@livekit/agents 0.5.2 → 0.6.0

package/dist/tts/tts.cjs.map

@@ -1 +1 @@
-{"version":3,"sources":["../../src/tts/tts.ts"],"sourcesContent":["// SPDX-FileCopyrightText: 2024 LiveKit, Inc.\n//\n// SPDX-License-Identifier: Apache-2.0\nimport type { AudioFrame } from '@livekit/rtc-node';\nimport { AsyncIterableQueue, mergeFrames } from '../utils.js';\n\n/** SynthesizedAudio is a packet of speech synthesis as returned by the TTS. */\nexport interface SynthesizedAudio {\n  /** Request ID (one segment could be made up of multiple requests) */\n  requestId: string;\n  /** Segment ID, each segment is separated by a flush */\n  segmentId: string;\n  /** Synthesized audio frame */\n  frame: AudioFrame;\n  /** Current segment of the synthesized audio */\n  deltaText?: string;\n}\n\n/**\n * Describes the capabilities of the TTS provider.\n *\n * @remarks\n * At present, only `streaming` is supplied to this interface, and the framework only supports\n * providers that do have a streaming endpoint.\n */\nexport interface TTSCapabilities {\n  streaming: boolean;\n}\n\n/**\n * An instance of a text-to-speech adapter.\n *\n * @remarks\n * This class is abstract, and as such cannot be used directly. Instead, use a provider plugin that\n * exports its own child TTS class, which inherits this class's methods.\n */\nexport abstract class TTS {\n  #capabilities: TTSCapabilities;\n  #sampleRate: number;\n  #numChannels: number;\n\n  constructor(sampleRate: number, numChannels: number, capabilities: TTSCapabilities) {\n    this.#capabilities = capabilities;\n    this.#sampleRate = sampleRate;\n    this.#numChannels = numChannels;\n  }\n\n  /** Returns this TTS's capabilities */\n  get capabilities(): TTSCapabilities {\n    return this.#capabilities;\n  }\n\n  /** Returns the sample rate of audio frames returned by this TTS */\n  get sampleRate(): number {\n    return this.#sampleRate;\n  }\n\n  /** Returns the channel count of audio frames returned by this TTS */\n  get numChannels(): number {\n    return this.#numChannels;\n  }\n\n  /**\n   * Receives text and returns synthesis in the form of a {@link ChunkedStream}\n   */\n  abstract synthesize(text: string): ChunkedStream;\n\n  /**\n   * Returns a {@link SynthesizeStream} that can be used to push text and receive audio data\n   */\n  abstract stream(): SynthesizeStream;\n}\n\n/**\n * An instance of a text-to-speech stream, as an asynchronous iterable iterator.\n *\n * @example Looping through frames\n * ```ts\n * for await (const event of stream) {\n *   await source.captureFrame(event.frame);\n * }\n * ```\n *\n * @remarks\n * This class is abstract, and as such cannot be used directly. Instead, use a provider plugin that\n * exports its own child SynthesizeStream class, which inherits this class's methods.\n */\nexport abstract class SynthesizeStream\n  implements AsyncIterableIterator<SynthesizedAudio | typeof SynthesizeStream.END_OF_STREAM>\n{\n  protected static readonly FLUSH_SENTINEL = Symbol('FLUSH_SENTINEL');\n  static readonly END_OF_STREAM = Symbol('END_OF_STREAM');\n  protected input = new AsyncIterableQueue<string | typeof SynthesizeStream.FLUSH_SENTINEL>();\n  protected queue = new AsyncIterableQueue<\n    SynthesizedAudio | typeof SynthesizeStream.END_OF_STREAM\n  >();\n  protected closed = false;\n\n  /** Push a string of text to the TTS */\n  pushText(text: string) {\n    if (this.input.closed) {\n      throw new Error('Input is closed');\n    }\n    if (this.closed) {\n      throw new Error('Stream is closed');\n    }\n    this.input.put(text);\n  }\n\n  /** Flush the TTS, causing it to process all pending text */\n  flush() {\n    if (this.input.closed) {\n      throw new Error('Input is closed');\n    }\n    if (this.closed) {\n      throw new Error('Stream is closed');\n    }\n    this.input.put(SynthesizeStream.FLUSH_SENTINEL);\n  }\n\n  /** Mark the input as ended and forbid additional pushes */\n  endInput() {\n    if (this.input.closed) {\n      throw new Error('Input is closed');\n    }\n    if (this.closed) {\n      throw new Error('Stream is closed');\n    }\n    this.input.close();\n  }\n\n  next(): Promise<IteratorResult<SynthesizedAudio | typeof SynthesizeStream.END_OF_STREAM>> {\n    return this.queue.next();\n  }\n\n  /** Close both the input and output of the TTS stream */\n  close() {\n    this.input.close();\n    this.queue.close();\n    this.closed = true;\n  }\n\n  [Symbol.asyncIterator](): SynthesizeStream {\n    return this;\n  }\n}\n\n/**\n * An instance of a text-to-speech response, as an asynchronous iterable iterator.\n *\n * @example Looping through frames\n * ```ts\n * for await (const event of stream) {\n *   await source.captureFrame(event.frame);\n * }\n * ```\n *\n * @remarks\n * This class is abstract, and as such cannot be used directly. Instead, use a provider plugin that\n * exports its own child ChunkedStream class, which inherits this class's methods.\n */\nexport abstract class ChunkedStream implements AsyncIterableIterator<SynthesizedAudio> {\n  protected queue = new AsyncIterableQueue<SynthesizedAudio>();\n  protected closed = false;\n\n  /** Collect every frame into one in a single call */\n  async collect(): Promise<AudioFrame> {\n    const frames = [];\n    for await (const event of this) {\n      frames.push(event.frame);\n    }\n    return mergeFrames(frames);\n  }\n\n  next(): Promise<IteratorResult<SynthesizedAudio>> {\n    return this.queue.next();\n  }\n\n  /** Close both the input and output of the TTS stream */\n  close() {\n    this.queue.close();\n    this.closed = true;\n  }\n\n  [Symbol.asyncIterator](): ChunkedStream {\n    return this;\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAIA,mBAAgD;AAgCzC,MAAe,IAAI;AAAA,EACxB;AAAA,EACA;AAAA,EACA;AAAA,EAEA,YAAY,YAAoB,aAAqB,cAA+B;AAClF,SAAK,gBAAgB;AACrB,SAAK,cAAc;AACnB,SAAK,eAAe;AAAA,EACtB;AAAA;AAAA,EAGA,IAAI,eAAgC;AAClC,WAAO,KAAK;AAAA,EACd;AAAA;AAAA,EAGA,IAAI,aAAqB;AACvB,WAAO,KAAK;AAAA,EACd;AAAA;AAAA,EAGA,IAAI,cAAsB;AACxB,WAAO,KAAK;AAAA,EACd;AAWF;AAgBO,MAAe,iBAEtB;AAAA,EACE,OAA0B,iBAAiB,OAAO,gBAAgB;AAAA,EAClE,OAAgB,gBAAgB,OAAO,eAAe;AAAA,EAC5C,QAAQ,IAAI,gCAAoE;AAAA,EAChF,QAAQ,IAAI,gCAEpB;AAAA,EACQ,SAAS;AAAA;AAAA,EAGnB,SAAS,MAAc;AACrB,QAAI,KAAK,MAAM,QAAQ;AACrB,YAAM,IAAI,MAAM,iBAAiB;AAAA,IACnC;AACA,QAAI,KAAK,QAAQ;AACf,YAAM,IAAI,MAAM,kBAAkB;AAAA,IACpC;AACA,SAAK,MAAM,IAAI,IAAI;AAAA,EACrB;AAAA;AAAA,EAGA,QAAQ;AACN,QAAI,KAAK,MAAM,QAAQ;AACrB,YAAM,IAAI,MAAM,iBAAiB;AAAA,IACnC;AACA,QAAI,KAAK,QAAQ;AACf,YAAM,IAAI,MAAM,kBAAkB;AAAA,IACpC;AACA,SAAK,MAAM,IAAI,iBAAiB,cAAc;AAAA,EAChD;AAAA;AAAA,EAGA,WAAW;AACT,QAAI,KAAK,MAAM,QAAQ;AACrB,YAAM,IAAI,MAAM,iBAAiB;AAAA,IACnC;AACA,QAAI,KAAK,QAAQ;AACf,YAAM,IAAI,MAAM,kBAAkB;AAAA,IACpC;AACA,SAAK,MAAM,MAAM;AAAA,EACnB;AAAA,EAEA,OAA0F;AACxF,WAAO,KAAK,MAAM,KAAK;AAAA,EACzB;AAAA;AAAA,EAGA,QAAQ;AACN,SAAK,MAAM,MAAM;AACjB,SAAK,MAAM,MAAM;AACjB,SAAK,SAAS;AAAA,EAChB;AAAA,EAEA,CAAC,OAAO,aAAa,IAAsB;AACzC,WAAO;AAAA,EACT;AACF;AAgBO,MAAe,cAAiE;AAAA,EAC3E,QAAQ,IAAI,gCAAqC;AAAA,EACjD,SAAS;AAAA;AAAA,EAGnB,MAAM,UAA+B;AACnC,UAAM,SAAS,CAAC;AAChB,qBAAiB,SAAS,MAAM;AAC9B,aAAO,KAAK,MAAM,KAAK;AAAA,IACzB;AACA,eAAO,0BAAY,MAAM;AAAA,EAC3B;AAAA,EAEA,OAAkD;AAChD,WAAO,KAAK,MAAM,KAAK;AAAA,EACzB;AAAA;AAAA,EAGA,QAAQ;AACN,SAAK,MAAM,MAAM;AACjB,SAAK,SAAS;AAAA,EAChB;AAAA,EAEA,CAAC,OAAO,aAAa,IAAmB;AACtC,WAAO;AAAA,EACT;AACF;","names":[]}
