@ai-ecoverse/pi-audio 1.23.0 → 1.25.0

package/README.md

@@ -23,6 +23,8 @@ pi-audio does not apply that patch. Hosts such as slicc-agent record monkeypatch
 
 `@ai-ecoverse/pi-audio/durable` appends a `voice.truncate` entry for a `VoiceCut`. The entry has no model messages. Its `edits` replace the target assistant message with the text heard so far and keep tool calls. The stored entry stays intact, and pi-durable derives the shorter message for the next model request. `@earendil-works/pi-durable` is an optional peer (`^1.1.0`); the root package does not import it. `VoiceCut.entryId` is the decimal string of pi-durable's numeric entry id.
 
+`@ai-ecoverse/pi-audio/wavespeed` registers WaveSpeed adapters when `registerWaveSpeed()` is called. Both are batch only. `wavespeed-ai/openai-whisper` submits JSON to `/wavespeed-ai/openai-whisper` and polls `/predictions/{id}/result`. Timestamps stay off, so the catalog price is $0.001 per second (`perInputMinute` 0.06). A byte input is uploaded to `/media/upload/binary` first. `minimax/speech-2.6-turbo` is the same submit/poll path for MP3 (`textStreaming: false`, `streaming: false`) at $0.06 per thousand characters. `voice_id` is a caller id (`cloning: true`). These models are not for live voice.
+
 `@ai-ecoverse/pi-audio/cloudflare` registers Workers AI speech adapters when `registerCloudflare()` is called. One Cloudflare token calls `POST /client/v4/accounts/{account_id}/ai/run/{model}`. Put the account in `baseUrl` or the `cf-account-id` header. Streaming models open `wss://gateway.ai.cloudflare.com/v1/{account}/{gateway}/workers-ai?model=…`. The gateway id is the `cf-aig-gateway-id` header, or `default`. `@cf/deepgram/nova-3` streams (`turnDetection: vad`, catalog price $0.0092 per minute). `@cf/deepgram/flux` streams with semantic end of turn ($0.0077 per minute) and has no batch transcribe. `@cf/openai/whisper-large-v3-turbo` is batch only ($0.000513 per minute). `@cf/deepgram/aura-1` ($0.015 per thousand characters), `aura-2-en`, and `aura-2-es` ($0.03) speak MP3 over HTTP and linear16 over the socket (`textStreaming: true`). `@cf/myshell-ai/melotts` is batch MP3 priced at $0.000205 per audio minute. `@cf/pipecat-ai/smart-turn-v2` is not a transcript model, so it is not registered.
 
 `@ai-ecoverse/pi-audio/google` registers Gemini speech adapters when `registerGoogle()` is called. `gemini-3.8-flash-tts` and `gemini-3.8-flash-lite-tts` post JSON to `models/{id}:generateContent` with `responseModalities: ["AUDIO"]` and a prebuilt `voiceName`. Unary audio is WAV at 24 kHz unless the format asks for `AUDIO_L16`, `AUDIO_MULAW`, or `AUDIO_ALAW`. `textStreaming` is false. Output audio is priced from 25 tokens per second: $0.0135 per minute for Flash and $0.009 per minute for Flash-Lite through December 31, 2026. Text input is $0.50 per million tokens and is not on `audioCost`. Auth is `x-goog-api-key`.
@@ -35,6 +37,8 @@ pi-audio does not apply that patch. Hosts such as slicc-agent record monkeypatch
 
 `calibrateEcho(out, mic, probe, durationMs, signal)` plays a probe and aligns the mic with `AudioOut.reference`. `delayMs` is the lag. `cancellationDb` is how far the leaked copy sits below the reference, and `residualDb` is that leak. A quiet leak recommends `interrupt`, a moderate one `onWords`, and a loud one `headphones`. `chooseCapture` keeps `none` and `wasm`, and moves `browser` to `wasm` when cancellation is under 12 dB. `micOpen('browser')` asks for `echoCancellation` and `playback: 'rtc-loopback'`. `wasm` and `none` turn browser cancellation off and use `webaudio`. `assertNodeEcho` rejects `browser` on Node. `openEchoMic` leaves `browser` and `none` audio as captured. `wasm` runs an injected `EchoCanceller` against `reference`. The homescoop `@ai-ecoverse/wasm-*` echo-canceller package does not exist yet, so this package does not import one. `voiceTurns` passes `options.echo` through `openEchoMic` before `listen`.
 
+`openNodeAudio({ sink: 'none' })` builds that mix on Node. `node-web-audio-api` (optional peer, BSD-3) is the `AudioContext`, created with `sinkId: { type: 'none' }` when the machine has no output device. `web-audio-api` (the audiojs package, MIT) is the pure-JS fallback: it accepts the same sink value but does not render unless a speaker or a stream is attached, so a silent sink is a discarded stream. Hosts pass the microphone as an `AudioSource`. The package does not open `node-web-audio-api`'s mic, and Node echo mode stays `wasm` or `none` with `playback: 'webaudio'`. Finished worklets post `pi-audio-release` and `bindWorkletNode` disconnects them. Returning `false` from `process` does not stop a processor in node-web-audio-api (issue 202). `openMixer(context)` is the same graph on a context the host already opened.
+
 `followBed(bed, soundscape)` cues a `BedChannel` from `VoicePhase`. A loop waits `graceMs` before it starts, so a short gap stays silent. `keep` leaves the current loop running across chained tools. Speech ducks the bed to `duckUnderSpeech`; listening ducks it to `duckUnderUser`. `voiceTurns` does this when `options.soundscape` is set and `out` is an `AudioOut` (`speech`, `bed`, `reference`, `playback`). A plain `AudioSink` still receives only the spoken audio.
 
 `voiceTurns` captions include `agent_spoken`. `upTo` follows `AudioSink` playback: per character when the chunk has alignment, and evenly across the sentence when it does not. `captionsFromHistory(entries)` rebuilds user captions from entries whose `data.source` is `voice`, and agent captions from the assistant text. A `voice.truncate` entry adds `agent_spoken` and `agent_cut` at `heardUpTo`, so the rest of the original text stays available as unheard.

package/dist/index.js

@@ -17,11 +17,13 @@ import {
   openEchoMic
 } from "./echo.js";
 import { VoiceRequestError } from "./errors.js";
+import { openMixer } from "./mixer.js";
 import {
   getVoiceModel,
   listVoiceModels,
   registerVoiceModel
 } from "./models.js";
+import { openNodeAudio } from "./node-host.js";
 import { planDelivery, planTruncation, RealtimeLedger, realtimeSocket } from "./realtime.js";
 import {
   connectRealtime,
@@ -40,15 +42,21 @@ import { TranscriptionEventStream } from "./transcription.js";
 import { voiceTurns } from "./turns.js";
 import { addAudioCounters, priceAudio } from "./usage.js";
 import { energyVad } from "./vad.js";
+import { bindWorkletNode, WORKLET_RELEASE } from "./worklet/release.js";
+import { AUDIO_TAP_PROCESSOR, audioTapWorkletUrl } from "./worklet/tap.js";
 export {
+  AUDIO_TAP_PROCESSOR,
   RealtimeLedger,
   SpeechEventStream,
   TranscriptionEventStream,
   VoiceRequestError,
+  WORKLET_RELEASE,
   addAudioCounters,
   askAgent,
   assertNodeEcho,
+  audioTapWorkletUrl,
   bindThinker,
+  bindWorkletNode,
   blankUsage,
   calibrateEcho,
   captionsFromHistory,
@@ -68,6 +76,8 @@ export {
   micOpen,
   mintRealtimeToken,
   openEchoMic,
+  openMixer,
+  openNodeAudio,
   planDelivery,
   planTruncation,
   priceAudio,

