC,QAAQ,IAAI,MAAM;AAChC;AAEA,SAAS,aACP,MACA,MACA,MACA,IACA,OACA;AACA,MAAI,KAAK,SAAU;AACnB,MAAI,KAAM,YAAW,SAAS,QAAQ,MAAM,KAAK,EAAG,MAAK,KAAK,KAAK;AAAA,OAC9D;AACH,SAAK,KAAK,EAAE,MAAM,SAAS,WAAW,KAAK,KAAK,CAAC;AACjD,SAAK,KAAK,EAAE,MAAM,YAAY,MAAM,GAAG,CAAC;AAAA,EAC1C;AACA,MAAI,CAAC,KAAK,UAAU;AAClB,SAAK,KAAK;AAAA,MACR,MAAM;AAAA,MACN,YAAY,EAAE,MAAM,KAAK,YAAY,MAAM,OAAO,WAAW,EAAE;AAAA,IACjE,CAAC;AAAA,EACH;AACF;AAEA,eAAe,WACb,OACA,OACA,SACA,QACA;AACA,QAAM,OAAO,WAAW,MAAM;AAC9B,QAAM,OAA8B,EAAE,MAAM,UAAU,QAAQ,CAAC,EAAE;AACjE,QAAM,MAAM,KAAK,IAAI;AACrB,MAAI,OAAO;AACX,MAAI,KAAK;AACT,MAAI,QAAQ;AACZ,MAAI;AACF,QAAI,SAAS,QAAQ,SAAS;AAC5B,WAAK,KAAK,EAAE,MAAM,SAAS,QAAQ,WAAW,OAAO,UAAU,CAAC;AAChE;AAAA,IACF;AACA,eAAW,SAAS,OAAO,MAAM,QAAQ,CAAC,GAAG,CAAC,EAAG,MAAK,KAAK,KAAK;AAChE,QAAI,CAAC,KAAK,UAAU;AAClB,uBAAiB,SAAS,OAAO;AAC/B,YAAI,KAAK,YAAY,SAAS,QAAQ,QAAS;AAC/C,aAAK,OAAO,KAAK,KAAK;AACtB,iBAAS;AACT,aAAK,MAAM;AACX,gBAAQ,YAAY,MAAM,IAAI;AAC9B,qBAAa,MAAM,MAAM,MAAM,OAAO,OAAO,IAAI;AAAA,MACnD;AAAA,IACF;AACA,QAAI,SAAS,QAAQ,WAAW,CAAC,KAAK,UAAU;AAC9C,WAAK,KAAK,EAAE,MAAM,SAAS,QAAQ,WAAW,OAAO,UAAU,CAAC;AAChE;AAAA,IACF;AACA,iBAAa,MAAM,MAAM,MAAM,MAAM,IAAI,KAAK;AAAA,EAChD,SAAS,OAAO;AACd,SAAK,KAAK,EAAE,MAAM,SAAS,QAAQ,SAAS,OAAO,aAAa,KAAK,EAAE,CAAC;AAAA,EAC1E;AACF;AAEA,SAAS,aACP,MACA,MACA,OACA,OACA,MACA;AACA,MAAI,MAAM;AACR,eAAW,SAAS,OAAO,MAAM,KAAK,EAAG,MAAK,KAAK,KAAK;AACxD;AAAA,EACF;AACA,MAAI,UAAU,EAAG,MAAK,KAAK,EAAE,MAAM,gBAAgB,IAAI,MAAM,UAAU,CAAC;AACxE,OAAK,KAAK,EAAE,MAAM,WAAW,WAAW,KAAK,KAAK,CAAC;AACrD;",
+  "names": []
+}