+{"version":3,"sources":["../../src/tts/tts.ts"],"sourcesContent":["// SPDX-FileCopyrightText: 2024 LiveKit, Inc.\n//\n// SPDX-License-Identifier: Apache-2.0\nimport type { AudioFrame } from '@livekit/rtc-node';\nimport type { TypedEventEmitter as TypedEmitter } from '@livekit/typed-emitter';\nimport { EventEmitter } from 'node:events';\nimport type { TTSMetrics } from '../metrics/base.js';\nimport { AsyncIterableQueue, mergeFrames } from '../utils.js';\n\n/** SynthesizedAudio is a packet of speech synthesis as returned by the TTS. */\nexport interface SynthesizedAudio {\n  /** Request ID (one segment could be made up of multiple requests) */\n  requestId: string;\n  /** Segment ID, each segment is separated by a flush */\n  segmentId: string;\n  /** Synthesized audio frame */\n  frame: AudioFrame;\n  /** Current segment of the synthesized audio */\n  deltaText?: string;\n  /** Whether this is the last frame of the segment (streaming only) */\n  final: boolean;\n}\n\n/**\n * Describes the capabilities of the TTS provider.\n *\n * @remarks\n * At present, only `streaming` is supplied to this interface, and the framework only supports\n * providers that do have a streaming endpoint.\n */\nexport interface TTSCapabilities {\n  streaming: boolean;\n}\n\nexport enum TTSEvent {\n  METRICS_COLLECTED,\n}\n\nexport type TTSCallbacks = {\n  [TTSEvent.METRICS_COLLECTED]: (metrics: TTSMetrics) => void;\n};\n\n/**\n * An instance of a text-to-speech adapter.\n *\n * @remarks\n * This class is abstract, and as such cannot be used directly. Instead, use a provider plugin that\n * exports its own child TTS class, which inherits this class's methods.\n */\nexport abstract class TTS extends (EventEmitter as new () => TypedEmitter<TTSCallbacks>) {\n  #capabilities: TTSCapabilities;\n  #sampleRate: number;\n  #numChannels: number;\n  abstract label: string;\n\n  constructor(sampleRate: number, numChannels: number, capabilities: TTSCapabilities) {\n    super();\n    this.#capabilities = capabilities;\n    this.#sampleRate = sampleRate;\n    this.#numChannels = numChannels;\n  }\n\n  /** Returns this TTS's capabilities */\n  get capabilities(): TTSCapabilities {\n    return this.#capabilities;\n  }\n\n  /** Returns the sample rate of audio frames returned by this TTS */\n  get sampleRate(): number {\n    return this.#sampleRate;\n  }\n\n  /** Returns the channel count of audio frames returned by this TTS */\n  get numChannels(): number {\n    return this.#numChannels;\n  }\n\n  /**\n   * Receives text and returns synthesis in the form of a {@link ChunkedStream}\n   */\n  abstract synthesize(text: string): ChunkedStream;\n\n  /**\n   * Returns a {@link SynthesizeStream} that can be used to push text and receive audio data\n   */\n  abstract stream(): SynthesizeStream;\n}\n\n/**\n * An instance of a text-to-speech stream, as an asynchronous iterable iterator.\n *\n * @example Looping through frames\n * ```ts\n * for await (const event of stream) {\n *   await source.captureFrame(event.frame);\n * }\n * ```\n *\n * @remarks\n * This class is abstract, and as such cannot be used directly. Instead, use a provider plugin that\n * exports its own child SynthesizeStream class, which inherits this class's methods.\n */\nexport abstract class SynthesizeStream\n  implements AsyncIterableIterator<SynthesizedAudio | typeof SynthesizeStream.END_OF_STREAM>\n{\n  protected static readonly FLUSH_SENTINEL = Symbol('FLUSH_SENTINEL');\n  static readonly END_OF_STREAM = Symbol('END_OF_STREAM');\n  protected input = new AsyncIterableQueue<string | typeof SynthesizeStream.FLUSH_SENTINEL>();\n  protected queue = new AsyncIterableQueue<\n    SynthesizedAudio | typeof SynthesizeStream.END_OF_STREAM\n  >();\n  protected output = new AsyncIterableQueue<\n    SynthesizedAudio | typeof SynthesizeStream.END_OF_STREAM\n  >();\n  protected closed = false;\n  abstract label: string;\n  #tts: TTS;\n  #metricsPendingTexts: string[] = [];\n  #metricsText = '';\n  #monitorMetricsTask?: Promise<void>;\n\n  constructor(tts: TTS) {\n    this.#tts = tts;\n  }\n\n  protected async monitorMetrics() {\n    const startTime = process.hrtime.bigint();\n    let audioDuration = 0;\n    let ttfb: bigint | undefined;\n    let requestId = '';\n\n    const emit = () => {\n      if (this.#metricsPendingTexts.length) {\n        const text = this.#metricsPendingTexts.shift()!;\n        const duration = process.hrtime.bigint() - startTime;\n        const metrics: TTSMetrics = {\n          timestamp: Date.now(),\n          requestId,\n          ttfb: Math.trunc(Number(ttfb! / BigInt(1000000))),\n          duration: Math.trunc(Number(duration / BigInt(1000000))),\n          charactersCount: text.length,\n          audioDuration,\n          cancelled: false, // XXX(nbsp)\n          label: this.label,\n          streamed: false,\n        };\n        this.#tts.emit(TTSEvent.METRICS_COLLECTED, metrics);\n      }\n    };\n\n    for await (const audio of this.queue) {\n      this.output.put(audio);\n      if (audio === SynthesizeStream.END_OF_STREAM) continue;\n      requestId = audio.requestId;\n      if (!ttfb) {\n        ttfb = process.hrtime.bigint() - startTime;\n      }\n      audioDuration += audio.frame.samplesPerChannel / audio.frame.sampleRate;\n      if (audio.final) {\n        emit();\n      }\n    }\n\n    if (requestId) {\n      emit();\n    }\n    this.output.close();\n  }\n\n  /** Push a string of text to the TTS */\n  pushText(text: string) {\n    if (!this.#monitorMetricsTask) {\n      this.#monitorMetricsTask = this.monitorMetrics();\n    }\n    this.#metricsText += text;\n\n    if (this.input.closed) {\n      throw new Error('Input is closed');\n    }\n    if (this.closed) {\n      throw new Error('Stream is closed');\n    }\n    this.input.put(text);\n  }\n\n  /** Flush the TTS, causing it to process all pending text */\n  flush() {\n    if (this.#metricsText) {\n      this.#metricsPendingTexts.push(this.#metricsText);\n      this.#metricsText = '';\n    }\n    if (this.input.closed) {\n      throw new Error('Input is closed');\n    }\n    if (this.closed) {\n      throw new Error('Stream is closed');\n    }\n    this.input.put(SynthesizeStream.FLUSH_SENTINEL);\n  }\n\n  /** Mark the input as ended and forbid additional pushes */\n  endInput() {\n    if (this.input.closed) {\n      throw new Error('Input is closed');\n    }\n    if (this.closed) {\n      throw new Error('Stream is closed');\n    }\n    this.input.close();\n  }\n\n  next(): Promise<IteratorResult<SynthesizedAudio | typeof SynthesizeStream.END_OF_STREAM>> {\n    return this.output.next();\n  }\n\n  /** Close both the input and output of the TTS stream */\n  close() {\n    this.input.close();\n    this.output.close();\n    this.closed = true;\n  }\n\n  [Symbol.asyncIterator](): SynthesizeStream {\n    return this;\n  }\n}\n\n/**\n * An instance of a text-to-speech response, as an asynchronous iterable iterator.\n *\n * @example Looping through frames\n * ```ts\n * for await (const event of stream) {\n *   await source.captureFrame(event.frame);\n * }\n * ```\n *\n * @remarks\n * This class is abstract, and as such cannot be used directly. Instead, use a provider plugin that\n * exports its own child ChunkedStream class, which inherits this class's methods.\n */\nexport abstract class ChunkedStream implements AsyncIterableIterator<SynthesizedAudio> {\n  protected queue = new AsyncIterableQueue<SynthesizedAudio>();\n  protected output = new AsyncIterableQueue<SynthesizedAudio>();\n  protected closed = false;\n  abstract label: string;\n  #text: string;\n  #tts: TTS;\n\n  constructor(text: string, tts: TTS) {\n    this.#text = text;\n    this.#tts = tts;\n\n    this.monitorMetrics();\n  }\n\n  protected async monitorMetrics() {\n    const startTime = process.hrtime.bigint();\n    let audioDuration = 0;\n    let ttfb: bigint | undefined;\n    let requestId = '';\n\n    for await (const audio of this.queue) {\n      this.output.put(audio);\n      requestId = audio.requestId;\n      if (!ttfb) {\n        ttfb = process.hrtime.bigint() - startTime;\n      }\n      audioDuration += audio.frame.samplesPerChannel / audio.frame.sampleRate;\n    }\n    this.output.close();\n\n    const duration = process.hrtime.bigint() - startTime;\n    const metrics: TTSMetrics = {\n      timestamp: Date.now(),\n      requestId,\n      ttfb: Math.trunc(Number(ttfb! / BigInt(1000000))),\n      duration: Math.trunc(Number(duration / BigInt(1000000))),\n      charactersCount: this.