package/dist/index.js.map

@@ -1,7 +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 type { CaptionHistoryEntry } from './captions.ts';\nexport { captionsFromHistory } from './captions.ts';\nexport {\n  convertAudio,\n  convertAudioSource,\n  decodeAudioBase64,\n  encodeAudioBase64,\n} from './convert.ts';\nexport type { EchoCanceller, EchoControl, EchoMode, EchoReport, MicOpen } from './echo.ts';\nexport {\n  assertNodeEcho,\n  calibrateEcho,\n  chooseCapture,\n  echoRuntime,\n  measureEcho,\n  micOpen,\n  openEchoMic,\n} from './echo.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  DeliveryPlan,\n  RealtimeApiProvider,\n  RealtimeContext,\n  RealtimeEvent,\n  RealtimeOptions,\n  RealtimeSession,\n  RealtimeSocket,\n  RealtimeToken,\n  TruncationPlan,\n} from './realtime.ts';\nexport { planDelivery, planTruncation, RealtimeLedger, realtimeSocket } from './realtime.ts';\nexport {\n  connectRealtime,\n  listen,\n  registerRealtimeApiProvider,\n  registerSpeechApiProvider,\n  registerTranscriptionApiProvider,\n  speak,\n  transcribe,\n} from './registry.ts';\nexport type { AudioOut, BedChannel, Cue, CueSource, Soundscape } from './soundscape.ts';\nexport { followBed } from './soundscape.ts';\nexport type { SpeechApiProvider, SpeechEvent, SpeechInput } from './speech.ts';\nexport { SpeechEventStream } from './speech.ts';\nexport type { VoiceAgentEvent, VoiceMessageChange } from './spoken.ts';\nexport type {\n  BindThinkerOptions,\n  BoundThinker,\n  Thinker,\n  ThinkerCaption,\n  ThinkerPhase,\n  ThinkerUpdate,\n} from './talker.ts';\nexport { askAgent, bindThinker, talkerTools } from './talker.ts';\nexport type { RealtimeConstraints } from './tokens.ts';\nexport { mintRealtimeToken } from './tokens.ts';\nexport type {\n  AudioInput,\n  ListenOptions,\n  Transcript,\n  TranscriptionApiProvider,\n  TranscriptionEvent,\n  TranscriptSegment,\n} from './transcription.ts';\nexport { TranscriptionEventStream } from './transcription.ts';\nexport type {\n  BargeIn,\n  CaptionEvent,\n  EchoRecommendation,\n  VoiceConversation,\n  VoiceCut,\n  VoiceModels,\n  VoicePhase,\n  VoiceSession,\n  VoiceTurnMeta,\n  VoiceTurnOptions,\n} from './turns.ts';\nexport { voiceTurns } from './turns.ts';\nexport type {\n  AudioCost,\n  AudioCounters,\n  AudioUsage,\n  RealtimeAudioCost,\n  SpeechAudioCost,\n  TranscriptionAudioCost,\n} from './usage.ts';\nexport { addAudioCounters, priceAudio } from './usage.ts';\nexport type { VadHook, VadUpdate } from './vad.ts';\nexport { energyVad } from './vad.ts';\n"],
-  "mappings": "AAUA,SAAS,wBAAwB;AACjC,SAAS,kBAAkB;AAE3B,SAAS,2BAA2B;AACpC;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;AAEP;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;AACP,SAAS,yBAAyB;AAiBlC;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,OACK;AAYP,SAAS,cAAc,gBAAgB,gBAAgB,sBAAsB;AAC7E;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;AAEP,SAAS,iBAAiB;AAE1B,SAAS,yBAAyB;AAUlC,SAAS,UAAU,aAAa,mBAAmB;AAEnD,SAAS,yBAAyB;AASlC,SAAS,gCAAgC;AAazC,SAAS,kBAAkB;AAS3B,SAAS,kBAAkB,kBAAkB;AAE7C,SAAS,iBAAiB;",
+  "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 type { CaptionHistoryEntry } from './captions.ts';\nexport { captionsFromHistory } from './captions.ts';\nexport {\n  convertAudio,\n  convertAudioSource,\n  decodeAudioBase64,\n  encodeAudioBase64,\n} from './convert.ts';\nexport type { EchoCanceller, EchoControl, EchoMode, EchoReport, MicOpen } from './echo.ts';\nexport {\n  assertNodeEcho,\n  calibrateEcho,\n  chooseCapture,\n  echoRuntime,\n  measureEcho,\n  micOpen,\n  openEchoMic,\n} from './echo.ts';\nexport { VoiceRequestError } from './errors.ts';\nexport type { Mixer, OpenMixerOptions } from './mixer.ts';\nexport { openMixer } from './mixer.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  NodeAudioEngine,\n  NodeAudioModule,\n  NodeAudioOptions,\n  NodeAudioSession,\n} from './node-host.ts';\nexport { openNodeAudio } from './node-host.ts';\nexport type {\n  DeliveryPlan,\n  RealtimeApiProvider,\n  RealtimeContext,\n  RealtimeEvent,\n  RealtimeOptions,\n  RealtimeSession,\n  RealtimeSocket,\n  RealtimeToken,\n  TruncationPlan,\n} from './realtime.ts';\nexport { planDelivery, planTruncation, RealtimeLedger, realtimeSocket } from './realtime.ts';\nexport {\n  connectRealtime,\n  listen,\n  registerRealtimeApiProvider,\n  registerSpeechApiProvider,\n  registerTranscriptionApiProvider,\n  speak,\n  transcribe,\n} from './registry.ts';\nexport type { AudioOut, BedChannel, Cue, CueSource, Soundscape } from './soundscape.ts';\nexport { followBed } from './soundscape.ts';\nexport type { SpeechApiProvider, SpeechEvent, SpeechInput } from './speech.ts';\nexport { SpeechEventStream } from './speech.ts';\nexport type { VoiceAgentEvent, VoiceMessageChange } from './spoken.ts';\nexport type {\n  BindThinkerOptions,\n  BoundThinker,\n  Thinker,\n  ThinkerCaption,\n  ThinkerPhase,\n  ThinkerUpdate,\n} from './talker.ts';\nexport { askAgent, bindThinker, talkerTools } from './talker.ts';\nexport type { RealtimeConstraints } from './tokens.ts';\nexport { mintRealtimeToken } from './tokens.ts';\nexport type {\n  AudioInput,\n  ListenOptions,\n  Transcript,\n  TranscriptionApiProvider,\n  TranscriptionEvent,\n  TranscriptSegment,\n} from './transcription.ts';\nexport { TranscriptionEventStream } from './transcription.ts';\nexport type {\n  BargeIn,\n  CaptionEvent,\n  EchoRecommendation,\n  VoiceConversation,\n  VoiceCut,\n  VoiceModels,\n  VoicePhase,\n  VoiceSession,\n  VoiceTurnMeta,\n  VoiceTurnOptions,\n} from './turns.ts';\nexport { voiceTurns } from './turns.ts';\nexport type {\n  AudioCost,\n  AudioCounters,\n  AudioUsage,\n  RealtimeAudioCost,\n  SpeechAudioCost,\n  TranscriptionAudioCost,\n} from './usage.ts';\nexport { addAudioCounters, priceAudio } from './usage.ts';\nexport type { VadHook, VadUpdate } from './vad.ts';\nexport { energyVad } from './vad.ts';\nexport { bindWorkletNode, WORKLET_RELEASE } from './worklet/release.ts';\nexport { AUDIO_TAP_PROCESSOR, audioTapWorkletUrl } from './worklet/tap.ts';\n"],
+  "mappings": "AAUA,SAAS,wBAAwB;AACjC,SAAS,kBAAkB;AAE3B,SAAS,2BAA2B;AACpC;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;AAEP;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;AACP,SAAS,yBAAyB;AAElC,SAAS,iBAAiB;AAiB1B;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,OACK;AAOP,SAAS,qBAAqB;AAY9B,SAAS,cAAc,gBAAgB,gBAAgB,sBAAsB;AAC7E;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OACK;AAEP,SAAS,iBAAiB;AAE1B,SAAS,yBAAyB;AAUlC,SAAS,UAAU,aAAa,mBAAmB;AAEnD,SAAS,yBAAyB;AASlC,SAAS,gCAAgC;AAazC,SAAS,kBAAkB;AAS3B,SAAS,kBAAkB,kBAAkB;AAE7C,SAAS,iBAAiB;AAC1B,SAAS,iBAAiB,uBAAuB;AACjD,SAAS,qBAAqB,0BAA0B;",
   "names": []
 }