package/dist/index.js

@@ -0,0 +1,37 @@
+import { resolveVoiceAuth } from "./auth.js";
+import { blankUsage } from "./blank-usage.js";
+import { VoiceRequestError } from "./errors.js";
+import {
+  getVoiceModel,
+  listVoiceModels,
+  registerVoiceModel
+} from "./models.js";
+import {
+  connectRealtime,
+  listen,
+  registerRealtimeApiProvider,
+  registerSpeechApiProvider,
+  registerTranscriptionApiProvider,
+  speak,
+  transcribe
+} from "./registry.js";
+import { SpeechEventStream } from "./speech.js";
+import { TranscriptionEventStream } from "./transcription.js";
+export {
+  SpeechEventStream,
+  TranscriptionEventStream,
+  VoiceRequestError,
+  blankUsage,
+  connectRealtime,
+  getVoiceModel,
+  listVoiceModels,
+  listen,
+  registerRealtimeApiProvider,
+  registerSpeechApiProvider,
+  registerTranscriptionApiProvider,
+  registerVoiceModel,
+  resolveVoiceAuth,
+  speak,
+  transcribe
+};
+//# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../src/index.ts"],
+  "sourcesContent": ["export type {\n  AudioChunk,\n  AudioEncoding,\n  AudioFormat,\n  AudioSink,\n  AudioSource,\n  AudioTag,\n  CharAlignment,\n  TextSpan,\n} from './audio.ts';\nexport { resolveVoiceAuth } from './auth.ts';\nexport { blankUsage } from './blank-usage.ts';\nexport { VoiceRequestError } from './errors.ts';\nexport type {\n  Delivery,\n  DeliverySupport,\n  KnownRealtimeApi,\n  KnownSpeechApi,\n  KnownTranscriptionApi,\n  RealtimeApi,\n  RealtimeModel,\n  SpeechApi,\n  SpeechModel,\n  TranscriptionApi,\n  TranscriptionModel,\n  VoiceInfo,\n  VoiceModel,\n  VoiceModelType,\n} from './models.ts';\nexport {\n  getVoiceModel,\n  listVoiceModels,\n  registerVoiceModel,\n} from './models.ts';\nexport type {\n  RealtimeApiProvider,\n  RealtimeContext,\n  RealtimeEvent,\n  RealtimeOptions,\n  RealtimeSession,\n  RealtimeToken,\n} from './realtime.ts';\nexport {\n  connectRealtime,\n  listen,\n  registerRealtimeApiProvider,\n  registerSpeechApiProvider,\n  registerTranscriptionApiProvider,\n  speak,\n  transcribe,\n} from './registry.ts';\nexport type { SpeechApiProvider, SpeechEvent, SpeechInput } from './speech.ts';\nexport { SpeechEventStream } from './speech.ts';\nexport type {\n  AudioInput,\n  ListenOptions,\n  Transcript,\n  TranscriptionApiProvider,\n  TranscriptionEvent,\n  TranscriptSegment,\n} from './transcription.ts';\nexport { TranscriptionEventStream } from './transcription.ts';\n"],
+  "mappings": "AAUA,SAAS,wBAAwB;AACjC,SAAS,kBAAkB;AAC3B,SAAS,yBAAyB;AAiBlC;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,OACK;AASP;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;AAEP,SAAS,yBAAyB;AASlC,SAAS,gCAAgC;",
+  "names": []
+}

package/dist/models.js

@@ -0,0 +1,29 @@
+const catalog = /* @__PURE__ */ new Map();
+function catalogKey(type, provider, id) {
+  return `${type}\0${provider}\0${id}`;
+}
+function registerVoiceModel(model) {
+  const key = catalogKey(model.type, model.provider, model.id);
+  catalog.set(key, model);
+  return () => {
+    if (catalog.get(key) === model) catalog.delete(key);
+  };
+}
+function getVoiceModel(type, provider, id) {
+  return catalog.get(catalogKey(type, provider, id));
+}
+function listVoiceModels(type, provider) {
+  const found = [];
+  for (const model of catalog.values()) {
+    if (model.type !== type) continue;
+    if (provider !== void 0 && model.provider !== provider) continue;
+    found.push(model);
+  }
+  return found;
+}
+export {
+  getVoiceModel,
+  listVoiceModels,
+  registerVoiceModel
+};
+//# sourceMappingURL=models.js.map