#text.length,\n      audioDuration,\n      cancelled: false, // XXX(nbsp)\n      label: this.label,\n      streamed: false,\n    };\n    this.#tts.emit(TTSEvent.METRICS_COLLECTED, metrics);\n  }\n\n  /** Collect every frame into one in a single call */\n  async collect(): Promise<AudioFrame> {\n    const frames = [];\n    for await (const event of this) {\n      frames.push(event.frame);\n    }\n    return mergeFrames(frames);\n  }\n\n  next(): Promise<IteratorResult<SynthesizedAudio>> {\n    return this.output.next();\n  }\n\n  /** Close both the input and output of the TTS stream */\n  close() {\n    this.queue.close();\n    this.output.close();\n    this.closed = true;\n  }\n\n  [Symbol.asyncIterator](): ChunkedStream {\n    return this;\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAKA,yBAA6B;AAE7B,mBAAgD;AA2BzC,IAAK,WAAL,kBAAKA,cAAL;AACL,EAAAA,oBAAA;AADU,SAAAA;AAAA,GAAA;AAeL,MAAe,YAAa,gCAAsD;AAAA,EACvF;AAAA,EACA;AAAA,EACA;AAAA,EAGA,YAAY,YAAoB,aAAqB,cAA+B;AAClF,UAAM;AACN,SAAK,gBAAgB;AACrB,SAAK,cAAc;AACnB,SAAK,eAAe;AAAA,EACtB;AAAA;AAAA,EAGA,IAAI,eAAgC;AAClC,WAAO,KAAK;AAAA,EACd;AAAA;AAAA,EAGA,IAAI,aAAqB;AACvB,WAAO,KAAK;AAAA,EACd;AAAA;AAAA,EAGA,IAAI,cAAsB;AACxB,WAAO,KAAK;AAAA,EACd;AAWF;AAgBO,MAAe,iBAEtB;AAAA,EACE,OAA0B,iBAAiB,OAAO,gBAAgB;AAAA,EAClE,OAAgB,gBAAgB,OAAO,eAAe;AAAA,EAC5C,QAAQ,IAAI,gCAAoE;AAAA,EAChF,QAAQ,IAAI,gCAEpB;AAAA,EACQ,SAAS,IAAI,gCAErB;AAAA,EACQ,SAAS;AAAA,EAEnB;AAAA,EACA,uBAAiC,CAAC;AAAA,EAClC,eAAe;AAAA,EACf;AAAA,EAEA,YAAY,KAAU;AACpB,SAAK,OAAO;AAAA,EACd;AAAA,EAEA,MAAgB,iBAAiB;AAC/B,UAAM,YAAY,QAAQ,OAAO,OAAO;AACxC,QAAI,gBAAgB;AACpB,QAAI;AACJ,QAAI,YAAY;AAEhB,UAAM,OAAO,MAAM;AACjB,UAAI,KAAK,qBAAqB,QAAQ;AACpC,cAAM,OAAO,KAAK,qBAAqB,MAAM;AAC7C,cAAM,WAAW,QAAQ,OAAO,OAAO,IAAI;AAC3C,cAAM,UAAsB;AAAA,UAC1B,WAAW,KAAK,IAAI;AAAA,UACpB;AAAA,UACA,MAAM,KAAK,MAAM,OAAO,OAAQ,OAAO,GAAO,CAAC,CAAC;AAAA,UAChD,UAAU,KAAK,MAAM,OAAO,WAAW,OAAO,GAAO,CAAC,CAAC;AAAA,UACvD,iBAAiB,KAAK;AAAA,UACtB;AAAA,UACA,WAAW;AAAA;AAAA,UACX,OAAO,KAAK;AAAA,UACZ,UAAU;AAAA,QACZ;AACA,aAAK,KAAK,KAAK,2BAA4B,OAAO;AAAA,MACpD;AAAA,IACF;AAEA,qBAAiB,SAAS,KAAK,OAAO;AACpC,WAAK,OAAO,IAAI,KAAK;AACrB,UAAI,UAAU,iBAAiB,cAAe;AAC9C,kBAAY,MAAM;AAClB,UAAI,CAAC,MAAM;AACT,eAAO,QAAQ,OAAO,OAAO,IAAI;AAAA,MACnC;AACA,uBAAiB,MAAM,MAAM,oBAAoB,MAAM,MAAM;AAC7D,UAAI,MAAM,OAAO;AACf,aAAK;AAAA,MACP;AAAA,IACF;AAEA,QAAI,WAAW;AACb,WAAK;AAAA,IACP;AACA,SAAK,OAAO,MAAM;AAAA,EACpB;AAAA;AAAA,EAGA,SAAS,MAAc;AACrB,QAAI,CAAC,KAAK,qBAAqB;AAC7B,WAAK,sBAAsB,KAAK,eAAe;AAAA,IACjD;AACA,SAAK,gBAAgB;AAErB,QAAI,KAAK,MAAM,QAAQ;AACrB,YAAM,IAAI,MAAM,iBAAiB;AAAA,IACnC;AACA,QAAI,KAAK,QAAQ;AACf,YAAM,IAAI,MAAM,kBAAkB;AAAA,IACpC;AACA,SAAK,MAAM,IAAI,IAAI;AAAA,EACrB;AAAA;AAAA,EAGA,QAAQ;AACN,QAAI,KAAK,cAAc;AACrB,WAAK,qBAAqB,KAAK,KAAK,YAAY;AAChD,WAAK,eAAe;AAAA,IACtB;AACA,QAAI,KAAK,MAAM,QAAQ;AACrB,YAAM,IAAI,MAAM,iBAAiB;AAAA,IACnC;AACA,QAAI,KAAK,QAAQ;AACf,YAAM,IAAI,MAAM,kBAAkB;AAAA,IACpC;AACA,SAAK,MAAM,IAAI,iBAAiB,cAAc;AAAA,EAChD;AAAA;AAAA,EAGA,WAAW;AACT,QAAI,KAAK,MAAM,QAAQ;AACrB,YAAM,IAAI,MAAM,iBAAiB;AAAA,IACnC;AACA,QAAI,KAAK,QAAQ;AACf,YAAM,IAAI,MAAM,kBAAkB;AAAA,IACpC;AACA,SAAK,MAAM,MAAM;AAAA,EACnB;AAAA,EAEA,OAA0F;AACxF,WAAO,KAAK,OAAO,KAAK;AAAA,EAC1B;AAAA;AAAA,EAGA,QAAQ;AACN,SAAK,MAAM,MAAM;AACjB,SAAK,OAAO,MAAM;AAClB,SAAK,SAAS;AAAA,EAChB;AAAA,EAEA,CAAC,OAAO,aAAa,IAAsB;AACzC,WAAO;AAAA,EACT;AACF;AAgBO,MAAe,cAAiE;AAAA,EAC3E,QAAQ,IAAI,gCAAqC;AAAA,EACjD,SAAS,IAAI,gCAAqC;AAAA,EAClD,SAAS;AAAA,EAEnB;AAAA,EACA;AAAA,EAEA,YAAY,MAAc,KAAU;AAClC,SAAK,QAAQ;AACb,SAAK,OAAO;AAEZ,SAAK,eAAe;AAAA,EACtB;AAAA,EAEA,MAAgB,iBAAiB;AAC/B,UAAM,YAAY,QAAQ,OAAO,OAAO;AACxC,QAAI,gBAAgB;AACpB,QAAI;AACJ,QAAI,YAAY;AAEhB,qBAAiB,SAAS,KAAK,OAAO;AACpC,WAAK,OAAO,IAAI,KAAK;AACrB,kBAAY,MAAM;AAClB,UAAI,CAAC,MAAM;AACT,eAAO,QAAQ,OAAO,OAAO,IAAI;AAAA,MACnC;AACA,uBAAiB,MAAM,MAAM,oBAAoB,MAAM,MAAM;AAAA,IAC/D;AACA,SAAK,OAAO,MAAM;AAElB,UAAM,WAAW,QAAQ,OAAO,OAAO,IAAI;AAC3C,UAAM,UAAsB;AAAA,MAC1B,WAAW,KAAK,IAAI;AAAA,MACpB;AAAA,MACA,MAAM,KAAK,MAAM,OAAO,OAAQ,OAAO,GAAO,CAAC,CAAC;AAAA,MAChD,UAAU,KAAK,MAAM,OAAO,WAAW,OAAO,GAAO,CAAC,CAAC;AAAA,MACvD,iBAAiB,KAAK,MAAM;AAAA,MAC5B;AAAA,MACA,WAAW;AAAA;AAAA,MACX,OAAO,KAAK;AAAA,MACZ,UAAU;AAAA,IACZ;AACA,SAAK,KAAK,KAAK,2BAA4B,OAAO;AAAA,EACpD;AAAA;AAAA,EAGA,MAAM,UAA+B;AACnC,UAAM,SAAS,CAAC;AAChB,qBAAiB,SAAS,MAAM;AAC9B,aAAO,KAAK,MAAM,KAAK;AAAA,IACzB;AACA,eAAO,0BAAY,MAAM;AAAA,EAC3B;AAAA,EAEA,OAAkD;AAChD,WAAO,KAAK,OAAO,KAAK;AAAA,EAC1B;AAAA;AAAA,EAGA,QAAQ;AACN,SAAK,MAAM,MAAM;AACjB,SAAK,OAAO,MAAM;AAClB,SAAK,SAAS;AAAA,EAChB;AAAA,EAEA,CAAC,OAAO,aAAa,IAAmB;AACtC,WAAO;AAAA,EACT;AACF;","names":["TTSEvent"]}

package/dist/tts/tts.d.ts

@@ -1,4 +1,6 @@
 import type { AudioFrame } from '@livekit/rtc-node';
+import type { TypedEventEmitter as TypedEmitter } from '@livekit/typed-emitter';
+import type { TTSMetrics } from '../metrics/base.js';
 import { AsyncIterableQueue } from '../utils.js';
 /** SynthesizedAudio is a packet of speech synthesis as returned by the TTS. */
 export interface SynthesizedAudio {
@@ -10,6 +12,8 @@ export interface SynthesizedAudio {
     frame: AudioFrame;
     /** Current segment of the synthesized audio */
     deltaText?: string;
+    /** Whether this is the last frame of the segment (streaming only) */
+    final: boolean;
 }
 /**
  * Describes the capabilities of the TTS provider.
@@ -21,6 +25,13 @@ export interface SynthesizedAudio {
 export interface TTSCapabilities {
     streaming: boolean;
 }
+export declare enum TTSEvent {
+    METRICS_COLLECTED = 0
+}
+export type TTSCallbacks = {
+    [TTSEvent.METRICS_COLLECTED]: (metrics: TTSMetrics) => void;
+};
+declare const TTS_base: new () => TypedEmitter<TTSCallbacks>;
 /**
  * An instance of a text-to-speech adapter.
  *
@@ -28,8 +39,9 @@ export interface TTSCapabilities {
  * This class is abstract, and as such cannot be used directly. Instead, use a provider plugin that
  * exports its own child TTS class, which inherits this class's methods.
  */
-export declare abstract class TTS {
+export declare abstract class TTS extends TTS_base {
     #private;
+    abstract label: string;
     constructor(sampleRate: number, numChannels: number, capabilities: TTSCapabilities);
     /** Returns this TTS's capabilities */
     get capabilities(): TTSCapabilities;
@@ -61,11 +73,16 @@ export declare abstract class TTS {
  * exports its own child SynthesizeStream class, which inherits this class's methods.
  */
 export declare abstract class SynthesizeStream implements AsyncIterableIterator<SynthesizedAudio | typeof SynthesizeStream.END_OF_STREAM> {
+    #private;
     protected static readonly FLUSH_SENTINEL: unique symbol;
     static readonly END_OF_STREAM: unique symbol;
     protected input: AsyncIterableQueue<string | typeof SynthesizeStream.FLUSH_SENTINEL>;
     protected queue: AsyncIterableQueue<SynthesizedAudio | typeof SynthesizeStream.END_OF_STREAM>;
+    protected output: AsyncIterableQueue<SynthesizedAudio | typeof SynthesizeStream.END_OF_STREAM>;
     protected closed: boolean;
+    abstract label: string;
+    constructor(tts: TTS);
+    protected monitorMetrics(): Promise<void>;
     /** Push a string of text to the TTS */
     pushText(text: string): void;
     /** Flush the TTS, causing it to process all pending text */
@@ -92,8 +109,13 @@ export declare abstract class SynthesizeStream implements AsyncIterableIterator<
  * exports its own child ChunkedStream class, which inherits this class's methods.