package/dist/mixer.js

@@ -0,0 +1,337 @@
+import { chunkSeconds } from "./audio.js";
+import { convertAudio } from "./convert.js";
+import { echoRuntime } from "./echo.js";
+import { VoiceRequestError } from "./errors.js";
+import { bindWorkletNode, isWorkletRelease } from "./worklet/release.js";
+import { AUDIO_TAP_PROCESSOR, loadAudioTap } from "./worklet/tap.js";
+async function openMixer(context, options = {}) {
+  const playback = playbackOf(options);
+  const mix = context.createGain();
+  const speechGain = context.createGain();
+  const bedGain = context.createGain();
+  speechGain.connect(mix);
+  bedGain.connect(mix);
+  await loadAudioTap(context.audioWorklet, options.tapUrl);
+  const tap = new (workletNode(options))(context, AUDIO_TAP_PROCESSOR, {
+    numberOfInputs: 1,
+    numberOfOutputs: 1,
+    channelCount: 2,
+    outputChannelCount: [2]
+  });
+  mix.connect(tap);
+  tap.connect(context.destination);
+  const queue = new ChunkQueue();
+  const release = watchTap(tap, queue, context);
+  const speech = new SpeechOut(context, speechGain);
+  const bed = new BedPlayer(context, bedGain);
+  return {
+    context,
+    speech,
+    bed,
+    reference: queue,
+    playback,
+    close() {
+      release();
+      bed.close();
+      speech.flush();
+      queue.close();
+      disconnect(speechGain);
+      disconnect(mix);
+    }
+  };
+}
+class ChunkQueue {
+  #chunks = [];
+  #waiters = [];
+  #closed = false;
+  push(chunk) {
+    if (this.#closed) return;
+    const waiter = this.#waiters.shift();
+    if (waiter) {
+      waiter({ value: chunk, done: false });
+      return;
+    }
+    if (this.#chunks.length === 32) this.#chunks.shift();
+    this.#chunks.push(chunk);
+  }
+  close() {
+    if (this.#closed) return;
+    this.#closed = true;
+    const waiters = this.#waiters.splice(0);
+    for (const waiter of waiters) waiter({ value: void 0, done: true });
+  }
+  [Symbol.asyncIterator]() {
+    return this;
+  }
+  next() {
+    const queued = this.#chunks.shift();
+    if (queued) return Promise.resolve({ value: queued, done: false });
+    if (this.#closed) return Promise.resolve({ value: void 0, done: true });
+    return new Promise((resolve) => {
+      this.#waiters.push(resolve);
+    });
+  }
+}
+class SpeechOut {
+  #context;
+  #gain;
+  #listeners = /* @__PURE__ */ new Set();
+  #sources = [];
+  #timers = [];
+  #origin = -1;
+  #writtenMs = 0;
+  #nextAt = 0;
+  constructor(context, gain) {
+    this.#context = context;
+    this.#gain = gain;
+  }
+  get playedMs() {
+    if (this.#origin < 0) return 0;
+    const elapsed = (this.#context.currentTime - this.#origin) * 1e3;
+    if (elapsed <= 0) return 0;
+    if (elapsed > this.#writtenMs) return this.#writtenMs;
+    return elapsed;
+  }
+  write(chunk, tag) {
+    const channels = planar(chunk);
+    const frames = channels[0].length;
+    if (frames === 0) return;
+    const buffer = this.#context.createBuffer(channels.length, frames, chunk.format.sampleRate);
+    writeBuffer(buffer, channels);
+    const source = this.#context.createBufferSource();
+    source.buffer = buffer;
+    source.connect(this.#gain);
+    const start = Math.max(this.#context.currentTime, this.#nextAt);
+    source.start(start);
+    this.#nextAt = start + chunkSeconds(chunk);
+    if (this.#origin < 0) this.#origin = start;
+    this.#writtenMs += chunkSeconds(chunk) * 1e3;
+    if (tag) this.#emit(tag, start);
+    source.onended = () => disconnect(source);
+    this.#sources.push(source);
+  }
+  flush() {
+    this.#stopSources();
+    this.#clearTimers();
+    this.#origin = -1;
+    this.#writtenMs = 0;
+    this.#nextAt = this.#context.currentTime;
+  }
+  onPlayback(listener) {
+    this.#listeners.add(listener);
+    return () => {
+      this.#listeners.delete(listener);
+    };
+  }
+  #emit(tag, start) {
+    const delay = (start - this.#context.currentTime) * 1e3;
+    const fire = () => {
+      for (const listener of this.#listeners) listener(tag, 0);
+    };
+    if (delay <= 0) {
+      fire();
+      return;
+    }
+    this.#timers.push(setTimeout(fire, delay));
+  }
+  #stopSources() {
+    const sources = this.#sources.splice(0);
+    for (const source of sources) {
+      source.stop();
+      disconnect(source);
+    }
+  }
+  #clearTimers() {
+    for (const timer of this.#timers.splice(0)) clearTimeout(timer);
+  }
+}
+class BedPlayer {
+  #context;
+  #output;
+  #loops = [];
+  #shots = [];
+  #timers = [];
+  #generation = 0;
+  constructor(context, output) {
+    this.#context = context;
+    this.#output = output;
+  }
+  play(cue, fadeMs) {
+    if (cue.kind === "keep" || cue.kind === "silence") return;
+    if (cue.kind === "loop") this.#clear(this.#loops, 0);
+    const gain = this.#context.createGain();
+    ramp(gain.gain, cue.gain, this.#context.currentTime, fadeMs);
+    gain.connect(this.#output);
+    const token = this.#generation;
+    const node = this.#start(cue, gain, token);
+    if (!node) return;
+    this.#keep(cue.kind, node);
+  }
+  duck(gain, rampMs) {
+    ramp(this.#output.gain, gain, this.#context.currentTime, rampMs);
+  }
+  stop(fadeMs) {
+    this.#generation += 1;
+    this.#clear(this.#loops, fadeMs);
+    this.#clear(this.#shots, fadeMs);
+  }
+  close() {
+    this.stop(0);
+    for (const timer of this.#timers.splice(0)) clearTimeout(timer);
+    disconnect(this.#output);
+  }
+  #start(cue, gain, token) {
+    const source = cue.source;
+    if ("url" in source) {
+      void this.#loadUrl(source.url, gain, cue.kind, token).catch(() => disconnect(gain));
+      return void 0;
+    }
+    if ("buffer" in source) return this.#buffer(source.buffer, gain, cue.kind);
+    return this.#synth(source, gain);
+  }
+  #synth(source, gain) {
+    const node = source.synth(this.#context);
+    node.connect(gain);
+    startNode(node);
+    return node;
+  }
+  #buffer(buffer, gain, kind) {
+    const source = this.#context.createBufferSource();
+    source.buffer = buffer;
+    source.loop = kind === "loop";
+    source.connect(gain);
+    source.start();
+    if (kind === "oneShot") source.onended = () => disconnect(source);
+    return source;
+  }
+  async #loadUrl(url, gain, kind, token) {
+    const response = await fetch(url).catch(() => void 0);
+    if (response === void 0 || !response.ok || token !== this.#generation) {
+      disconnect(gain);
+      return;
+    }
+    const bytes = await response.arrayBuffer();
+    if (token !== this.#generation) {
+      disconnect(gain);
+      return;
+    }
+    const buffer = await decoded(this.#context, bytes);
+    if (buffer === void 0 || token !== this.#generation) {
+      disconnect(gain);
+      return;
+    }
+    this.#keep(kind, this.#buffer(buffer, gain, kind));
+  }
+  #keep(kind, node) {
+    if (kind === "loop") this.#loops.push(node);
+    else this.#shots.push(node);
+  }
+  #clear(list, fadeMs) {
+    const nodes = list.splice(0);
+    for (const node of nodes) this.#end(node, fadeMs);
+  }
+  #end(node, fadeMs) {
+    const stoppable = node;
+    if (fadeMs <= 0) {
+      if (typeof stoppable.stop === "function") stoppable.stop();
+      disconnect(node);
+      return;
+    }
+    if (typeof stoppable.stop === "function") {
+      stoppable.stop(this.#context.currentTime + fadeMs / 1e3);
+    }
+    this.#timers.push(setTimeout(() => disconnect(node), fadeMs));
+  }
+}
+function playbackOf(options) {
+  const playback = options.playback ?? "webaudio";
+  const runtime = options.runtime ?? echoRuntime(hasMediaDevices());
+  if (runtime === "node" && playback === "rtc-loopback") {
+    throw new VoiceRequestError("error", "Node playback is webaudio");
+  }
+  return playback;
+}
+function hasMediaDevices() {
+  return typeof navigator !== "undefined" && navigator.mediaDevices !== void 0;
+}
+function workletNode(options) {
+  if (options.workletNode) return options.workletNode;
+  const ctor = globalThis.AudioWorkletNode;
+  if (ctor) return ctor;
+  throw new VoiceRequestError("error", "AudioWorkletNode is not available");
+}
+function watchTap(tap, queue, context) {
+  tap.port.onmessage = (event) => {
+    const samples = readSamples(event.data);
+    if (!samples) return;
+    queue.push({
+      data: new Uint8Array(samples.buffer, samples.byteOffset, samples.byteLength).slice(),
+      format: { encoding: "pcmf32", sampleRate: context.sampleRate, channels: 1 },
+      timestamp: context.currentTime * 1e3
+    });
+  };
+  return bindWorkletNode(tap);
+}
+function readSamples(data) {
+  if (isWorkletRelease(data)) return void 0;
+  if (typeof data !== "object" || data === null || !("samples" in data)) return void 0;
+  if (!(data.samples instanceof Float32Array) || data.samples.length === 0) return void 0;
+  return data.samples;
+}
+function planar(chunk) {
+  const pcm = chunk.format.encoding === "pcmf32" ? chunk : convertAudio(chunk, { ...chunk.format, encoding: "pcmf32" });
+  const channels = pcm.format.channels;
+  const view = new DataView(pcm.data.buffer, pcm.data.byteOffset, pcm.data.byteLength);
+  const frames = Math.floor(pcm.data.byteLength / (4 * channels));
+  const out = [];
+  for (let channel = 0; channel < channels; channel += 1) {
+    const samples = new Float32Array(frames);
+    for (let frame = 0; frame < frames; frame += 1) {
+      samples[frame] = view.getFloat32((frame * channels + channel) * 4, true);
+    }
+    out.push(samples);
+  }
+  return out;
+}
+function writeBuffer(buffer, channels) {
+  if (typeof buffer.copyToChannel === "function") {
+    for (let index = 0; index < channels.length; index += 1) {
+      buffer.copyToChannel(channels[index], index);
+    }
+    return;
+  }
+  for (let index = 0; index < channels.length; index += 1) {
+    buffer.getChannelData(index).set(channels[index]);
+  }
+}
+function ramp(param, value, when, fadeMs) {
+  param.cancelScheduledValues(when);
+  if (fadeMs <= 0) {
+    param.setValueAtTime(value, when);
+    return;
+  }
+  param.setValueAtTime(param.value, when);
+  param.linearRampToValueAtTime(value, when + fadeMs / 1e3);
+}
+function startNode(node) {
+  const startable = node;
+  if (typeof startable.start === "function") startable.start();
+}
+async function decoded(context, bytes) {
+  try {
+    return await context.decodeAudioData(bytes);
+  } catch {
+    return void 0;
+  }
+}
+function disconnect(node) {
+  try {
+    node.disconnect();
+  } catch {
+    return;
+  }
+}
+export {
+  openMixer
+};
+//# sourceMappingURL=mixer.js.map