package/dist/models.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../src/models.ts"],
+  "sourcesContent": ["import type { BaseModel } from '@earendil-works/pi-ai';\nimport type { AudioFormat } from './audio.ts';\n\nexport type KnownTranscriptionApi =\n  | 'openai-transcriptions'\n  | 'openai-realtime-transcription'\n  | 'xai-stt'\n  | 'elevenlabs-scribe'\n  | 'mistral-transcriptions'\n  | 'fish-asr'\n  | 'cloudflare-ai-run';\n\nexport type TranscriptionApi = KnownTranscriptionApi | (string & {});\n\nexport type KnownSpeechApi =\n  | 'openai-speech'\n  | 'xai-tts'\n  | 'elevenlabs-tts'\n  | 'mistral-speech'\n  | 'fish-tts'\n  | 'google-tts'\n  | 'cloudflare-ai-run'\n  | 'wavespeed';\n\nexport type SpeechApi = KnownSpeechApi | (string & {});\n\nexport type KnownRealtimeApi =\n  | 'openai-realtime'\n  | 'xai-realtime'\n  | 'gemini-live'\n  | 'elevenlabs-agents'\n  | 'fish-agents';\n\nexport type RealtimeApi = KnownRealtimeApi | (string & {});\n\nexport type Delivery = 'interrupt' | 'when_idle' | 'silent';\n\nexport type DeliverySupport = 'native' | 'approximate' | 'unsupported';\n\nexport interface VoiceInfo {\n  id: string;\n  name: string;\n  languages?: string[];\n}\n\nexport interface TranscriptionModel<TApi extends TranscriptionApi>\n  extends Omit<BaseModel<TApi>, 'input'> {\n  type: 'transcription';\n  streaming: boolean;\n  turnDetection: 'none' | 'vad' | 'semantic';\n  inputFormats: AudioFormat[];\n  languages?: string[];\n  diarization?: boolean;\n}\n\nexport interface SpeechModel<TApi extends SpeechApi> extends Omit<BaseModel<TApi>, 'input'> {\n  type: 'speech';\n  textStreaming: boolean;\n  outputFormats: AudioFormat[];\n  voices?: VoiceInfo[];\n  cloning?: boolean;\n}\n\nexport interface RealtimeModel<TApi extends RealtimeApi> extends Omit<BaseModel<TApi>, 'input'> {\n  type: 'realtime';\n  transports: ('websocket' | 'webrtc')[];\n  toolCalls: 'blocking' | 'async';\n  inputFormat: AudioFormat;\n  outputFormat: AudioFormat;\n  video?: { maxFps: number };\n  maxSessionMs?: number;\n  resumable: boolean;\n  ephemeralAuth: boolean;\n  voices: VoiceInfo[];\n  cascade?: { stt: string; llm: string; tts: string };\n  provisioning: 'none' | 'agent';\n  delivery: Record<Delivery, DeliverySupport>;\n  truncation: 'client' | 'server' | 'none';\n  liveToolUpdate: boolean;\n  toolTimeoutMs?: { min: number; max: number };\n}\n\ndeclare module '@earendil-works/pi-ai' {\n  interface ModelTypeMap {\n    transcription: TranscriptionModel<TranscriptionApi>;\n    speech: SpeechModel<SpeechApi>;\n    realtime: RealtimeModel<RealtimeApi>;\n  }\n}\n\nexport type VoiceModel =\n  | TranscriptionModel<TranscriptionApi>\n  | SpeechModel<SpeechApi>\n  | RealtimeModel<RealtimeApi>;\n\nexport type VoiceModelType = VoiceModel['type'];\n\nconst catalog = new Map<string, VoiceModel>();\n\nfunction catalogKey(type: string, provider: string, id: string): string {\n  return `${type}\\0${provider}\\0${id}`;\n}\n\nexport function registerVoiceModel(model: VoiceModel): () => void {\n  const key = catalogKey(model.type, model.provider, model.id);\n  catalog.set(key, model);\n  return () => {\n    if (catalog.get(key) === model) catalog.delete(key);\n  };\n}\n\nexport function getVoiceModel<TType extends VoiceModelType>(\n  type: TType,\n  provider: string,\n  id: string\n): Extract<VoiceModel, { type: TType }> | undefined {\n  return catalog.get(catalogKey(type, provider, id)) as\n    | Extract<VoiceModel, { type: TType }>\n    | undefined;\n}\n\nexport function listVoiceModels<TType extends VoiceModelType>(\n  type: TType,\n  provider?: string\n): Extract<VoiceModel, { type: TType }>[] {\n  const found: Extract<VoiceModel, { type: TType }>[] = [];\n  for (const model of catalog.values()) {\n    if (model.type !== type) continue;\n    if (provider !== undefined && model.provider !== provider) continue;\n    found.push(model as Extract<VoiceModel, { type: TType }>);\n  }\n  return found;\n}\n"],
+  "mappings": "AAiGA,MAAM,UAAU,oBAAI,IAAwB;AAE5C,SAAS,WAAW,MAAc,UAAkB,IAAoB;AACtE,SAAO,GAAG,IAAI,KAAK,QAAQ,KAAK,EAAE;AACpC;AAEO,SAAS,mBAAmB,OAA+B;AAChE,QAAM,MAAM,WAAW,MAAM,MAAM,MAAM,UAAU,MAAM,EAAE;AAC3D,UAAQ,IAAI,KAAK,KAAK;AACtB,SAAO,MAAM;AACX,QAAI,QAAQ,IAAI,GAAG,MAAM,MAAO,SAAQ,OAAO,GAAG;AAAA,EACpD;AACF;AAEO,SAAS,cACd,MACA,UACA,IACkD;AAClD,SAAO,QAAQ,IAAI,WAAW,MAAM,UAAU,EAAE,CAAC;AAGnD;AAEO,SAAS,gBACd,MACA,UACwC;AACxC,QAAM,QAAgD,CAAC;AACvD,aAAW,SAAS,QAAQ,OAAO,GAAG;AACpC,QAAI,MAAM,SAAS,KAAM;AACzB,QAAI,aAAa,UAAa,MAAM,aAAa,SAAU;AAC3D,UAAM,KAAK,KAA6C;AAAA,EAC1D;AACA,SAAO;AACT;",
+  "names": []
+}

package/dist/realtime.js

@@ -0,0 +1 @@
+//# sourceMappingURL=realtime.js.map

package/dist/realtime.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": [],
+  "sourcesContent": [],
+  "mappings": "",
+  "names": []
+}