  */
 export declare abstract class ChunkedStream implements AsyncIterableIterator<SynthesizedAudio> {
+    #private;
     protected queue: AsyncIterableQueue<SynthesizedAudio>;
+    protected output: AsyncIterableQueue<SynthesizedAudio>;
     protected closed: boolean;
+    abstract label: string;
+    constructor(text: string, tts: TTS);
+    protected monitorMetrics(): Promise<void>;
     /** Collect every frame into one in a single call */
     collect(): Promise<AudioFrame>;
     next(): Promise<IteratorResult<SynthesizedAudio>>;
@@ -101,4 +123,5 @@ export declare abstract class ChunkedStream implements AsyncIterableIterator<Syn
     close(): void;
     [Symbol.asyncIterator](): ChunkedStream;
 }
+export {};
 //# sourceMappingURL=tts.d.ts.map

package/dist/tts/tts.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"tts.d.ts","sourceRoot":"","sources":["../../src/tts/tts.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,EAAE,kBAAkB,EAAe,MAAM,aAAa,CAAC;AAE9D,+EAA+E;AAC/E,MAAM,WAAW,gBAAgB;IAC/B,qEAAqE;IACrE,SAAS,EAAE,MAAM,CAAC;IAClB,uDAAuD;IACvD,SAAS,EAAE,MAAM,CAAC;IAClB,8BAA8B;IAC9B,KAAK,EAAE,UAAU,CAAC;IAClB,+CAA+C;IAC/C,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED;;;;;;GAMG;AACH,MAAM,WAAW,eAAe;IAC9B,SAAS,EAAE,OAAO,CAAC;CACpB;AAED;;;;;;GAMG;AACH,8BAAsB,GAAG;;gBAKX,UAAU,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,YAAY,EAAE,eAAe;IAMlF,sCAAsC;IACtC,IAAI,YAAY,IAAI,eAAe,CAElC;IAED,mEAAmE;IACnE,IAAI,UAAU,IAAI,MAAM,CAEvB;IAED,qEAAqE;IACrE,IAAI,WAAW,IAAI,MAAM,CAExB;IAED;;OAEG;IACH,QAAQ,CAAC,UAAU,CAAC,IAAI,EAAE,MAAM,GAAG,aAAa;IAEhD;;OAEG;IACH,QAAQ,CAAC,MAAM,IAAI,gBAAgB;CACpC;AAED;;;;;;;;;;;;;GAaG;AACH,8BAAsB,gBACpB,YAAW,qBAAqB,CAAC,gBAAgB,GAAG,OAAO,gBAAgB,CAAC,aAAa,CAAC;IAE1F,SAAS,CAAC,MAAM,CAAC,QAAQ,CAAC,cAAc,gBAA4B;IACpE,MAAM,CAAC,QAAQ,CAAC,aAAa,gBAA2B;IACxD,SAAS,CAAC,KAAK,sEAA6E;IAC5F,SAAS,CAAC,KAAK,+EAEX;IACJ,SAAS,CAAC,MAAM,UAAS;IAEzB,uCAAuC;IACvC,QAAQ,CAAC,IAAI,EAAE,MAAM;IAUrB,4DAA4D;IAC5D,KAAK;IAUL,2DAA2D;IAC3D,QAAQ;IAUR,IAAI,IAAI,OAAO,CAAC,cAAc,CAAC,gBAAgB,GAAG,OAAO,gBAAgB,CAAC,aAAa,CAAC,CAAC;IAIzF,wDAAwD;IACxD,KAAK;IAML,CAAC,MAAM,CAAC,aAAa,CAAC,IAAI,gBAAgB;CAG3C;AAED;;;;;;;;;;;;;GAaG;AACH,8BAAsB,aAAc,YAAW,qBAAqB,CAAC,gBAAgB,CAAC;IACpF,SAAS,CAAC,KAAK,uCAA8C;IAC7D,SAAS,CAAC,MAAM,UAAS;IAEzB,oDAAoD;IAC9C,OAAO,IAAI,OAAO,CAAC,UAAU,CAAC;IAQpC,IAAI,IAAI,OAAO,CAAC,cAAc,CAAC,gBAAgB,CAAC,CAAC;IAIjD,wDAAwD;IACxD,KAAK;IAKL,CAAC,MAAM,CAAC,aAAa,CAAC,IAAI,aAAa;CAGxC"}
+{"version":3,"file":"tts.d.ts","sourceRoot":"","sources":["../../src/tts/tts.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,KAAK,EAAE,iBAAiB,IAAI,YAAY,EAAE,MAAM,wBAAwB,CAAC;AAEhF,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,oBAAoB,CAAC;AACrD,OAAO,EAAE,kBAAkB,EAAe,MAAM,aAAa,CAAC;AAE9D,+EAA+E;AAC/E,MAAM,WAAW,gBAAgB;IAC/B,qEAAqE;IACrE,SAAS,EAAE,MAAM,CAAC;IAClB,uDAAuD;IACvD,SAAS,EAAE,MAAM,CAAC;IAClB,8BAA8B;IAC9B,KAAK,EAAE,UAAU,CAAC;IAClB,+CAA+C;IAC/C,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,qEAAqE;IACrE,KAAK,EAAE,OAAO,CAAC;CAChB;AAED;;;;;;GAMG;AACH,MAAM,WAAW,eAAe;IAC9B,SAAS,EAAE,OAAO,CAAC;CACpB;AAED,oBAAY,QAAQ;IAClB,iBAAiB,IAAA;CAClB;AAED,MAAM,MAAM,YAAY,GAAG;IACzB,CAAC,QAAQ,CAAC,iBAAiB,CAAC,EAAE,CAAC,OAAO,EAAE,UAAU,KAAK,IAAI,CAAC;CAC7D,CAAC;kCAS2D,aAAa,YAAY,CAAC;AAPvF;;;;;;GAMG;AACH,8BAAsB,GAAI,SAAQ,QAAsD;;IAItF,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;gBAEX,UAAU,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,YAAY,EAAE,eAAe;IAOlF,sCAAsC;IACtC,IAAI,YAAY,IAAI,eAAe,CAElC;IAED,mEAAmE;IACnE,IAAI,UAAU,IAAI,MAAM,CAEvB;IAED,qEAAqE;IACrE,IAAI,WAAW,IAAI,MAAM,CAExB;IAED;;OAEG;IACH,QAAQ,CAAC,UAAU,CAAC,IAAI,EAAE,MAAM,GAAG,aAAa;IAEhD;;OAEG;IACH,QAAQ,CAAC,MAAM,IAAI,gBAAgB;CACpC;AAED;;;;;;;;;;;;;GAaG;AACH,8BAAsB,gBACpB,YAAW,qBAAqB,CAAC,gBAAgB,GAAG,OAAO,gBAAgB,CAAC,aAAa,CAAC;;IAE1F,SAAS,CAAC,MAAM,CAAC,QAAQ,CAAC,cAAc,gBAA4B;IACpE,MAAM,CAAC,QAAQ,CAAC,aAAa,gBAA2B;IACxD,SAAS,CAAC,KAAK,sEAA6E;IAC5F,SAAS,CAAC,KAAK,+EAEX;IACJ,SAAS,CAAC,MAAM,+EAEZ;IACJ,SAAS,CAAC,MAAM,UAAS;IACzB,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;gBAMX,GAAG,EAAE,GAAG;cAIJ,cAAc;IA4C9B,uCAAuC;IACvC,QAAQ,CAAC,IAAI,EAAE,MAAM;IAerB,4DAA4D;IAC5D,KAAK;IAcL,2DAA2D;IAC3D,QAAQ;IAUR,IAAI,IAAI,OAAO,CAAC,cAAc,CAAC,gBAAgB,GAAG,OAAO,gBAAgB,CAAC,aAAa,CAAC,CAAC;IAIzF,wDAAwD;IACxD,KAAK;IAML,CAAC,MAAM,CAAC,aAAa,CAAC,IAAI,gBAAgB;CAG3C;AAED;;;;;;;;;;;;;GAaG;AACH,8BAAsB,aAAc,YAAW,qBAAqB,CAAC,gBAAgB,CAAC;;IACpF,SAAS,CAAC,KAAK,uCAA8C;IAC7D,SAAS,CAAC,MAAM,uCAA8C;IAC9D,SAAS,CAAC,MAAM,UAAS;IACzB,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;gBAIX,IAAI,EAAE,MAAM,EAAE,GAAG,EAAE,GAAG;cAOlB,cAAc;IA+B9B,oDAAoD;IAC9C,OAAO,IAAI,OAAO,CAAC,UAAU,CAAC;IAQpC,IAAI,IAAI,OAAO,CAAC,cAAc,CAAC,gBAAgB,CAAC,CAAC;IAIjD,wDAAwD;IACxD,KAAK;IAML,CAAC,MAAM,CAAC,aAAa,CAAC,IAAI,aAAa;CAGxC"}

package/dist/tts/tts.js

@@ -1,9 +1,15 @@
+import { EventEmitter } from "node:events";
 import { AsyncIterableQueue, mergeFrames } from "../utils.js";
-class TTS {
+var TTSEvent = /* @__PURE__ */ ((TTSEvent2) => {
+  TTSEvent2[TTSEvent2["METRICS_COLLECTED"] = 0] = "METRICS_COLLECTED";
+  return TTSEvent2;
+})(TTSEvent || {});
+class TTS extends EventEmitter {
   #capabilities;
   #sampleRate;
   #numChannels;
   constructor(sampleRate, numChannels, capabilities) {
+    super();
     this.#capabilities = capabilities;
     this.#sampleRate = sampleRate;
     this.#numChannels = numChannels;
@@ -26,9 +32,62 @@ class SynthesizeStream {
   static END_OF_STREAM = Symbol("END_OF_STREAM");
   input = new AsyncIterableQueue();
   queue = new AsyncIterableQueue();
+  output = new AsyncIterableQueue();
   closed = false;
+  #tts;
+  #metricsPendingTexts = [];
+  #metricsText = "";
+  #monitorMetricsTask;
+  constructor(tts) {
+    this.#tts = tts;
+  }
+  async monitorMetrics() {
+    const startTime = process.hrtime.bigint();
+    let audioDuration = 0;
+    let ttfb;
+    let requestId = "";
+    const emit = () => {
+      if (this.#metricsPendingTexts.length) {
+        const text = this.#metricsPendingTexts.shift();
+        const duration = process.hrtime.bigint() - startTime;
+        const metrics = {
+          timestamp: Date.now(),
+          requestId,
+          ttfb: Math.trunc(Number(ttfb / BigInt(1e6))),
+          duration: Math.trunc(Number(duration / BigInt(1e6))),
+          charactersCount: text.length,
+          audioDuration,
+          cancelled: false,
+          // XXX(nbsp)
+          label: this.label,
+          streamed: false
+        };
+        this.#tts.emit(0 /* METRICS_COLLECTED */, metrics);
+      }
+    };
+    for await (const audio of this.queue) {
+      this.output.put(audio);
+      if (audio === SynthesizeStream.END_OF_STREAM) continue;
+      requestId = audio.requestId;
+      if (!ttfb) {
+        ttfb = process.hrtime.bigint() - startTime;
+      }
+      audioDuration += audio.frame.samplesPerChannel / audio.frame.sampleRate;
+      if (audio.final) {
+        emit();
+      }
+    }
+    if (requestId) {
+      emit();
+    }
+    this.output.close();
+  }
   /** Push a string of text to the TTS */
   pushText(text) {
+    if (!this.#monitorMetricsTask) {
+      this.#monitorMetricsTask = this.monitorMetrics();
+    }
+    this.#metricsText += text;
     if (this.input.closed) {
       throw new Error("Input is closed");
     }
@@ -39,6 +98,10 @@ class SynthesizeStream {
   }
   /** Flush the TTS, causing it to process all pending text */
   flush() {
+    if (this.#metricsText) {
+      this.#metricsPendingTexts.push(this.#metricsText);
+      this.#metricsText = "";
+    }
     if (this.input.closed) {
       throw new Error("Input is closed");
     }
@@ -58,12 +121,12 @@ class SynthesizeStream {