package/dist/mixer.js.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../src/mixer.ts"],
+  "sourcesContent": ["import { type AudioChunk, type AudioSink, type AudioTag, chunkSeconds } from './audio.ts';\nimport { convertAudio } from './convert.ts';\nimport { echoRuntime } from './echo.ts';\nimport { VoiceRequestError } from './errors.ts';\nimport type { AudioOut, Cue, CueSource } from './soundscape.ts';\nimport { bindWorkletNode, isWorkletRelease, type WorkletNodeLike } from './worklet/release.ts';\nimport { AUDIO_TAP_PROCESSOR, loadAudioTap } from './worklet/tap.ts';\n\nexport interface Mixer extends AudioOut {\n  close(): void;\n}\n\nexport interface MixerWorkletNode extends WorkletNodeLike {\n  connect(destination: AudioNode): AudioNode;\n}\n\nexport interface OpenMixerOptions {\n  playback?: 'webaudio' | 'rtc-loopback';\n  runtime?: 'node' | 'browser';\n  tapUrl?: string;\n  workletNode?: new (\n    context: BaseAudioContext,\n    name: string,\n    options?: AudioWorkletNodeOptions\n  ) => MixerWorkletNode;\n}\n\nexport async function openMixer(\n  context: BaseAudioContext,\n  options: OpenMixerOptions = {}\n): Promise<Mixer> {\n  const playback = playbackOf(options);\n  const mix = context.createGain();\n  const speechGain = context.createGain();\n  const bedGain = context.createGain();\n  speechGain.connect(mix);\n  bedGain.connect(mix);\n  await loadAudioTap(context.audioWorklet, options.tapUrl);\n  const tap = new (workletNode(options))(context, AUDIO_TAP_PROCESSOR, {\n    numberOfInputs: 1,\n    numberOfOutputs: 1,\n    channelCount: 2,\n    outputChannelCount: [2],\n  });\n  mix.connect(tap as unknown as AudioNode);\n  tap.connect(context.destination);\n  const queue = new ChunkQueue();\n  const release = watchTap(tap, queue, context);\n  const speech = new SpeechOut(context, speechGain);\n  const bed = new BedPlayer(context, bedGain);\n  return {\n    context,\n    speech,\n    bed,\n    reference: queue,\n    playback,\n    close() {\n      release();\n      bed.close();\n      speech.flush();\n      queue.close();\n      disconnect(speechGain);\n      disconnect(mix);\n    },\n  };\n}\n\nclass ChunkQueue implements AsyncIterable<AudioChunk> {\n  readonly #chunks: AudioChunk[] = [];\n  readonly #waiters: Array<(result: IteratorResult<AudioChunk>) => void> = [];\n  #closed = false;\n\n  push(chunk: AudioChunk): void {\n    if (this.#closed) return;\n    const waiter = this.#waiters.shift();\n    if (waiter) {\n      waiter({ value: chunk, done: false });\n      return;\n    }\n    if (this.#chunks.length === 32) this.#chunks.shift();\n    this.#chunks.push(chunk);\n  }\n\n  close(): void {\n    if (this.#closed) return;\n    this.#closed = true;\n    const waiters = this.#waiters.splice(0);\n    for (const waiter of waiters) waiter({ value: undefined, done: true });\n  }\n\n  [Symbol.asyncIterator](): AsyncIterator<AudioChunk> {\n    return this;\n  }\n\n  next(): Promise<IteratorResult<AudioChunk>> {\n    const queued = this.#chunks.shift();\n    if (queued) return Promise.resolve({ value: queued, done: false });\n    if (this.#closed) return Promise.resolve({ value: undefined, done: true });\n    return new Promise((resolve) => {\n      this.#waiters.push(resolve);\n    });\n  }\n}\n\nclass SpeechOut implements AudioSink {\n  readonly #context: BaseAudioContext;\n  readonly #gain: GainNode;\n  readonly #listeners = new Set<(tag: AudioTag, offsetMs: number) => void>();\n  readonly #sources: AudioBufferSourceNode[] = [];\n  readonly #timers: Array<ReturnType<typeof setTimeout>> = [];\n  #origin = -1;\n  #writtenMs = 0;\n  #nextAt = 0;\n\n  constructor(context: BaseAudioContext, gain: GainNode) {\n    this.#context = context;\n    this.#gain = gain;\n  }\n\n  get playedMs(): number {\n    if (this.#origin < 0) return 0;\n    const elapsed = (this.#context.currentTime - this.#origin) * 1000;\n    if (elapsed <= 0) return 0;\n    if (elapsed > this.#writtenMs) return this.#writtenMs;\n    return elapsed;\n  }\n\n  write(chunk: AudioChunk, tag?: AudioTag): void {\n    const channels = planar(chunk);\n    const frames = channels[0].length;\n    if (frames === 0) return;\n    const buffer = this.#context.createBuffer(channels.length, frames, chunk.format.sampleRate);\n    writeBuffer(buffer, channels);\n    const source = this.#context.createBufferSource();\n    source.buffer = buffer;\n    source.connect(this.#gain);\n    const start = Math.max(this.#context.currentTime, this.#nextAt);\n    source.start(start);\n    this.#nextAt = start + chunkSeconds(chunk);\n    if (this.#origin < 0) this.#origin = start;\n    this.#writtenMs += chunkSeconds(chunk) * 1000;\n    if (tag) this.#emit(tag, start);\n    source.onended = () => disconnect(source);\n    this.#sources.push(source);\n  }\n\n  flush(): void {\n    this.#stopSources();\n    this.#clearTimers();\n    this.#origin = -1;\n    this.#writtenMs = 0;\n    this.#nextAt = this.#context.currentTime;\n  }\n\n  onPlayback(listener: (tag: AudioTag, offsetMs: number) => void): () => void {\n    this.#listeners.add(listener);\n    return () => {\n      this.#listeners.delete(listener);\n    };\n  }\n\n  #emit(tag: AudioTag, start: number): void {\n    const delay = (start - this.