package/dist/registry.js

@@ -0,0 +1,57 @@
+const transcriptionApis = /* @__PURE__ */ new Map();
+const speechApis = /* @__PURE__ */ new Map();
+const realtimeApis = /* @__PURE__ */ new Map();
+function missing(api) {
+  return new Error(`No API provider registered for api: ${api}`);
+}
+function registerTranscriptionApiProvider(provider) {
+  const entry = { provider };
+  transcriptionApis.set(provider.api, entry);
+  return () => {
+    if (transcriptionApis.get(provider.api) === entry) transcriptionApis.delete(provider.api);
+  };
+}
+function registerSpeechApiProvider(provider) {
+  const entry = { provider };
+  speechApis.set(provider.api, entry);
+  return () => {
+    if (speechApis.get(provider.api) === entry) speechApis.delete(provider.api);
+  };
+}
+function registerRealtimeApiProvider(provider) {
+  const entry = { provider };
+  realtimeApis.set(provider.api, entry);
+  return () => {
+    if (realtimeApis.get(provider.api) === entry) realtimeApis.delete(provider.api);
+  };
+}
+function transcribe(model, input, options) {
+  const transcribeModel = transcriptionApis.get(model.api)?.provider.transcribe;
+  if (!transcribeModel) return Promise.reject(missing(model.api));
+  return transcribeModel(model, input, options);
+}
+function listen(model, audio, options) {
+  const listenModel = transcriptionApis.get(model.api)?.provider.listen;
+  if (!listenModel) throw missing(model.api);
+  return listenModel(model, audio, options);
+}
+function speak(model, input, options) {
+  const speakModel = speechApis.get(model.api)?.provider.speak;
+  if (!speakModel) throw missing(model.api);
+  return speakModel(model, input, options);
+}
+function connectRealtime(model, context, options) {
+  const connect = realtimeApis.get(model.api)?.provider.connect;
+  if (!connect) return Promise.reject(missing(model.api));
+  return connect(model, context, options);
+}
+export {
+  connectRealtime,
+  listen,
+  registerRealtimeApiProvider,
+  registerSpeechApiProvider,
+  registerTranscriptionApiProvider,
+  speak,
+  transcribe
+};
+//# sourceMappingURL=registry.js.map

package/dist/registry.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../src/registry.ts"],
+  "sourcesContent": ["import type { ProviderRequestOptions } from '@earendil-works/pi-ai';\nimport type { AudioSource } from './audio.ts';\nimport type { RealtimeModel, SpeechModel, TranscriptionModel } from './models.ts';\nimport type {\n  RealtimeApiProvider,\n  RealtimeContext,\n  RealtimeOptions,\n  RealtimeSession,\n} from './realtime.ts';\nimport type { SpeechApiProvider, SpeechEventStream, SpeechInput } from './speech.ts';\nimport type {\n  AudioInput,\n  ListenOptions,\n  Transcript,\n  TranscriptionApiProvider,\n  TranscriptionEventStream,\n} from './transcription.ts';\n\ninterface TranscriptionEntry {\n  provider: TranscriptionApiProvider;\n}\n\ninterface SpeechEntry {\n  provider: SpeechApiProvider;\n}\n\ninterface RealtimeEntry {\n  provider: RealtimeApiProvider;\n}\n\nconst transcriptionApis = new Map<string, TranscriptionEntry>();\nconst speechApis = new Map<string, SpeechEntry>();\nconst realtimeApis = new Map<string, RealtimeEntry>();\n\nfunction missing(api: string): Error {\n  return new Error(`No API provider registered for api: ${api}`);\n}\n\nexport function registerTranscriptionApiProvider(provider: TranscriptionApiProvider): () => void {\n  const entry = { provider };\n  transcriptionApis.set(provider.api, entry);\n  return () => {\n    if (transcriptionApis.get(provider.api) === entry) transcriptionApis.delete(provider.api);\n  };\n}\n\nexport function registerSpeechApiProvider(provider: SpeechApiProvider): () => void {\n  const entry = { provider };\n  speechApis.set(provider.api, entry);\n  return () => {\n    if (speechApis.get(provider.api) === entry) speechApis.delete(provider.api);\n  };\n}\n\nexport function registerRealtimeApiProvider(provider: RealtimeApiProvider): () => void {\n  const entry = { provider };\n  realtimeApis.set(provider.api, entry);\n  return () => {\n    if (realtimeApis.get(provider.api) === entry) realtimeApis.delete(provider.api);\n  };\n}\n\nexport function transcribe(\n  model: TranscriptionModel<string>,\n  input: AudioInput,\n  options?: ProviderRequestOptions\n): Promise<Transcript> {\n  const transcribeModel = transcriptionApis.get(model.api)?.provider.transcribe;\n  if (!transcribeModel) return Promise.reject(missing(model.api));\n  return transcribeModel(model, input, options);\n}\n\nexport function listen(\n  model: TranscriptionModel<string>,\n  audio: AudioSource,\n  options?: ListenOptions\n): TranscriptionEventStream {\n  const listenModel = transcriptionApis.get(model.api)?.provider.listen;\n  if (!listenModel) throw missing(model.api);\n  return listenModel(model, audio, options);\n}\n\nexport function speak(\n  model: SpeechModel<string>,\n  input: SpeechInput,\n  options?: ProviderRequestOptions\n): SpeechEventStream {\n  const speakModel = speechApis.get(model.api)?.provider.speak;\n  if (!speakModel) throw missing(model.api);\n  return speakModel(model, input, options);\n}\n\nexport function connectRealtime(\n  model: RealtimeModel<string>,\n  context: RealtimeContext,\n  options?: RealtimeOptions\n): Promise<RealtimeSession> {\n  const connect = realtimeApis.get(model.api)?.provider.connect;\n  if (!connect) return Promise.reject(missing(model.api));\n  return connect(model, context, options);\n}\n"],
+  "mappings": "AA8BA,MAAM,oBAAoB,oBAAI,IAAgC;AAC9D,MAAM,aAAa,oBAAI,IAAyB;AAChD,MAAM,eAAe,oBAAI,IAA2B;AAEpD,SAAS,QAAQ,KAAoB;AACnC,SAAO,IAAI,MAAM,uCAAuC,GAAG,EAAE;AAC/D;AAEO,SAAS,iCAAiC,UAAgD;AAC/F,QAAM,QAAQ,EAAE,SAAS;AACzB,oBAAkB,IAAI,SAAS,KAAK,KAAK;AACzC,SAAO,MAAM;AACX,QAAI,kBAAkB,IAAI,SAAS,GAAG,MAAM,MAAO,mBAAkB,OAAO,SAAS,GAAG;AAAA,EAC1F;AACF;AAEO,SAAS,0BAA0B,UAAyC;AACjF,QAAM,QAAQ,EAAE,SAAS;AACzB,aAAW,IAAI,SAAS,KAAK,KAAK;AAClC,SAAO,MAAM;AACX,QAAI,WAAW,IAAI,SAAS,GAAG,MAAM,MAAO,YAAW,OAAO,SAAS,GAAG;AAAA,EAC5E;AACF;AAEO,SAAS,4BAA4B,UAA2C;AACrF,QAAM,QAAQ,EAAE,SAAS;AACzB,eAAa,IAAI,SAAS,KAAK,KAAK;AACpC,SAAO,MAAM;AACX,QAAI,aAAa,IAAI,SAAS,GAAG,MAAM,MAAO,cAAa,OAAO,SAAS,GAAG;AAAA,EAChF;AACF;AAEO,SAAS,WACd,OACA,OACA,SACqB;AACrB,QAAM,kBAAkB,kBAAkB,IAAI,MAAM,GAAG,GAAG,SAAS;AACnE,MAAI,CAAC,gBAAiB,QAAO,QAAQ,OAAO,QAAQ,MAAM,GAAG,CAAC;AAC9D,SAAO,gBAAgB,OAAO,OAAO,OAAO;AAC9C;AAEO,SAAS,OACd,OACA,OACA,SAC0B;AAC1B,QAAM,cAAc,kBAAkB,IAAI,MAAM,GAAG,GAAG,SAAS;AAC/D,MAAI,CAAC,YAAa,OAAM,QAAQ,MAAM,GAAG;AACzC,SAAO,YAAY,OAAO,OAAO,OAAO;AAC1C;AAEO,SAAS,MACd,OACA,OACA,SACmB;AACnB,QAAM,aAAa,WAAW,IAAI,MAAM,GAAG,GAAG,SAAS;AACvD,MAAI,CAAC,WAAY,OAAM,QAAQ,MAAM,GAAG;AACxC,SAAO,WAAW,OAAO,OAAO,OAAO;AACzC;AAEO,SAAS,gBACd,OACA,SACA,SAC0B;AAC1B,QAAM,UAAU,aAAa,IAAI,MAAM,GAAG,GAAG,SAAS;AACtD,MAAI,CAAC,QAAS,QAAO,QAAQ,OAAO,QAAQ,MAAM,GAAG,CAAC;AACtD,SAAO,QAAQ,OAAO,SAAS,OAAO;AACxC;",
+  "names": []
+}