     this.input.close();
   }
   next() {
-    return this.queue.next();
+    return this.output.next();
   }
   /** Close both the input and output of the TTS stream */
   close() {
     this.input.close();
-    this.queue.close();
+    this.output.close();
     this.closed = true;
   }
   [Symbol.asyncIterator]() {
@@ -72,7 +135,44 @@ class SynthesizeStream {
 }
 class ChunkedStream {
   queue = new AsyncIterableQueue();
+  output = new AsyncIterableQueue();
   closed = false;
+  #text;
+  #tts;
+  constructor(text, tts) {
+    this.#text = text;
+    this.#tts = tts;
+    this.monitorMetrics();
+  }
+  async monitorMetrics() {
+    const startTime = process.hrtime.bigint();
+    let audioDuration = 0;
+    let ttfb;
+    let requestId = "";
+    for await (const audio of this.queue) {
+      this.output.put(audio);
+      requestId = audio.requestId;
+      if (!ttfb) {
+        ttfb = process.hrtime.bigint() - startTime;
+      }
+      audioDuration += audio.frame.samplesPerChannel / audio.frame.sampleRate;
+    }
+    this.output.close();
+    const duration = process.hrtime.bigint() - startTime;
+    const metrics = {
+      timestamp: Date.now(),
+      requestId,
+      ttfb: Math.trunc(Number(ttfb / BigInt(1e6))),
+      duration: Math.trunc(Number(duration / BigInt(1e6))),
+      charactersCount: this.#text.length,
+      audioDuration,
+      cancelled: false,
+      // XXX(nbsp)
+      label: this.label,
+      streamed: false
+    };
+    this.#tts.emit(0 /* METRICS_COLLECTED */, metrics);
+  }
   /** Collect every frame into one in a single call */
   async collect() {
     const frames = [];
@@ -82,11 +182,12 @@ class ChunkedStream {
     return mergeFrames(frames);
   }
   next() {
-    return this.queue.next();
+    return this.output.next();
   }
   /** Close both the input and output of the TTS stream */
   close() {
     this.queue.close();
+    this.output.close();
     this.closed = true;
   }
   [Symbol.asyncIterator]() {
@@ -96,6 +197,7 @@ class ChunkedStream {
 export {
   ChunkedStream,
   SynthesizeStream,
-  TTS
+  TTS,
+  TTSEvent
 };
 //# sourceMappingURL=tts.js.map

package/dist/tts/tts.js.map

@@ -1 +1 @@
-{"version":3,"sources":["../../src/tts/tts.ts"],"sourcesContent":["// SPDX-FileCopyrightText: 2024 LiveKit, Inc.\n//\n// SPDX-License-Identifier: Apache-2.0\nimport type { AudioFrame } from '@livekit/rtc-node';\nimport { AsyncIterableQueue, mergeFrames } from '../utils.js';\n\n/** SynthesizedAudio is a packet of speech synthesis as returned by the TTS. */\nexport interface SynthesizedAudio {\n  /** Request ID (one segment could be made up of multiple requests) */\n  requestId: string;\n  /** Segment ID, each segment is separated by a flush */\n  segmentId: string;\n  /** Synthesized audio frame */\n  frame: AudioFrame;\n  /** Current segment of the synthesized audio */\n  deltaText?: string;\n}\n\n/**\n * Describes the capabilities of the TTS provider.\n *\n * @remarks\n * At present, only `streaming` is supplied to this interface, and the framework only supports\n * providers that do have a streaming endpoint.\n */\nexport interface TTSCapabilities {\n  streaming: boolean;\n}\n\n/**\n * An instance of a text-to-speech adapter.\n *\n * @remarks\n * This class is abstract, and as such cannot be used directly. Instead, use a provider plugin that\n * exports its own child TTS class, which inherits this class's methods.\n */\nexport abstract class TTS {\n  #capabilities: TTSCapabilities;\n  #sampleRate: number;\n  #numChannels: number;\n\n  constructor(sampleRate: number, numChannels: number, capabilities: TTSCapabilities) {\n    this.#capabilities = capabilities;\n    this.#sampleRate = sampleRate;\n    this.#numChannels = numChannels;\n  }\n\n  /** Returns this TTS's capabilities */\n  get capabilities(): TTSCapabilities {\n    return this.#capabilities;\n  }\n\n  /** Returns the sample rate of audio frames returned by this TTS */\n  get sampleRate(): number {\n    return this.#sampleRate;\n  }\n\n  /** Returns the channel count of audio frames returned by this TTS */\n  get numChannels(): number {\n    return this.#numChannels;\n  }\n\n  /**\n   * Receives text and returns synthesis in the form of a {@link ChunkedStream}\n   */\n  abstract synthesize(text: string): ChunkedStream;\n\n  /**\n   * Returns a {@link SynthesizeStream} that can be used to push text and receive audio data\n   */\n  abstract stream(): SynthesizeStream;\n}\n\n/**\n * An instance of a text-to-speech stream, as an asynchronous iterable iterator.\n *\n * @example Looping through frames\n * ```ts\n * for await (const event of stream) {\n *   await source.captureFrame(event.frame);\n * }\n * ```\n *\n * @remarks\n * This class is abstract, and as such cannot be used directly. Instead, use a provider plugin that\n * exports its own child SynthesizeStream class, which inherits this class's methods.\n */\nexport abstract class SynthesizeStream\n  implements AsyncIterableIterator<SynthesizedAudio | typeof SynthesizeStream.END_OF_STREAM>\n{\n  protected static readonly FLUSH_SENTINEL = Symbol('FLUSH_SENTINEL');\n  static readonly END_OF_STREAM = Symbol('END_OF_STREAM');\n  protected input = new AsyncIterableQueue<string | typeof SynthesizeStream.FLUSH_SENTINEL>();\n  protected queue = new AsyncIterableQueue<\n    SynthesizedAudio | typeof SynthesizeStream.END_OF_STREAM\n  >();\n  protected closed = false;\n\n  /** Push a string of text to the TTS */\n  pushText(text: string) {\n    if (this.input.closed) {\n      throw new Error('Input is closed');\n    }\n    if (this.closed) {\n      throw new Error('Stream is closed');\n    }\n    this.input.put(text);\n  }\n\n  /** Flush the TTS, causing it to process all pending text */\n  flush() {\n    if (this.input.closed) {\n      throw new Error('Input is closed');\n    }\n    if (this.closed) {\n      throw new Error('Stream is closed');\n    }\n    this.input.put(SynthesizeStream.FLUSH_SENTINEL);\n  }\n\n  /** Mark the input as ended and forbid additional pushes */\n  endInput() {\n    if (this.input.closed) {\n      throw new Error('Input is closed');\n    }\n    if (this.closed) {\n      throw new Error('Stream is closed');\n    }\n    this.input.close();\n  }\n\n  next(): Promise<IteratorResult<SynthesizedAudio | typeof SynthesizeStream.END_OF_STREAM>> {\n    return this.queue.next();\n  }\n\n  /** Close both the input and output of the TTS stream */\n  close() {\n    this.input.close();\n    this.queue.close();\n    this.closed = true;\n  }\n\n  [Symbol.asyncIterator](): SynthesizeStream {\n    return this;\n  }\n}\n\n/**\n * An instance of a text-to-speech response, as an asynchronous iterable iterator.\n *\n * @example Looping through frames\n * ```ts\n * for await (const event of stream) {\n *   await source.captureFrame(event.frame);\n * }\n * ```\n *\n * @remarks\n * This class is abstract, and as such cannot be used directly. Instead, use a provider plugin that\n * exports its own child ChunkedStream class, which inherits this class's methods.