#context.currentTime) * 1000;\n    const fire = () => {\n      for (const listener of this.#listeners) listener(tag, 0);\n    };\n    if (delay <= 0) {\n      fire();\n      return;\n    }\n    this.#timers.push(setTimeout(fire, delay));\n  }\n\n  #stopSources(): void {\n    const sources = this.#sources.splice(0);\n    for (const source of sources) {\n      source.stop();\n      disconnect(source);\n    }\n  }\n\n  #clearTimers(): void {\n    for (const timer of this.#timers.splice(0)) clearTimeout(timer);\n  }\n}\n\nclass BedPlayer {\n  readonly #context: BaseAudioContext;\n  readonly #output: GainNode;\n  readonly #loops: AudioNode[] = [];\n  readonly #shots: AudioNode[] = [];\n  readonly #timers: Array<ReturnType<typeof setTimeout>> = [];\n  #generation = 0;\n\n  constructor(context: BaseAudioContext, output: GainNode) {\n    this.#context = context;\n    this.#output = output;\n  }\n\n  play(cue: Cue, fadeMs: number): void {\n    if (cue.kind === 'keep' || cue.kind === 'silence') return;\n    if (cue.kind === 'loop') this.#clear(this.#loops, 0);\n    const gain = this.#context.createGain();\n    ramp(gain.gain, cue.gain, this.#context.currentTime, fadeMs);\n    gain.connect(this.#output);\n    const token = this.#generation;\n    const node = this.#start(cue, gain, token);\n    if (!node) return;\n    this.#keep(cue.kind, node);\n  }\n\n  duck(gain: number, rampMs: number): void {\n    ramp(this.#output.gain, gain, this.#context.currentTime, rampMs);\n  }\n\n  stop(fadeMs: number): void {\n    this.#generation += 1;\n    this.#clear(this.#loops, fadeMs);\n    this.#clear(this.#shots, fadeMs);\n  }\n\n  close(): void {\n    this.stop(0);\n    for (const timer of this.#timers.splice(0)) clearTimeout(timer);\n    disconnect(this.#output);\n  }\n\n  #start(\n    cue: Extract<Cue, { kind: 'loop' | 'oneShot' }>,\n    gain: GainNode,\n    token: number\n  ): AudioNode | undefined {\n    const source = cue.source;\n    if ('url' in source) {\n      void this.#loadUrl(source.url, gain, cue.kind, token).catch(() => disconnect(gain));\n      return undefined;\n    }\n    if ('buffer' in source) return this.#buffer(source.buffer, gain, cue.kind);\n    return this.#synth(source, gain);\n  }\n\n  #synth(source: Extract<CueSource, { synth: unknown }>, gain: GainNode): AudioNode {\n    const node = source.synth(this.#context);\n    node.connect(gain);\n    startNode(node);\n    return node;\n  }\n\n  #buffer(buffer: AudioBuffer, gain: GainNode, kind: 'loop' | 'oneShot'): AudioNode {\n    const source = this.#context.createBufferSource();\n    source.buffer = buffer;\n    source.loop = kind === 'loop';\n    source.connect(gain);\n    source.start();\n    if (kind === 'oneShot') source.onended = () => disconnect(source);\n    return source;\n  }\n\n  async #loadUrl(\n    url: string,\n    gain: GainNode,\n    kind: 'loop' | 'oneShot',\n    token: number\n  ): Promise<void> {\n    const response = await fetch(url).catch(() => undefined);\n    if (response === undefined || !response.ok || token !== this.#generation) {\n      disconnect(gain);\n      return;\n    }\n    const bytes = await response.arrayBuffer();\n    if (token !== this.#generation) {\n      disconnect(gain);\n      return;\n    }\n    const buffer = await decoded(this.#context, bytes);\n    if (buffer === undefined || token !== this.#generation) {\n      disconnect(gain);\n      return;\n    }\n    this.#keep(kind, this.#buffer(buffer, gain, kind));\n  }\n\n  #keep(kind: 'loop' | 'oneShot', node: AudioNode): void {\n    if (kind === 'loop') this.#loops.push(node);\n    else this.#shots.push(node);\n  }\n\n  #clear(list: AudioNode[], fadeMs: number): void {\n    const nodes = list.splice(0);\n    for (const node of nodes) this.#end(node, fadeMs);\n  }\n\n  #end(node: AudioNode, fadeMs: number): void {\n    const stoppable = node as AudioNode & { stop?: (when?: number) => void };\n    if (fadeMs <= 0) {\n      if (typeof stoppable.stop === 'function') stoppable.stop();\n      disconnect(node);\n      return;\n    }\n    if (typeof stoppable.stop === 'function') {\n      stoppable.stop(this.#context.currentTime + fadeMs / 1000);\n    }\n    this.#timers.push(setTimeout(() => disconnect(node), fadeMs));\n  }\n}\n\nfunction playbackOf(options: OpenMixerOptions): 'webaudio' | 'rtc-loopback' {\n  const playback = options.playback ?? 'webaudio';\n  const runtime = options.runtime ?? echoRuntime(hasMediaDevices());\n  if (runtime === 'node' && playback === 'rtc-loopback') {\n    throw new VoiceRequestError('error', 'Node playback is webaudio');\n  }\n  return playback;\n}\n\nfunction hasMediaDevices(): boolean {\n  return typeof navigator !== 'undefined' && navigator.mediaDevices !== undefined;\n}\n\nfunction workletNode(options: OpenMixerOptions): NonNullable<OpenMixerOptions['workletNode']> {\n  if (options.workletNode) return options.workletNode;\n  const ctor = (globalThis as { AudioWorkletNode?: OpenMixerOptions['workletNode'] })\n    .AudioWorkletNode;\n  if (ctor) return ctor;\n  throw new VoiceRequestError('error', 'AudioWorkletNode is not available');\n}\n\nfunction watchTap(tap: WorkletNodeLike, queue: ChunkQueue, context: BaseAudioContext): () => void {\n  tap.port.onmessage = (event) => {\n    const samples = readSamples(event.data);\n    if (!samples) return;\n    queue.push({\n      data: new Uint8Array(samples.buffer, samples.byteOffset, samples.byteLength).slice(),\n      format: { encoding: 'pcmf32', sampleRate: context.sampleRate, channels: 1 },\n      timestamp: context.currentTime * 1000,\n    });\n  };\n  return bindWorkletNode(tap);\n}\n\nfunction readSamples(data: unknown): Float32Array | undefined {\n  if (isWorkletRelease(data)) return undefined;\n  if (typeof data !