package/dist/speech.js

@@ -0,0 +1,31 @@
+import { EventStream } from "@earendil-works/pi-ai";
+import { blankUsage } from "./blank-usage.js";
+import { VoiceRequestError } from "./errors.js";
+import { streamResult } from "./stream-result.js";
+class SpeechEventStream extends EventStream {
+  #settled = streamResult();
+  constructor() {
+    super(
+      (event) => event.type === "done" || event.type === "error",
+      (event) => event.type === "done" ? event.usage : blankUsage()
+    );
+  }
+  push(event) {
+    super.push(event);
+    if (event.type === "done") this.#settled.finish(event.usage);
+    else if (event.type === "error")
+      this.#settled.fail(new VoiceRequestError(event.reason, event.error));
+  }
+  end(result) {
+    super.end(result);
+    if (result !== void 0) this.#settled.finish(result);
+    else this.#settled.fail(new VoiceRequestError("error", "Speech ended without usage"));
+  }
+  result() {
+    return this.#settled.result;
+  }
+}
+export {
+  SpeechEventStream
+};
+//# sourceMappingURL=speech.js.map

package/dist/speech.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../src/speech.ts"],
+  "sourcesContent": ["import type { ProviderRequestOptions, Usage } from '@earendil-works/pi-ai';\nimport { EventStream } from '@earendil-works/pi-ai';\nimport type { AudioChunk, AudioFormat, CharAlignment, TextSpan } from './audio.ts';\nimport { blankUsage } from './blank-usage.ts';\nimport { VoiceRequestError } from './errors.ts';\nimport type { SpeechModel } from './models.ts';\nimport { streamResult } from './stream-result.ts';\n\nexport interface SpeechInput {\n  text: string | AsyncIterable<string>;\n  voice: string;\n  language?: string;\n  instructions?: string;\n  format?: AudioFormat;\n}\n\nexport type SpeechEvent =\n  | { type: 'start'; format: AudioFormat }\n  | { type: 'audio'; chunk: AudioChunk; span: TextSpan; alignment?: CharAlignment }\n  | { type: 'done'; usage: Usage }\n  | { type: 'error'; reason: 'aborted' | 'error'; error: string };\n\nexport class SpeechEventStream extends EventStream<SpeechEvent, Usage> {\n  readonly #settled = streamResult<Usage>();\n\n  constructor() {\n    super(\n      (event) => event.type === 'done' || event.type === 'error',\n      (event) => (event.type === 'done' ? event.usage : blankUsage())\n    );\n  }\n\n  override push(event: SpeechEvent): void {\n    super.push(event);\n    if (event.type === 'done') this.#settled.finish(event.usage);\n    else if (event.type === 'error')\n      this.#settled.fail(new VoiceRequestError(event.reason, event.error));\n  }\n\n  override end(result?: Usage): void {\n    super.end(result);\n    if (result !== undefined) this.#settled.finish(result);\n    else this.#settled.fail(new VoiceRequestError('error', 'Speech ended without usage'));\n  }\n\n  override result(): Promise<Usage> {\n    return this.#settled.result;\n  }\n}\n\nexport interface SpeechApiProvider<TApi extends string = string> {\n  api: TApi;\n  speak: (\n    model: SpeechModel<TApi>,\n    input: SpeechInput,\n    options?: ProviderRequestOptions\n  ) => SpeechEventStream;\n}\n"],
+  "mappings": "AACA,SAAS,mBAAmB;AAE5B,SAAS,kBAAkB;AAC3B,SAAS,yBAAyB;AAElC,SAAS,oBAAoB;AAgBtB,MAAM,0BAA0B,YAAgC;AAAA,EAC5D,WAAW,aAAoB;AAAA,EAExC,cAAc;AACZ;AAAA,MACE,CAAC,UAAU,MAAM,SAAS,UAAU,MAAM,SAAS;AAAA,MACnD,CAAC,UAAW,MAAM,SAAS,SAAS,MAAM,QAAQ,WAAW;AAAA,IAC/D;AAAA,EACF;AAAA,EAES,KAAK,OAA0B;AACtC,UAAM,KAAK,KAAK;AAChB,QAAI,MAAM,SAAS,OAAQ,MAAK,SAAS,OAAO,MAAM,KAAK;AAAA,aAClD,MAAM,SAAS;AACtB,WAAK,SAAS,KAAK,IAAI,kBAAkB,MAAM,QAAQ,MAAM,KAAK,CAAC;AAAA,EACvE;AAAA,EAES,IAAI,QAAsB;AACjC,UAAM,IAAI,MAAM;AAChB,QAAI,WAAW,OAAW,MAAK,SAAS,OAAO,MAAM;AAAA,QAChD,MAAK,SAAS,KAAK,IAAI,kBAAkB,SAAS,4BAA4B,CAAC;AAAA,EACtF;AAAA,EAES,SAAyB;AAChC,WAAO,KAAK,SAAS;AAAA,EACvB;AACF;",
+  "names": []
+}

package/dist/stream-result.js

@@ -0,0 +1,28 @@
+function streamResult() {
+  let settled = false;
+  let resolve;
+  let reject;
+  const result = new Promise((res, rej) => {
+    resolve = res;
+    reject = rej;
+  });
+  void result.catch(() => {
+  });
+  return {
+    result,
+    finish(value) {
+      if (settled) return;
+      settled = true;
+      resolve(value);
+    },
+    fail(error) {
+      if (settled) return;
+      settled = true;
+      reject(error);
+    }
+  };
+}
+export {
+  streamResult
+};
+//# sourceMappingURL=stream-result.js.map

package/dist/stream-result.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../src/stream-result.ts"],
+  "sourcesContent": ["export interface StreamResult<T> {\n  readonly result: Promise<T>;\n  finish(value: T): void;\n  fail(error: Error): void;\n}\n\nexport function streamResult<T>(): StreamResult<T> {\n  let settled = false;\n  let resolve!: (value: T) => void;\n  let reject!: (error: Error) => void;\n  const result = new Promise<T>((res, rej) => {\n    resolve = res;\n    reject = rej;\n  });\n  void result.catch(() => {});\n  return {\n    result,\n    finish(value) {\n      if (settled) return;\n      settled = true;\n      resolve(value);\n    },\n    fail(error) {\n      if (settled) return;\n      settled = true;\n      reject(error);\n    },\n  };\n}\n"],
+  "mappings": "AAMO,SAAS,eAAmC;AACjD,MAAI,UAAU;AACd,MAAI;AACJ,MAAI;AACJ,QAAM,SAAS,IAAI,QAAW,CAAC,KAAK,QAAQ;AAC1C,cAAU;AACV,aAAS;AAAA,EACX,CAAC;AACD,OAAK,OAAO,MAAM,MAAM;AAAA,EAAC,CAAC;AAC1B,SAAO;AAAA,IACL;AAAA,IACA,OAAO,OAAO;AACZ,UAAI,QAAS;AACb,gBAAU;AACV,cAAQ,KAAK;AAAA,IACf;AAAA,IACA,KAAK,OAAO;AACV,UAAI,QAAS;AACb,gBAAU;AACV,aAAO,KAAK;AAAA,IACd;AAAA,EACF;AACF;",
+  "names": []
+}