\n */\nexport abstract class ChunkedStream implements AsyncIterableIterator<SynthesizedAudio> {\n  protected queue = new AsyncIterableQueue<SynthesizedAudio>();\n  protected closed = false;\n\n  /** Collect every frame into one in a single call */\n  async collect(): Promise<AudioFrame> {\n    const frames = [];\n    for await (const event of this) {\n      frames.push(event.frame);\n    }\n    return mergeFrames(frames);\n  }\n\n  next(): Promise<IteratorResult<SynthesizedAudio>> {\n    return this.queue.next();\n  }\n\n  /** Close both the input and output of the TTS stream */\n  close() {\n    this.queue.close();\n    this.closed = true;\n  }\n\n  [Symbol.asyncIterator](): ChunkedStream {\n    return this;\n  }\n}\n"],"mappings":"AAIA,SAAS,oBAAoB,mBAAmB;AAgCzC,MAAe,IAAI;AAAA,EACxB;AAAA,EACA;AAAA,EACA;AAAA,EAEA,YAAY,YAAoB,aAAqB,cAA+B;AAClF,SAAK,gBAAgB;AACrB,SAAK,cAAc;AACnB,SAAK,eAAe;AAAA,EACtB;AAAA;AAAA,EAGA,IAAI,eAAgC;AAClC,WAAO,KAAK;AAAA,EACd;AAAA;AAAA,EAGA,IAAI,aAAqB;AACvB,WAAO,KAAK;AAAA,EACd;AAAA;AAAA,EAGA,IAAI,cAAsB;AACxB,WAAO,KAAK;AAAA,EACd;AAWF;AAgBO,MAAe,iBAEtB;AAAA,EACE,OAA0B,iBAAiB,OAAO,gBAAgB;AAAA,EAClE,OAAgB,gBAAgB,OAAO,eAAe;AAAA,EAC5C,QAAQ,IAAI,mBAAoE;AAAA,EAChF,QAAQ,IAAI,mBAEpB;AAAA,EACQ,SAAS;AAAA;AAAA,EAGnB,SAAS,MAAc;AACrB,QAAI,KAAK,MAAM,QAAQ;AACrB,YAAM,IAAI,MAAM,iBAAiB;AAAA,IACnC;AACA,QAAI,KAAK,QAAQ;AACf,YAAM,IAAI,MAAM,kBAAkB;AAAA,IACpC;AACA,SAAK,MAAM,IAAI,IAAI;AAAA,EACrB;AAAA;AAAA,EAGA,QAAQ;AACN,QAAI,KAAK,MAAM,QAAQ;AACrB,YAAM,IAAI,MAAM,iBAAiB;AAAA,IACnC;AACA,QAAI,KAAK,QAAQ;AACf,YAAM,IAAI,MAAM,kBAAkB;AAAA,IACpC;AACA,SAAK,MAAM,IAAI,iBAAiB,cAAc;AAAA,EAChD;AAAA;AAAA,EAGA,WAAW;AACT,QAAI,KAAK,MAAM,QAAQ;AACrB,YAAM,IAAI,MAAM,iBAAiB;AAAA,IACnC;AACA,QAAI,KAAK,QAAQ;AACf,YAAM,IAAI,MAAM,kBAAkB;AAAA,IACpC;AACA,SAAK,MAAM,MAAM;AAAA,EACnB;AAAA,EAEA,OAA0F;AACxF,WAAO,KAAK,MAAM,KAAK;AAAA,EACzB;AAAA;AAAA,EAGA,QAAQ;AACN,SAAK,MAAM,MAAM;AACjB,SAAK,MAAM,MAAM;AACjB,SAAK,SAAS;AAAA,EAChB;AAAA,EAEA,CAAC,OAAO,aAAa,IAAsB;AACzC,WAAO;AAAA,EACT;AACF;AAgBO,MAAe,cAAiE;AAAA,EAC3E,QAAQ,IAAI,mBAAqC;AAAA,EACjD,SAAS;AAAA;AAAA,EAGnB,MAAM,UAA+B;AACnC,UAAM,SAAS,CAAC;AAChB,qBAAiB,SAAS,MAAM;AAC9B,aAAO,KAAK,MAAM,KAAK;AAAA,IACzB;AACA,WAAO,YAAY,MAAM;AAAA,EAC3B;AAAA,EAEA,OAAkD;AAChD,WAAO,KAAK,MAAM,KAAK;AAAA,EACzB;AAAA;AAAA,EAGA,QAAQ;AACN,SAAK,MAAM,MAAM;AACjB,SAAK,SAAS;AAAA,EAChB;AAAA,EAEA,CAAC,OAAO,aAAa,IAAmB;AACtC,WAAO;AAAA,EACT;AACF;","names":[]}
+{"version":3,"sources":["../../src/tts/tts.ts"],"sourcesContent":["// SPDX-FileCopyrightText: 2024 LiveKit, Inc.\n//\n// SPDX-License-Identifier: Apache-2.0\nimport type { AudioFrame } from '@livekit/rtc-node';\nimport type { TypedEventEmitter as TypedEmitter } from '@livekit/typed-emitter';\nimport { EventEmitter } from 'node:events';\nimport type { TTSMetrics } from '../metrics/base.js';\nimport { AsyncIterableQueue, mergeFrames } from '../utils.js';\n\n/** SynthesizedAudio is a packet of speech synthesis as returned by the TTS. */\nexport interface SynthesizedAudio {\n  /** Request ID (one segment could be made up of multiple requests) */\n  requestId: string;\n  /** Segment ID, each segment is separated by a flush */\n  segmentId: string;\n  /** Synthesized audio frame */\n  frame: AudioFrame;\n  /** Current segment of the synthesized audio */\n  deltaText?: string;\n  /** Whether this is the last frame of the segment (streaming only) */\n  final: boolean;\n}\n\n/**\n * Describes the capabilities of the TTS provider.\n *\n * @remarks\n * At present, only `streaming` is supplied to this interface, and the framework only supports\n * providers that do have a streaming endpoint.\n */\nexport interface TTSCapabilities {\n  streaming: boolean;\n}\n\nexport enum TTSEvent {\n  METRICS_COLLECTED,\n}\n\nexport type TTSCallbacks = {\n  [TTSEvent.METRICS_COLLECTED]: (metrics: TTSMetrics) => void;\n};\n\n/**\n * An instance of a text-to-speech adapter.\n *\n * @remarks\n * This class is abstract, and as such cannot be used directly. Instead, use a provider plugin that\n * exports its own child TTS class, which inherits this class's methods.\n */\nexport abstract class TTS extends (EventEmitter as new () => TypedEmitter<TTSCallbacks>) {\n  #capabilities: TTSCapabilities;\n  #sampleRate: number;\n  #numChannels: number;\n  abstract label: string;\n\n  constructor(sampleRate: number, numChannels: number, capabilities: TTSCapabilities) {\n    super();\n    this.#capabilities = capabilities;\n    this.#sampleRate = sampleRate;\n    this.#numChannels = numChannels;\n  }\n\n  /** Returns this TTS's capabilities */\n  get capabilities(): TTSCapabilities {\n    return this.#capabilities;\n  }\n\n  /** Returns the sample rate of audio frames returned by this TTS */\n  get sampleRate(): number {\n    return this.#sampleRate;\n  }\n\n  /** Returns the channel count of audio frames returned by this TTS */\n  get numChannels(): number {\n    return this.#numChannels;\n  }\n\n  /**\n   * Receives text and returns synthesis in the form of a {@link ChunkedStream}\n   */\n  abstract synthesize(text: string): ChunkedStream;\n\n  /**\n   * Returns a {@link SynthesizeStream} that can be used to push text and receive audio data\n   */\n  abstract stream(): SynthesizeStream;\n}\n\n/**\n * An instance of a text-to-speech stream, as an asynchronous iterable iterator.\n *\n * @example Looping through frames\n * ```ts\n * for await (const event of stream) {\n *   await source.captureFrame(event.frame);\n * }\n * ```\n *\n * @remarks\n * This class is abstract, and as such cannot be used directly. Instead, use a provider plugin that\n * exports its own child SynthesizeStream class, which inherits this class's methods.\n */\nexport abstract class SynthesizeStream\n  implements AsyncIterableIterator<SynthesizedAudio | typeof SynthesizeStream.END_OF_STREAM>\n{\n  protected static readonly FLUSH_SENTINEL = Symbol('FLUSH_SENTINEL');\n  static readonly END_OF_STREAM = Symbol('END_OF_STREAM');\n  protected input = new AsyncIterableQueue<string | typeof SynthesizeStream.FLUSH_SENTINEL>();\n  protected queue = new AsyncIterableQueue<\n    SynthesizedAudio | typeof SynthesizeStream.END_OF_STREAM\n  >();\n  protected output = new AsyncIterableQueue<\n    SynthesizedAudio | typeof SynthesizeStream.END_OF_STREAM\n  >();\n  protected closed = false;\n  abstract label: string;\n  #tts: TTS;\n  #metricsPendingTexts: string[] = [];\n  #metricsText = '';\n  #monitorMetricsTask?: Promise<void>;\n\n  constructor(tts: TTS) {\n    this.#tts = tts;\n  }\n\n  protected async monitorMetrics() {\n    const startTime = process.hrtime.bigint();\n    let audioDuration = 0;\n    let ttfb: bigint | undefined;\n    let requestId = '';\n\n    const emit = () => {\n      if (this.#metricsPendingTexts.length) {\n        const text = this.#metricsPendingTexts.shift()!;\n        const duration = process.hrtime.bigint() - startTime;\n        const metrics: TTSMetrics = {\n          timestamp: Date.now(),\n          requestId,\n          ttfb: Math.trunc(Number(ttfb! / BigInt(1000000))),\n          duration: Math.trunc(Number(duration / BigInt(1000000))),\n          charactersCount: text.length,\n          audioDuration,\n          cancelled: false, // XXX(nbsp)\n          label: this.label,\n          streamed: false,\n        };\n        this.#tts.emit(TTSEvent.METRICS_COLLECTED, metrics);\n      }\n    };\n\n    for await (const audio of this.queue) {\n      this.output.put(audio);\n      if (audio === SynthesizeStream.END_OF_STREAM) continue;\n      requestId = audio.requestId;\n      if (!ttfb) {\n        ttfb = process.hrtime.bigint() - startTime;\n      }\n      audioDuration += audio.frame.samplesPerChannel / audio.frame.sampleRate;\n      if (audio.final) {\n        emit();\n      }\n    }\n\n    if (requestId) {\n      emit();\n    }\n    this.output.close();\n  }\n\n  /** Push a string of text to the TTS */\n  pushText(text: string) {\n    if (!this.#monitorMetricsTask) {\n      this.#monitorMetricsTask = this.monitorMetrics();\n    }\n    this.#metricsText += text;\n\n    if (this.input.closed) {\n      throw new Error('Input is closed');\n    }\n    if (this.closed) {\n      throw new Error('Stream is closed');\n    }\n    this.input.put(text);\n  }\n\n  /** Flush the TTS, causing it to process all pending text */\n  flush() {\n    if (this.