== 'object' || data === null || !('samples' in data)) return undefined;\n  if (!(data.samples instanceof Float32Array) || data.samples.length === 0) return undefined;\n  return data.samples;\n}\n\nfunction planar(chunk: AudioChunk): Float32Array[] {\n  const pcm =\n    chunk.format.encoding === 'pcmf32'\n      ? chunk\n      : convertAudio(chunk, { ...chunk.format, encoding: 'pcmf32' });\n  const channels = pcm.format.channels;\n  const view = new DataView(pcm.data.buffer, pcm.data.byteOffset, pcm.data.byteLength);\n  const frames = Math.floor(pcm.data.byteLength / (4 * channels));\n  const out: Float32Array[] = [];\n  for (let channel = 0; channel < channels; channel += 1) {\n    const samples = new Float32Array(frames);\n    for (let frame = 0; frame < frames; frame += 1) {\n      samples[frame] = view.getFloat32((frame * channels + channel) * 4, true);\n    }\n    out.push(samples);\n  }\n  return out;\n}\n\nfunction writeBuffer(buffer: AudioBuffer, channels: Float32Array[]): void {\n  if (typeof buffer.copyToChannel === 'function') {\n    for (let index = 0; index < channels.length; index += 1) {\n      buffer.copyToChannel(channels[index] as Float32Array<ArrayBuffer>, index);\n    }\n    return;\n  }\n  for (let index = 0; index < channels.length; index += 1) {\n    buffer.getChannelData(index).set(channels[index] as Float32Array<ArrayBuffer>);\n  }\n}\n\nfunction ramp(param: AudioParam, value: number, when: number, fadeMs: number): void {\n  param.cancelScheduledValues(when);\n  if (fadeMs <= 0) {\n    param.setValueAtTime(value, when);\n    return;\n  }\n  param.setValueAtTime(param.value, when);\n  param.linearRampToValueAtTime(value, when + fadeMs / 1000);\n}\n\nfunction startNode(node: AudioNode): void {\n  const startable = node as AudioNode & { start?: (when?: number) => void };\n  if (typeof startable.start === 'function') startable.start();\n}\n\nasync function decoded(\n  context: BaseAudioContext,\n  bytes: ArrayBuffer\n): Promise<AudioBuffer | undefined> {\n  try {\n    return await context.decodeAudioData(bytes);\n  } catch {\n    return undefined;\n  }\n}\n\nfunction disconnect(node: { disconnect(): void }): void {\n  try {\n    node.disconnect();\n  } catch {\n    return;\n  }\n}\n"],
+  "mappings": "AAAA,SAAyD,oBAAoB;AAC7E,SAAS,oBAAoB;AAC7B,SAAS,mBAAmB;AAC5B,SAAS,yBAAyB;AAElC,SAAS,iBAAiB,wBAA8C;AACxE,SAAS,qBAAqB,oBAAoB;AAqBlD,eAAsB,UACpB,SACA,UAA4B,CAAC,GACb;AAChB,QAAM,WAAW,WAAW,OAAO;AACnC,QAAM,MAAM,QAAQ,WAAW;AAC/B,QAAM,aAAa,QAAQ,WAAW;AACtC,QAAM,UAAU,QAAQ,WAAW;AACnC,aAAW,QAAQ,GAAG;AACtB,UAAQ,QAAQ,GAAG;AACnB,QAAM,aAAa,QAAQ,cAAc,QAAQ,MAAM;AACvD,QAAM,MAAM,KAAK,YAAY,OAAO,GAAG,SAAS,qBAAqB;AAAA,IACnE,gBAAgB;AAAA,IAChB,iBAAiB;AAAA,IACjB,cAAc;AAAA,IACd,oBAAoB,CAAC,CAAC;AAAA,EACxB,CAAC;AACD,MAAI,QAAQ,GAA2B;AACvC,MAAI,QAAQ,QAAQ,WAAW;AAC/B,QAAM,QAAQ,IAAI,WAAW;AAC7B,QAAM,UAAU,SAAS,KAAK,OAAO,OAAO;AAC5C,QAAM,SAAS,IAAI,UAAU,SAAS,UAAU;AAChD,QAAM,MAAM,IAAI,UAAU,SAAS,OAAO;AAC1C,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA;AAAA,IACA,WAAW;AAAA,IACX;AAAA,IACA,QAAQ;AACN,cAAQ;AACR,UAAI,MAAM;AACV,aAAO,MAAM;AACb,YAAM,MAAM;AACZ,iBAAW,UAAU;AACrB,iBAAW,GAAG;AAAA,IAChB;AAAA,EACF;AACF;AAEA,MAAM,WAAgD;AAAA,EAC3C,UAAwB,CAAC;AAAA,EACzB,WAAgE,CAAC;AAAA,EAC1E,UAAU;AAAA,EAEV,KAAK,OAAyB;AAC5B,QAAI,KAAK,QAAS;AAClB,UAAM,SAAS,KAAK,SAAS,MAAM;AACnC,QAAI,QAAQ;AACV,aAAO,EAAE,OAAO,OAAO,MAAM,MAAM,CAAC;AACpC;AAAA,IACF;AACA,QAAI,KAAK,QAAQ,WAAW,GAAI,MAAK,QAAQ,MAAM;AACnD,SAAK,QAAQ,KAAK,KAAK;AAAA,EACzB;AAAA,EAEA,QAAc;AACZ,QAAI,KAAK,QAAS;AAClB,SAAK,UAAU;AACf,UAAM,UAAU,KAAK,SAAS,OAAO,CAAC;AACtC,eAAW,UAAU,QAAS,QAAO,EAAE,OAAO,QAAW,MAAM,KAAK,CAAC;AAAA,EACvE;AAAA,EAEA,CAAC,OAAO,aAAa,IAA+B;AAClD,WAAO;AAAA,EACT;AAAA,EAEA,OAA4C;AAC1C,UAAM,SAAS,KAAK,QAAQ,MAAM;AAClC,QAAI,OAAQ,QAAO,QAAQ,QAAQ,EAAE,OAAO,QAAQ,MAAM,MAAM,CAAC;AACjE,QAAI,KAAK,QAAS,QAAO,QAAQ,QAAQ,EAAE,OAAO,QAAW,MAAM,KAAK,CAAC;AACzE,WAAO,IAAI,QAAQ,CAAC,YAAY;AAC9B,WAAK,SAAS,KAAK,OAAO;AAAA,IAC5B,CAAC;AAAA,EACH;AACF;AAEA,MAAM,UAA+B;AAAA,EAC1B;AAAA,EACA;AAAA,EACA,aAAa,oBAAI,IAA+C;AAAA,EAChE,WAAoC,CAAC;AAAA,EACrC,UAAgD,CAAC;AAAA,EAC1D,UAAU;AAAA,EACV,aAAa;AAAA,EACb,UAAU;AAAA,EAEV,YAAY,SAA2B,MAAgB;AACrD,SAAK,WAAW;AAChB,SAAK,QAAQ;AAAA,EACf;AAAA,EAEA,IAAI,WAAmB;AACrB,QAAI,KAAK,UAAU,EAAG,QAAO;AAC7B,UAAM,WAAW,KAAK,SAAS,cAAc,KAAK,WAAW;AAC7D,QAAI,WAAW,EAAG,QAAO;AACzB,QAAI,UAAU,KAAK,WAAY,QAAO,KAAK;AAC3C,WAAO;AAAA,EACT;AAAA,EAEA,MAAM,OAAmB,KAAsB;AAC7C,UAAM,WAAW,OAAO,KAAK;AAC7B,UAAM,SAAS,SAAS,CAAC,EAAE;AAC3B,QAAI,WAAW,EAAG;AAClB,UAAM,SAAS,KAAK,SAAS,aAAa,SAAS,QAAQ,QAAQ,MAAM,OAAO,UAAU;AAC1F,gBAAY,QAAQ,QAAQ;AAC5B,UAAM,SAAS,KAAK,SAAS,mBAAmB;AAChD,WAAO,SAAS;AAChB,WAAO,QAAQ,KAAK,KAAK;AACzB,UAAM,QAAQ,KAAK,IAAI,KAAK,SAAS,aAAa,KAAK,OAAO;AAC9D,WAAO,MAAM,KAAK;AAClB,SAAK,UAAU,QAAQ,aAAa,KAAK;AACzC,QAAI,KAAK,UAAU,EAAG,MAAK,UAAU;AACrC,SAAK,cAAc,aAAa,KAAK,IAAI;AACzC,QAAI,IAAK,MAAK,MAAM,KAAK,KAAK;AAC9B,WAAO,UAAU,MAAM,WAAW,MAAM;AACxC,SAAK,SAAS,KAAK,MAAM;AAAA,EAC3B;AAAA,EAEA,QAAc;AACZ,SAAK,aAAa;AAClB,SAAK,aAAa;AAClB,SAAK,UAAU;AACf,SAAK,aAAa;AAClB,SAAK,UAAU,KAAK,SAAS;AAAA,EAC/B;AAAA,EAEA,WAAW,UAAiE;AAC1E,SAAK,WAAW,IAAI,QAAQ;AAC5B,WAAO,MAAM;AACX,WAAK,WAAW,OAAO,QAAQ;AAAA,IACjC;AAAA,EACF;AAAA,EAEA,MAAM,KAAe,OAAqB;AACxC,UAAM,SAAS,QAAQ,KAAK,SAAS,eAAe;AACpD,UAAM,OAAO,MAAM;AACjB,iBAAW,YAAY,KAAK,WAAY,UAAS,KAAK,CAAC;AAAA,IACzD;AACA,QAAI,SAAS,GAAG;AACd,WAAK;AACL;AAAA,IACF;AACA,SAAK,QAAQ,KAAK,WAAW,MAAM,KAAK,CAAC;AAAA,EAC3C;AAAA,EAEA,eAAqB;AACnB,UAAM,UAAU,KAAK,SAAS,OAAO,CAAC;AACtC,eAAW,UAAU,SAAS;AAC5B,aAAO,KAAK;AACZ,iBAAW,MAAM;AAAA,IACnB;AAAA,EACF;AAAA,EAEA,eAAqB;AACnB,eAAW,SAAS,KAAK,QAAQ,OAAO,CAAC,EAAG,cAAa,KAAK;AAAA,EAChE;AACF;AAEA,MAAM,UAAU;AAAA,EACL;AAAA,EACA;AAAA,EACA,SAAsB,CAAC;AAAA,EACvB,SAAsB,CAAC;AAAA,EACvB,UAAgD,CAAC;AAAA,EAC1D,cAAc;AAAA,EAEd,YAAY,SAA2B,QAAkB;AACvD,SAAK,WAAW;AAChB,SAAK,UAAU;AAAA,EACjB;AAAA,EAEA,KAAK,KAAU,QAAsB;AACnC,QAAI,IAAI,SAAS,UAAU,IAAI,SAAS,UAAW;AACnD,QAAI,IAAI,SAAS,OAAQ,MAAK,OAAO,KAAK,QAAQ,CAAC;AACnD,UAAM,OAAO,KAAK,SAAS,WAAW;AACtC,SAAK,KAAK,MAAM,IAAI,MAAM,KAAK,SAAS,aAAa,MAAM;AAC3D,SAAK,QAAQ,KAAK,OAAO;AACzB,UAAM,QAAQ,KAAK;AACnB,UAAM,OAAO,KAAK,OAAO,KAAK,MAAM,KAAK;AACzC,QAAI,CAAC,KAAM;AACX,SAAK,MAAM,IAAI,MAAM,IAAI;AAAA,EAC3B;AAAA,EAEA,KAAK,MAAc,QAAsB;AACvC,SAAK,KAAK,QAAQ,MAAM,MAAM,KAAK,SAAS,aAAa,MAAM;AAAA,EACjE;AAAA,EAEA,KAAK,QAAsB;AACzB,SAAK,eAAe;AACpB,SAAK,OAAO,KAAK,QAAQ,MAAM;AAC/B,SAAK,OAAO,KAAK,QAAQ,MAAM;AAAA,EACjC;AAAA,EAEA,QAAc;AACZ,SAAK,KAAK,CAAC;AACX,eAAW,SAAS,KAAK,QAAQ,OAAO,CAAC,EAAG,cAAa,KAAK;AAC9D,eAAW,KAAK,OAAO;AAAA,EACzB;AAAA,EAEA,OACE,KACA,MACA,OACuB;AACvB,UAAM,SAAS,IAAI;AACnB,QAAI,SAAS,QAAQ;AACnB,WAAK,KAAK,SAAS,OAAO,KAAK,MAAM,IAAI,MAAM,KAAK,EAAE,MAAM,MAAM,WAAW,IAAI,CAAC;AAClF,aAAO;AAAA,IACT;AACA,QAAI,YAAY,OAAQ,QAAO,KAAK,QAAQ,OAAO,QAAQ,MAAM,IAAI,IAAI;AACzE,WAAO,KAAK,OAAO,QAAQ,IAAI;AAAA,EACjC;AAAA,EAEA,OAAO,QAAgD,MAA2B;AAChF,UAAM,OAAO,OAAO,MAAM,KAAK,QAAQ;AACvC,SAAK,QAAQ,IAAI;AACjB,cAAU,IAAI;AACd,WAAO;AAAA,EACT;AAAA,EAEA,QAAQ,QAAqB,MAAgB,MAAqC;AAChF,UAAM,SAAS,KAAK,SAAS,mBAAmB;AAChD,WAAO,SAAS;AAChB,WAAO,OAAO,SAAS;AACvB,WAAO,QAAQ,IAAI;AACnB,WAAO,MAAM;AACb,QAAI,SAAS,UAAW,QAAO,UAAU,MAAM,WAAW,MAAM;AAChE,WAAO;AAAA,EACT;AAAA,EAEA,MAAM,SACJ,KACA,MACA,MACA,OACe;AACf,UAAM,WAAW,MAAM,MAAM,GAAG,EAAE,MAAM,MAAM,MAAS;AACvD,QAAI,aAAa,UAAa,CAAC,SAAS,MAAM,UAAU,KAAK,aAAa;AACxE,iBAAW,IAAI;AACf;AAAA,IACF;AACA,UAAM,QAAQ,MAAM,SAAS,YAAY;AACzC,QAAI,UAAU,KAAK,aAAa;AAC9B,iBAAW,IAAI;AACf;AAAA,IACF;AACA,UAAM,SAAS,MAAM,QAAQ,KAAK,UAAU,KAAK;AACjD,QAAI,WAAW,UAAa,UAAU,KAAK,aAAa;AACtD,iBAAW,IAAI;AACf;AAAA,IACF;AACA,SAAK,MAAM,MAAM,KAAK,QAAQ,QAAQ,MAAM,IAAI,CAAC;AAAA,EACnD;AAAA,EAEA,MAAM,MAA0B,MAAuB;AACrD,QAAI,SAAS,OAAQ,MAAK,OAAO,KAAK,IAAI;AAAA,QACrC,MAAK,OAAO,KAAK,IAAI;AAAA,EAC5B;AAAA,EAEA,OAAO,MAAmB,QAAsB;AAC9C,UAAM,QAAQ,KAAK,OAAO,CAAC;AAC3B,eAAW,QAAQ,MAAO,MAAK,KAAK,MAAM,MAAM;AAAA,EAClD;AAAA,EAEA,KAAK,MAAiB,QAAsB;AAC1C,UAAM,YAAY;AAClB,QAAI,UAAU,GAAG;AACf,UAAI,OAAO,UAAU,SAAS,WAAY,WAAU,KAAK;AACzD,iBAAW,IAAI;AACf;AAAA,IACF;AACA,QAAI,OAAO,UAAU,SAAS,YAAY;AACxC,gBAAU,KAAK,KAAK,SAAS,cAAc,SAAS,GAAI;AAAA,IAC1D;AACA,SAAK,QAAQ,KAAK,WAAW,MAAM,WAAW,IAAI,GAAG,MAAM,CAAC;AAAA,EAC9D;AACF;AAEA,SAAS,WAAW,SAAwD;AAC1E,QAAM,WAAW,QAAQ,YAAY;AACrC,QAAM,UAAU,QAAQ,WAAW,YAAY,gBAAgB,CAAC;AAChE,MAAI,YAAY,UAAU,aAAa,gBAAgB;AACrD,UAAM,IAAI,kBAAkB,SAAS,2BAA2B;AAAA,EAClE;AACA,SAAO;AACT;AAEA,SAAS,kBAA2B;AAClC,SAAO,OAAO,cAAc,eAAe,UAAU,iBAAiB;AACxE;AAEA,SAAS,YAAY,SAAyE;AAC5F,MAAI,QAAQ,YAAa,QAAO,QAAQ;AACxC,QAAM,OAAQ,WACX;AACH,MAAI,KAAM,QAAO;AACjB,QAAM,IAAI,kBAAkB,SAAS,mCAAmC;AAC1E;AAEA,SAAS,SAAS,KAAsB,OAAmB,SAAuC;AAChG,MAAI,KAAK,YAAY,CAAC,UAAU;AAC9B,UAAM,UAAU,YAAY,MAAM,IAAI;AACtC,QAAI,CAAC,QAAS;AACd,UAAM,KAAK;AAAA,MACT,MAAM,IAAI,WAAW,QAAQ,QAAQ,QAAQ,YAAY,QAAQ,UAAU,EAAE,MAAM;AAAA,MACnF,QAAQ,EAAE,UAAU,UAAU,YAAY,QAAQ,YAAY,UAAU,EAAE;AAAA,MAC1E,WAAW,QAAQ,cAAc;AAAA,IACnC,CAAC;AAAA,EACH;AACA,SAAO,gBAAgB,GAAG;AAC5B;AAEA,SAAS,YAAY,MAAyC;AAC5D,MAAI,iBAAiB,IAAI,EAAG,QAAO;AACnC,MAAI,OAAO,SAAS,YAAY,SAAS,QAAQ,EAAE,aAAa,MAAO,QAAO;AAC9E,MAAI,EAAE,KAAK,mBAAmB,iBAAiB,KAAK,QAAQ,WAAW,EAAG,QAAO;AACjF,SAAO,KAAK;AACd;AAEA,SAAS,OAAO,OAAmC;AACjD,QAAM,MACJ,MAAM,OAAO,aAAa,WACtB,QACA,aAAa,OAAO,EAAE,GAAG,MAAM,QAAQ,UAAU,SAAS,CAAC;AACjE,QAAM,WAAW,IAAI,OAAO;AAC5B,QAAM,OAAO,IAAI,SAAS,IAAI,KAAK,QAAQ,IAAI,KAAK,YAAY,IAAI,KAAK,UAAU;AACnF,QAAM,SAAS,KAAK,MAAM,IAAI,KAAK,cAAc,IAAI,SAAS;AAC9D,QAAM,MAAsB,CAAC;AAC7B,WAAS,UAAU,GAAG,UAAU,UAAU,WAAW,GAAG;AACtD,UAAM,UAAU,IAAI,aAAa,MAAM;AACvC,aAAS,QAAQ,GAAG,QAAQ,QAAQ,SAAS,GAAG;AAC9C,cAAQ,KAAK,IAAI,KAAK,YAAY,QAAQ,WAAW,WAAW,GAAG,IAAI;AAAA,IACzE;AACA,QAAI,KAAK,OAAO;AAAA,EAClB;AACA,SAAO;AACT;AAEA,SAAS,YAAY,QAAqB,UAAgC;AACxE,MAAI,OAAO,OAAO,kBAAkB,YAAY;AAC9C,aAAS,QAAQ,GAAG,QAAQ,SAAS,QAAQ,SAAS,GAAG;AACvD,aAAO,cAAc,SAAS,KAAK,GAAgC,KAAK;AAAA,IAC1E;AACA;AAAA,EACF;AACA,WAAS,QAAQ,GAAG,QAAQ,SAAS,QAAQ,SAAS,GAAG;AACvD,WAAO,eAAe,KAAK,EAAE,IAAI,SAAS,KAAK,CAA8B;AAAA,EAC/E;AACF;AAEA,SAAS,KAAK,OAAmB,OAAe,MAAc,QAAsB;AAClF,QAAM,sBAAsB,IAAI;AAChC,MAAI,UAAU,GAAG;AACf,UAAM,eAAe,OAAO,IAAI;AAChC;AAAA,EACF;AACA,QAAM,eAAe,MAAM,OAAO,IAAI;AACtC,QAAM,wBAAwB,OAAO,OAAO,SAAS,GAAI;AAC3D;AAEA,SAAS,UAAU,MAAuB;AACxC,QAAM,YAAY;AAClB,MAAI,OAAO,UAAU,UAAU,WAAY,WAAU,MAAM;AAC7D;AAEA,eAAe,QACb,SACA,OACkC;AAClC,MAAI;AACF,WAAO,MAAM,QAAQ,gBAAgB,KAAK;AAAA,EAC5C,QAAQ;AACN,WAAO;AAAA,EACT;AACF;AAEA,SAAS,WAAW,MAAoC;AACtD,MAAI;AACF,SAAK,WAAW;AAAA,EAClB,QAAQ;AACN;AAAA,EACF;AACF;",
+  "names": []
+}