package/dist/transcription.js

@@ -0,0 +1,37 @@
+import { EventStream } from "@earendil-works/pi-ai";
+import { blankUsage } from "./blank-usage.js";
+import { VoiceRequestError } from "./errors.js";
+import { streamResult } from "./stream-result.js";
+function blankTranscript() {
+  return { text: "", usage: blankUsage() };
+}
+class TranscriptionEventStream extends EventStream {
+  #settled = streamResult();
+  constructor() {
+    super(
+      (event) => event.type === "done" || event.type === "error",
+      (event) => event.type === "done" ? event.transcript : blankTranscript()
+    );
+  }
+  push(event) {
+    super.push(event);
+    if (event.type === "done") this.#settled.finish(event.transcript);
+    else if (event.type === "error")
+      this.#settled.fail(new VoiceRequestError(event.reason, event.error));
+  }
+  end(result) {
+    super.end(result);
+    if (result !== void 0) this.#settled.finish(result);
+    else
+      this.#settled.fail(
+        new VoiceRequestError("error", "Transcription ended without a transcript")
+      );
+  }
+  result() {
+    return this.#settled.result;
+  }
+}
+export {
+  TranscriptionEventStream
+};
+//# sourceMappingURL=transcription.js.map

package/dist/transcription.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../src/transcription.ts"],
+  "sourcesContent": ["import type { ProviderRequestOptions, Usage } from '@earendil-works/pi-ai';\nimport { EventStream } from '@earendil-works/pi-ai';\nimport type { AudioSource } from './audio.ts';\nimport { blankUsage } from './blank-usage.ts';\nimport { VoiceRequestError } from './errors.ts';\nimport type { TranscriptionModel } from './models.ts';\nimport { streamResult } from './stream-result.ts';\n\nexport type AudioInput = { data: Uint8Array; mimeType: string } | { url: string };\n\nexport interface TranscriptSegment {\n  text: string;\n  start: number;\n  end: number;\n  speaker?: string;\n}\n\nexport interface Transcript {\n  text: string;\n  language?: string;\n  segments?: TranscriptSegment[];\n  usage: Usage;\n}\n\nexport type TranscriptionEvent =\n  | { type: 'speech_start'; at: number }\n  | { type: 'partial'; segmentId: string; text: string }\n  | { type: 'final'; segmentId: string; text: string; confidence?: number }\n  | { type: 'turn_end'; text: string; at: number }\n  | { type: 'done'; transcript: Transcript }\n  | { type: 'error'; reason: 'aborted' | 'error'; error: string };\n\nexport interface ListenOptions extends ProviderRequestOptions {\n  language?: string;\n  vocabulary?: string[];\n}\n\nfunction blankTranscript(): Transcript {\n  return { text: '', usage: blankUsage() };\n}\n\nexport class TranscriptionEventStream extends EventStream<TranscriptionEvent, Transcript> {\n  readonly #settled = streamResult<Transcript>();\n\n  constructor() {\n    super(\n      (event) => event.type === 'done' || event.type === 'error',\n      (event) => (event.type === 'done' ? event.transcript : blankTranscript())\n    );\n  }\n\n  override push(event: TranscriptionEvent): void {\n    super.push(event);\n    if (event.type === 'done') this.#settled.finish(event.transcript);\n    else if (event.type === 'error')\n      this.#settled.fail(new VoiceRequestError(event.reason, event.error));\n  }\n\n  override end(result?: Transcript): void {\n    super.end(result);\n    if (result !== undefined) this.#settled.finish(result);\n    else\n      this.#settled.fail(\n        new VoiceRequestError('error', 'Transcription ended without a transcript')\n      );\n  }\n\n  override result(): Promise<Transcript> {\n    return this.#settled.result;\n  }\n}\n\nexport interface TranscriptionApiProvider<TApi extends string = string> {\n  api: TApi;\n  transcribe?: (\n    model: TranscriptionModel<TApi>,\n    input: AudioInput,\n    options?: ProviderRequestOptions\n  ) => Promise<Transcript>;\n  listen?: (\n    model: TranscriptionModel<TApi>,\n    audio: AudioSource,\n    options?: ListenOptions\n  ) => TranscriptionEventStream;\n}\n"],
+  "mappings": "AACA,SAAS,mBAAmB;AAE5B,SAAS,kBAAkB;AAC3B,SAAS,yBAAyB;AAElC,SAAS,oBAAoB;AA+B7B,SAAS,kBAA8B;AACrC,SAAO,EAAE,MAAM,IAAI,OAAO,WAAW,EAAE;AACzC;AAEO,MAAM,iCAAiC,YAA4C;AAAA,EAC/E,WAAW,aAAyB;AAAA,EAE7C,cAAc;AACZ;AAAA,MACE,CAAC,UAAU,MAAM,SAAS,UAAU,MAAM,SAAS;AAAA,MACnD,CAAC,UAAW,MAAM,SAAS,SAAS,MAAM,aAAa,gBAAgB;AAAA,IACzE;AAAA,EACF;AAAA,EAES,KAAK,OAAiC;AAC7C,UAAM,KAAK,KAAK;AAChB,QAAI,MAAM,SAAS,OAAQ,MAAK,SAAS,OAAO,MAAM,UAAU;AAAA,aACvD,MAAM,SAAS;AACtB,WAAK,SAAS,KAAK,IAAI,kBAAkB,MAAM,QAAQ,MAAM,KAAK,CAAC;AAAA,EACvE;AAAA,EAES,IAAI,QAA2B;AACtC,UAAM,IAAI,MAAM;AAChB,QAAI,WAAW,OAAW,MAAK,SAAS,OAAO,MAAM;AAAA;AAEnD,WAAK,SAAS;AAAA,QACZ,IAAI,kBAAkB,SAAS,0CAA0C;AAAA,MAC3E;AAAA,EACJ;AAAA,EAES,SAA8B;AACrC,WAAO,KAAK,SAAS;AAAA,EACvB;AACF;",
+  "names": []
+}