#metricsText) {\n      this.#metricsPendingTexts.push(this.#metricsText);\n      this.#metricsText = '';\n    }\n    if (this.input.closed) {\n      throw new Error('Input is closed');\n    }\n    if (this.closed) {\n      throw new Error('Stream is closed');\n    }\n    this.input.put(SynthesizeStream.FLUSH_SENTINEL);\n  }\n\n  /** Mark the input as ended and forbid additional pushes */\n  endInput() {\n    if (this.input.closed) {\n      throw new Error('Input is closed');\n    }\n    if (this.closed) {\n      throw new Error('Stream is closed');\n    }\n    this.input.close();\n  }\n\n  next(): Promise<IteratorResult<SynthesizedAudio | typeof SynthesizeStream.END_OF_STREAM>> {\n    return this.output.next();\n  }\n\n  /** Close both the input and output of the TTS stream */\n  close() {\n    this.input.close();\n    this.output.close();\n    this.closed = true;\n  }\n\n  [Symbol.asyncIterator](): SynthesizeStream {\n    return this;\n  }\n}\n\n/**\n * An instance of a text-to-speech response, as an asynchronous iterable iterator.\n *\n * @example Looping through frames\n * ```ts\n * for await (const event of stream) {\n *   await source.captureFrame(event.frame);\n * }\n * ```\n *\n * @remarks\n * This class is abstract, and as such cannot be used directly. Instead, use a provider plugin that\n * exports its own child ChunkedStream class, which inherits this class's methods.\n */\nexport abstract class ChunkedStream implements AsyncIterableIterator<SynthesizedAudio> {\n  protected queue = new AsyncIterableQueue<SynthesizedAudio>();\n  protected output = new AsyncIterableQueue<SynthesizedAudio>();\n  protected closed = false;\n  abstract label: string;\n  #text: string;\n  #tts: TTS;\n\n  constructor(text: string, tts: TTS) {\n    this.#text = text;\n    this.#tts = tts;\n\n    this.monitorMetrics();\n  }\n\n  protected async monitorMetrics() {\n    const startTime = process.hrtime.bigint();\n    let audioDuration = 0;\n    let ttfb: bigint | undefined;\n    let requestId = '';\n\n    for await (const audio of this.queue) {\n      this.output.put(audio);\n      requestId = audio.requestId;\n      if (!ttfb) {\n        ttfb = process.hrtime.bigint() - startTime;\n      }\n      audioDuration += audio.frame.samplesPerChannel / audio.frame.sampleRate;\n    }\n    this.output.close();\n\n    const duration = process.hrtime.bigint() - startTime;\n    const metrics: TTSMetrics = {\n      timestamp: Date.now(),\n      requestId,\n      ttfb: Math.trunc(Number(ttfb! / BigInt(1000000))),\n      duration: Math.trunc(Number(duration / BigInt(1000000))),\n      charactersCount: this.#text.length,\n      audioDuration,\n      cancelled: false, // XXX(nbsp)\n      label: this.label,\n      streamed: false,\n    };\n    this.#tts.emit(TTSEvent.METRICS_COLLECTED, metrics);\n  }\n\n  /** Collect every frame into one in a single call */\n  async collect(): Promise<AudioFrame> {\n    const frames = [];\n    for await (const event of this) {\n      frames.push(event.frame);\n    }\n    return mergeFrames(frames);\n  }\n\n  next(): Promise<IteratorResult<SynthesizedAudio>> {\n    return this.output.next();\n  }\n\n  /** Close both the input and output of the TTS stream */\n  close() {\n    this.queue.close();\n    this.output.close();\n    this.closed = true;\n  }\n\n  [Symbol.asyncIterator](): ChunkedStream {\n    return this;\n  }\n}\n"],"mappings":"AAKA,SAAS,oBAAoB;AAE7B,SAAS,oBAAoB,mBAAmB;AA2BzC,IAAK,WAAL,kBAAKA,cAAL;AACL,EAAAA,oBAAA;AADU,SAAAA;AAAA,GAAA;AAeL,MAAe,YAAa,aAAsD;AAAA,EACvF;AAAA,EACA;AAAA,EACA;AAAA,EAGA,YAAY,YAAoB,aAAqB,cAA+B;AAClF,UAAM;AACN,SAAK,gBAAgB;AACrB,SAAK,cAAc;AACnB,SAAK,eAAe;AAAA,EACtB;AAAA;AAAA,EAGA,IAAI,eAAgC;AAClC,WAAO,KAAK;AAAA,EACd;AAAA;AAAA,EAGA,IAAI,aAAqB;AACvB,WAAO,KAAK;AAAA,EACd;AAAA;AAAA,EAGA,IAAI,cAAsB;AACxB,WAAO,KAAK;AAAA,EACd;AAWF;AAgBO,MAAe,iBAEtB;AAAA,EACE,OAA0B,iBAAiB,OAAO,gBAAgB;AAAA,EAClE,OAAgB,gBAAgB,OAAO,eAAe;AAAA,EAC5C,QAAQ,IAAI,mBAAoE;AAAA,EAChF,QAAQ,IAAI,mBAEpB;AAAA,EACQ,SAAS,IAAI,mBAErB;AAAA,EACQ,SAAS;AAAA,EAEnB;AAAA,EACA,uBAAiC,CAAC;AAAA,EAClC,eAAe;AAAA,EACf;AAAA,EAEA,YAAY,KAAU;AACpB,SAAK,OAAO;AAAA,EACd;AAAA,EAEA,MAAgB,iBAAiB;AAC/B,UAAM,YAAY,QAAQ,OAAO,OAAO;AACxC,QAAI,gBAAgB;AACpB,QAAI;AACJ,QAAI,YAAY;AAEhB,UAAM,OAAO,MAAM;AACjB,UAAI,KAAK,qBAAqB,QAAQ;AACpC,cAAM,OAAO,KAAK,qBAAqB,MAAM;AAC7C,cAAM,WAAW,QAAQ,OAAO,OAAO,IAAI;AAC3C,cAAM,UAAsB;AAAA,UAC1B,WAAW,KAAK,IAAI;AAAA,UACpB;AAAA,UACA,MAAM,KAAK,MAAM,OAAO,OAAQ,OAAO,GAAO,CAAC,CAAC;AAAA,UAChD,UAAU,KAAK,MAAM,OAAO,WAAW,OAAO,GAAO,CAAC,CAAC;AAAA,UACvD,iBAAiB,KAAK;AAAA,UACtB;AAAA,UACA,WAAW;AAAA;AAAA,UACX,OAAO,KAAK;AAAA,UACZ,UAAU;AAAA,QACZ;AACA,aAAK,KAAK,KAAK,2BAA4B,OAAO;AAAA,MACpD;AAAA,IACF;AAEA,qBAAiB,SAAS,KAAK,OAAO;AACpC,WAAK,OAAO,IAAI,KAAK;AACrB,UAAI,UAAU,iBAAiB,cAAe;AAC9C,kBAAY,MAAM;AAClB,UAAI,CAAC,MAAM;AACT,eAAO,QAAQ,OAAO,OAAO,IAAI;AAAA,MACnC;AACA,uBAAiB,MAAM,MAAM,oBAAoB,MAAM,MAAM;AAC7D,UAAI,MAAM,OAAO;AACf,aAAK;AAAA,MACP;AAAA,IACF;AAEA,QAAI,WAAW;AACb,WAAK;AAAA,IACP;AACA,SAAK,OAAO,MAAM;AAAA,EACpB;AAAA;AAAA,EAGA,SAAS,MAAc;AACrB,QAAI,CAAC,KAAK,qBAAqB;AAC7B,WAAK,sBAAsB,KAAK,eAAe;AAAA,IACjD;AACA,SAAK,gBAAgB;AAErB,QAAI,KAAK,MAAM,QAAQ;AACrB,YAAM,IAAI,MAAM,iBAAiB;AAAA,IACnC;AACA,QAAI,KAAK,QAAQ;AACf,YAAM,IAAI,MAAM,kBAAkB;AAAA,IACpC;AACA,SAAK,MAAM,IAAI,IAAI;AAAA,EACrB;AAAA;AAAA,EAGA,QAAQ;AACN,QAAI,KAAK,cAAc;AACrB,WAAK,qBAAqB,KAAK,KAAK,YAAY;AAChD,WAAK,eAAe;AAAA,IACtB;AACA,QAAI,KAAK,MAAM,QAAQ;AACrB,YAAM,IAAI,MAAM,iBAAiB;AAAA,IACnC;AACA,QAAI,KAAK,QAAQ;AACf,YAAM,IAAI,MAAM,kBAAkB;AAAA,IACpC;AACA,SAAK,MAAM,IAAI,iBAAiB,cAAc;AAAA,EAChD;AAAA;AAAA,EAGA,WAAW;AACT,QAAI,KAAK,MAAM,QAAQ;AACrB,YAAM,IAAI,MAAM,iBAAiB;AAAA,IACnC;AACA,QAAI,KAAK,QAAQ;AACf,YAAM,IAAI,MAAM,kBAAkB;AAAA,IACpC;AACA,SAAK,MAAM,MAAM;AAAA,EACnB;AAAA,EAEA,OAA0F;AACxF,WAAO,KAAK,OAAO,KAAK;AAAA,EAC1B;AAAA;AAAA,EAGA,QAAQ;AACN,SAAK,MAAM,MAAM;AACjB,SAAK,OAAO,MAAM;AAClB,SAAK,SAAS;AAAA,EAChB;AAAA,EAEA,CAAC,OAAO,aAAa,IAAsB;AACzC,WAAO;AAAA,EACT;AACF;AAgBO,MAAe,cAAiE;AAAA,EAC3E,QAAQ,IAAI,mBAAqC;AAAA,EACjD,SAAS,IAAI,mBAAqC;AAAA,EAClD,SAAS;AAAA,EAEnB;AAAA,EACA;AAAA,EAEA,YAAY,MAAc,KAAU;AAClC,SAAK,QAAQ;AACb,SAAK,OAAO;AAEZ,SAAK,eAAe;AAAA,EACtB;AAAA,EAEA,MAAgB,iBAAiB;AAC/B,UAAM,YAAY,QAAQ,OAAO,OAAO;AACxC,QAAI,gBAAgB;AACpB,QAAI;AACJ,QAAI,YAAY;AAEhB,qBAAiB,SAAS,KAAK,OAAO;AACpC,WAAK,OAAO,IAAI,KAAK;AACrB,kBAAY,MAAM;AAClB,UAAI,CAAC,MAAM;AACT,eAAO,QAAQ,OAAO,OAAO,IAAI;AAAA,MACnC;AACA,uBAAiB,MAAM,MAAM,oBAAoB,MAAM,MAAM;AAAA,IAC/D;AACA,SAAK,OAAO,MAAM;AAElB,UAAM,WAAW,QAAQ,OAAO,OAAO,IAAI;AAC3C,UAAM,UAAsB;AAAA,MAC1B,WAAW,KAAK,IAAI;AAAA,MACpB;AAAA,MACA,MAAM,KAAK,MAAM,OAAO,OAAQ,OAAO,GAAO,CAAC,CAAC;AAAA,MAChD,UAAU,KAAK,MAAM,OAAO,WAAW,OAAO,GAAO,CAAC,CAAC;AAAA,MACvD,iBAAiB,KAAK,MAAM;AAAA,MAC5B;AAAA,MACA,WAAW;AAAA;AAAA,MACX,OAAO,KAAK;AAAA,MACZ,UAAU;AAAA,IACZ;AACA,SAAK,KAAK,KAAK,2BAA4B,OAAO;AAAA,EACpD;AAAA;AAAA,EAGA,MAAM,UAA+B;AACnC,UAAM,SAAS,CAAC;AAChB,qBAAiB,SAAS,MAAM;AAC9B,aAAO,KAAK,MAAM,KAAK;AAAA,IACzB;AACA,WAAO,YAAY,MAAM;AAAA,EAC3B;AAAA,EAEA,OAAkD;AAChD,WAAO,KAAK,OAAO,KAAK;AAAA,EAC1B;AAAA;AAAA,EAGA,QAAQ;AACN,SAAK,MAAM,MAAM;AACjB,SAAK,OAAO,MAAM;AAClB,SAAK,SAAS;AAAA,EAChB;AAAA,EAEA,CAAC,OAAO,aAAa,IAAmB;AACtC,WAAO;AAAA,EACT;AACF;","names":["TTSEvent"]}

package/dist/vad.cjs

@@ -23,16 +23,19 @@ __export(vad_exports, {
   VADStream: () => VADStream
 });
 module.exports = __toCommonJS(vad_exports);
+var import_node_events = require("node:events");
 var import_utils = require("./utils.cjs");
 var VADEventType = /* @__PURE__ */ ((VADEventType2) => {
   VADEventType2[VADEventType2["START_OF_SPEECH"] = 0] = "START_OF_SPEECH";
   VADEventType2[VADEventType2["INFERENCE_DONE"] = 1] = "INFERENCE_DONE";