package/dist/models.js.map

@@ -1,7 +1,7 @@
 {
   "version": 3,
   "sources": ["../src/models.ts"],
-  "sourcesContent": ["import type { BaseModel } from '@earendil-works/pi-ai';\nimport type { AudioFormat } from './audio.ts';\nimport type { RealtimeAudioCost, SpeechAudioCost, TranscriptionAudioCost } from './usage.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  audioCost: TranscriptionAudioCost;\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  audioCost: SpeechAudioCost;\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  audioCost: RealtimeAudioCost;\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": "AAqGA,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;",
+  "sourcesContent": ["import type { BaseModel } from '@earendil-works/pi-ai';\nimport type { AudioFormat } from './audio.ts';\nimport type { RealtimeAudioCost, SpeechAudioCost, TranscriptionAudioCost } from './usage.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  | 'wavespeed';\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  audioCost: TranscriptionAudioCost;\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  audioCost: SpeechAudioCost;\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  audioCost: RealtimeAudioCost;\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": "AAsGA,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/node-host.js

@@ -0,0 +1,132 @@
+import { PassThrough } from "node:stream";
+import { VoiceRequestError } from "./errors.js";
+import { openMixer } from "./mixer.js";
+const TAP_NAME = "pi-audio-tap.js";
+const AUDIO_TAP_CLASSIC = [
+  "class PiAudioTap extends AudioWorkletProcessor {",
+  "  constructor() {",
+  "    super();",
+  "    this.released = false;",
+  "    this.posted = false;",
+  "    var port = this.port;",
+  "    if (port) {",
+  "      port.onmessage = function (event) {",
+  '        if (event.data && event.data.type === "pi-audio-release") this.released = true;',
+  "      }.bind(this);",
+  "    }",
+  "  }",
+  "  process(inputs, outputs) {",
+  "    if (this.released) {",
+  "      if (!this.posted) {",
+  "        this.posted = true;",
+  '        if (this.port) this.port.postMessage({ type: "pi-audio-release" });',
+  "      }",
+  "      return true;",
+  "    }",
+  "    var input = (inputs && inputs[0]) || [];",
+  "    var output = (outputs && outputs[0]) || [];",
+  "    for (var channel = 0; channel < output.length; channel++) {",
+  "      var dst = output[channel];",
+  "      if (!dst) continue;",
+  "      var src = input[channel] || input[0];",
+  "      if (!src) { dst.fill(0); continue; }",
+  "      if (src.length === dst.length) dst.set(src);",
+  "      else {",
+  "        var count = Math.min(src.length, dst.length);",
+  "        for (var frame = 0; frame < count; frame++) dst[frame] = src[frame];",
+  "      }",
+  "    }",
+  "    var heard = input[0];",
+  "    if (heard && heard.length && this.port) {",
+  "      var copy = new Float32Array(heard.length);",
+  "      copy.set(heard);",
+  "      this.port.postMessage({ samples: copy });",
+  "    }",
+  "    return true;",
+  "  }",
+  "}",
+  "registerProcessor('pi-audio-tap', PiAudioTap);",
+  ""
+].join("\n");
+async function loadNodeAudioEngine(engine) {
+  if (engine === "node-web-audio-api") {
+    return await import("node-web-audio-api");
+  }
+  return await import("web-audio-api");
+}
+async function openNodeAudio(options = {}) {
+  const engine = options.engine ?? "auto";
+  const sink = options.sink ?? "none";
+  const load = options.load ?? loadNodeAudioEngine;
+  const names = engine === "auto" ? ["node-web-audio-api", "web-audio-api"] : [engine];
+  const tapUrl = await writeTap();
+  let failure;
+  for (const name of names) {
+    try {
+      return await startEngine(name, sink, options.sampleRate, load, tapUrl);
+    } catch (error) {
+      failure = error;
+      if (engine !== "auto") break;
+    }
+  }
+  throw new VoiceRequestError("error", failureText(failure));
+}
+async function startEngine(engine, sink, sampleRate, load, tapUrl) {
+  const loaded = await load(engine);
+  const built = await nodeContextOptions(engine, sink, sampleRate);
+  const context = new loaded.AudioContext(built.options);
+  try {
+    const out = await openMixer(context, {
+      playback: "webaudio",
+      runtime: "node",
+      tapUrl,
+      workletNode: loaded.AudioWorkletNode
+    });
+    if (context.state === "suspended") await context.resume();
+    return {
+      engine,
+      sink,
+      context,
+      out,
+      close: () => closeSession(context, out, built.stream)
+    };
+  } catch (error) {
+    await closeSession(context, void 0, built.stream);
+    throw error;
+  }
+}
+async function nodeContextOptions(engine, sink, sampleRate) {
+  const options = {};
+  if (sampleRate !== void 0) options.sampleRate = sampleRate;
+  if (sink === "default") return { options };
+  if (engine === "node-web-audio-api") {
+    options.sinkId = { type: "none" };
+    return { options };
+  }
+  const stream = new PassThrough();
+  stream.resume();
+  options.sinkId = stream;
+  return { options, stream };
+}
+async function closeSession(context, out, stream) {
+  if (out) out.close();
+  if (stream) stream.destroy();
+  await context.close();
+}
+function failureText(error) {
+  if (error instanceof Error) return error.message;
+  return "Node audio engine is unavailable";
+}
+async function writeTap() {
+  const { writeFileSync } = await import("node:fs");
+  const { tmpdir } = await import("node:os");
+  const { join } = await import("node:path");
+  const file = join(tmpdir(), TAP_NAME);
+  writeFileSync(file, AUDIO_TAP_CLASSIC);
+  return file;
+}
+export {
+  loadNodeAudioEngine,
+  openNodeAudio
+};
+//# sourceMappingURL=node-host.js.map