package/dist/types/audio.d.ts

@@ -0,0 +1,36 @@
+export type AudioEncoding = 'pcm16' | 'pcmf32' | 'mulaw' | 'alaw' | 'opus' | 'mp3' | 'wav' | 'flac';
+export interface AudioFormat {
+    encoding: AudioEncoding;
+    sampleRate: number;
+    channels: 1 | 2;
+}
+export interface AudioChunk {
+    data: Uint8Array;
+    format: AudioFormat;
+    timestamp: number;
+}
+export type AudioSource = AsyncIterable<AudioChunk>;
+export type TextSpan = {
+    entryId: string;
+    contentIndex: number;
+    start: number;
+    end: number;
+} | {
+    itemId: string;
+    start: number;
+    end: number;
+};
+export interface CharAlignment {
+    startMs: number[];
+    durationMs: number[];
+}
+export interface AudioTag {
+    span: TextSpan;
+    alignment?: CharAlignment;
+}
+export interface AudioSink {
+    write(chunk: AudioChunk, tag?: AudioTag): void;
+    flush(): void;
+    readonly playedMs: number;
+    onPlayback(listener: (tag: AudioTag, offsetMs: number) => void): () => void;
+}

package/dist/types/auth.d.ts

@@ -0,0 +1,3 @@
+import type { AuthResult, Models } from '@earendil-works/pi-ai';
+import type { VoiceModel } from './models.ts';
+export declare function resolveVoiceAuth(models: Models, model: VoiceModel, overrides?: Parameters<Models['getAuth']>[1]): Promise<AuthResult | undefined>;

package/dist/types/blank-usage.d.ts

@@ -0,0 +1,2 @@
+import type { Usage } from '@earendil-works/pi-ai';
+export declare function blankUsage(): Usage;

package/dist/types/errors.d.ts

@@ -0,0 +1,4 @@
+export declare class VoiceRequestError extends Error {
+    readonly reason: 'aborted' | 'error';
+    constructor(reason: 'aborted' | 'error', message: string);
+}

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

@@ -0,0 +1,3 @@
+export { type FauxRealtime, type FauxRealtimeCall, type FauxRealtimeFlags, type FauxRealtimeOptions, registerFauxRealtime, } from './realtime.ts';
+export { type FauxSpeech, type FauxSpeechOptions, registerFauxSpeech, } from './speech.ts';
+export { type FauxTranscription, type FauxTranscriptionCall, type FauxTranscriptionCue, type FauxTranscriptionOptions, registerFauxTranscription, } from './transcription.ts';

package/dist/types/faux/realtime.d.ts

@@ -0,0 +1,53 @@
+import type { AudioChunk } from '../audio.ts';
+import { type Delivery, type DeliverySupport, type RealtimeModel } from '../models.ts';
+import type { RealtimeEvent, RealtimeSession } from '../realtime.ts';
+export type FauxRealtimeCall = {
+    method: 'appendAudio';
+    chunk: AudioChunk;
+} | {
+    method: 'appendVideoFrame';
+    jpeg: Uint8Array;
+} | {
+    method: 'sendText';
+    text: string;
+} | {
+    method: 'inject';
+    text: string;
+    delivery: Delivery;
+} | {
+    method: 'respond';
+    callId: string;
+    result: Parameters<RealtimeSession['respond']>[1];
+    delivery?: Delivery;
+} | {
+    method: 'interrupt';
+    heard?: {
+        itemId: string;
+        heardMs: number;
+    };
+} | {
+    method: 'update';
+    patch: Parameters<RealtimeSession['update']>[0];
+} | {
+    method: 'close';
+};
+export interface FauxRealtimeFlags {
+    toolCalls?: RealtimeModel<string>['toolCalls'];
+    delivery?: Record<Delivery, DeliverySupport>;
+    truncation?: RealtimeModel<string>['truncation'];
+}
+export interface FauxRealtimeOptions extends FauxRealtimeFlags {
+    api?: string;
+    provider?: string;
+    id?: string;
+    name?: string;
+    events?: RealtimeEvent[];
+}
+export interface FauxRealtime {
+    model: RealtimeModel<string>;
+    calls: FauxRealtimeCall[];
+    emit(event: RealtimeEvent): void;
+    setFlags(flags: FauxRealtimeFlags): void;
+    unregister(): void;
+}
+export declare function registerFauxRealtime(options?: FauxRealtimeOptions): FauxRealtime;