   VADEventType2[VADEventType2["END_OF_SPEECH"] = 2] = "END_OF_SPEECH";
+  VADEventType2[VADEventType2["METRICS_COLLECTED"] = 3] = "METRICS_COLLECTED";
   return VADEventType2;
 })(VADEventType || {});
-class VAD {
+class VAD extends import_node_events.EventEmitter {
   #capabilities;
   constructor(capabilities) {
+    super();
     this.#capabilities = capabilities;
   }
   get capabilities() {
@@ -43,7 +46,44 @@ class VADStream {
   static FLUSH_SENTINEL = Symbol("FLUSH_SENTINEL");
   input = new import_utils.AsyncIterableQueue();
   queue = new import_utils.AsyncIterableQueue();
+  output = new import_utils.AsyncIterableQueue();
   closed = false;
+  #vad;
+  #lastActivityTime = BigInt(0);
+  constructor(vad) {
+    this.#vad = vad;
+    this.monitorMetrics();
+  }
+  async monitorMetrics() {
+    let inferenceDurationTotal = 0;
+    let inferenceCount = 0;
+    for await (const event of this.queue) {
+      this.output.put(event);
+      switch (event.type) {
+        case 0 /* START_OF_SPEECH */:
+          inferenceCount++;
+          if (inferenceCount >= 1 / this.#vad.capabilities.updateInterval) {
+            this.#vad.emit(3 /* METRICS_COLLECTED */, {
+              timestamp: Date.now(),
+              idleTime: Math.trunc(
+                Number((process.hrtime.bigint() - this.#lastActivityTime) / BigInt(1e6))
+              ),
+              inferenceDurationTotal,
+              inferenceCount,
+              label: this.#vad.label
+            });
+            inferenceCount = 0;
+            inferenceDurationTotal = 0;
+          }
+          break;
+        case 1 /* INFERENCE_DONE */:
+        case 2 /* END_OF_SPEECH */:
+          this.#lastActivityTime = process.hrtime.bigint();
+          break;
+      }
+    }
+    this.output.close();
+  }
   pushFrame(frame) {
     if (this.input.closed) {
       throw new Error("Input is closed");
@@ -72,11 +112,12 @@ class VADStream {
     this.input.close();
   }
   next() {
-    return this.queue.next();
+    return this.output.next();
   }
   close() {
     this.input.close();
     this.queue.close();
+    this.output.close();
     this.closed = true;
   }
   [Symbol.asyncIterator]() {

package/dist/vad.cjs.map

@@ -1 +1 @@
-{"version":3,"sources":["../src/vad.ts"],"sourcesContent":["// SPDX-FileCopyrightText: 2024 LiveKit, Inc.\n//\n// SPDX-License-Identifier: Apache-2.0\nimport type { AudioFrame } from '@livekit/rtc-node';\nimport { AsyncIterableQueue } from './utils.js';\n\nexport enum VADEventType {\n  START_OF_SPEECH,\n  INFERENCE_DONE,\n  END_OF_SPEECH,\n}\n\nexport interface VADEvent {\n  /** Type of the VAD event (e.g., start of speech, end of speech, inference done). */\n  type: VADEventType;\n  /**\n   * Index of the audio sample where the event occurred, relative to the inference sample rate.\n   */\n  samplesIndex: number;\n  /** Timestamp when the event was fired. */\n  timestamp: number;\n  /** Duration of the detected speech segment in seconds. */\n  speechDuration: number;\n  /** Duration of the silence segment preceding or following the speech, in seconds. */\n  silenceDuration: number;\n  /**\n   * List of audio frames associated with the speech.\n   *\n   * @remarks\n   * - For `start_of_speech` events, this contains the audio chunks that triggered the detection.\n   * - For `inference_done` events, this contains the audio chunks that were processed.\n   * - For `end_of_speech` events, this contains the complete user speech.\n   */\n  frames: AudioFrame[];\n  /** Probability that speech is present (only for `INFERENCE_DONE` events). */\n  probability: number;\n  /** Time taken to perform the inference, in seconds (only for `INFERENCE_DONE` events). */\n  inferenceDuration: number;\n  /** Indicates whether speech was detected in the frames. */\n  speaking: boolean;\n}\n\nexport interface VADCapabilities {\n  updateInterval: number;\n}\n\nexport abstract class VAD {\n  #capabilities: VADCapabilities;\n  constructor(capabilities: VADCapabilities) {\n    this.#capabilities = capabilities;\n  }\n\n  get capabilities(): VADCapabilities {\n    return this.#capabilities;\n  }\n\n  /**\n   * Returns a {@link VADStream} that can be used to push audio frames and receive VAD events.\n   */\n  abstract stream(): VADStream;\n}\n\nexport abstract class VADStream implements AsyncIterableIterator<VADEvent> {\n  protected static readonly FLUSH_SENTINEL = Symbol('FLUSH_SENTINEL');\n  protected input = new AsyncIterableQueue<AudioFrame | typeof VADStream.FLUSH_SENTINEL>();\n  protected queue = new AsyncIterableQueue<VADEvent>();\n  protected closed = false;\n\n  pushFrame(frame: AudioFrame) {\n    if (this.input.closed) {\n      throw new Error('Input is closed');\n    }\n    if (this.closed) {\n      throw new Error('Stream is closed');\n    }\n    this.input.put(frame);\n  }\n\n  flush() {\n    if (this.input.closed) {\n      throw new Error('Input is closed');\n    }\n    if (this.closed) {\n      throw new Error('Stream is closed');\n    }\n    this.input.put(VADStream.FLUSH_SENTINEL);\n  }\n\n  endInput() {\n    if (this.input.closed) {\n      throw new Error('Input is closed');\n    }\n    if (this.closed) {\n      throw new Error('Stream is closed');\n    }\n    this.input.close();\n  }\n\n  next(): Promise<IteratorResult<VADEvent>> {\n    return this.queue.next();\n  }\n\n  close() {\n    this.input.close();\n    this.queue.close();\n    this.closed = true;\n  }\n\n  [Symbol.asyncIterator](): VADStream {\n    return this;\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAIA,mBAAmC;AAE5B,IAAK,eAAL,kBAAKA,kBAAL;AACL,EAAAA,4BAAA;AACA,EAAAA,4BAAA;AACA,EAAAA,4BAAA;AAHU,SAAAA;AAAA,GAAA;AAwCL,MAAe,IAAI;AAAA,EACxB;AAAA,EACA,YAAY,cAA+B;AACzC,SAAK,gBAAgB;AAAA,EACvB;AAAA,EAEA,IAAI,eAAgC;AAClC,WAAO,KAAK;AAAA,EACd;AAMF;AAEO,MAAe,UAAqD;AAAA,EACzE,OAA0B,iBAAiB,OAAO,gBAAgB;AAAA,EACxD,QAAQ,IAAI,gCAAiE;AAAA,EAC7E,QAAQ,IAAI,gCAA6B;AAAA,EACzC,SAAS;AAAA,EAEnB,UAAU,OAAmB;AAC3B,QAAI,KAAK,MAAM,QAAQ;AACrB,YAAM,IAAI,MAAM,iBAAiB;AAAA,IACnC;AACA,QAAI,KAAK,QAAQ;AACf,YAAM,IAAI,MAAM,kBAAkB;AAAA,IACpC;AACA,SAAK,MAAM,IAAI,KAAK;AAAA,EACtB;AAAA,EAEA,QAAQ;AACN,QAAI,KAAK,MAAM,QAAQ;AACrB,YAAM,IAAI,MAAM,iBAAiB;AAAA,IACnC;AACA,QAAI,KAAK,QAAQ;AACf,YAAM,IAAI,MAAM,kBAAkB;AAAA,IACpC;AACA,SAAK,MAAM,IAAI,UAAU,cAAc;AAAA,EACzC;AAAA,EAEA,WAAW;AACT,QAAI,KAAK,MAAM,QAAQ;AACrB,YAAM,IAAI,MAAM,iBAAiB;AAAA,IACnC;AACA,QAAI,KAAK,QAAQ;AACf,YAAM,IAAI,MAAM,kBAAkB;AAAA,IACpC;AACA,SAAK,MAAM,MAAM;AAAA,EACnB;AAAA,EAEA,OAA0C;AACxC,WAAO,KAAK,MAAM,KAAK;AAAA,EACzB;AAAA,EAEA,QAAQ;AACN,SAAK,MAAM,MAAM;AACjB,SAAK,MAAM,MAAM;AACjB,SAAK,SAAS;AAAA,EAChB;AAAA,EAEA,CAAC,OAAO,aAAa,IAAe;AAClC,WAAO;AAAA,EACT;AACF;","names":["VADEventType"]}
+{"version":3,"sources":["../src/vad.ts"],"sourcesContent":["// SPDX-FileCopyrightText: 2024 LiveKit, Inc.\n//\n// SPDX-License-Identifier: Apache-2.0\nimport type { AudioFrame } from '@livekit/rtc-node';\nimport type { TypedEventEmitter as TypedEmitter } from '@livekit/typed-emitter';\nimport { EventEmitter } from 'node:events';\nimport type { VADMetrics } from './metrics/base.js';\nimport { AsyncIterableQueue } from './utils.js';\n\nexport enum VADEventType {\n  START_OF_SPEECH,\n  INFERENCE_DONE,\n  END_OF_SPEECH,\n  METRICS_COLLECTED,\n}\n\nexport interface VADEvent {\n  /** Type of the VAD event (e.g., start of speech, end of speech, inference done). */\n  type: VADEventType;\n  /**\n   * Index of the audio sample where the event occurred, relative to the inference sample rate.\n   */\n  samplesIndex: number;\n  /** Timestamp when the event was fired. */\n  timestamp: number;\n  /** Duration of the speech segment. */\n  speechDuration: number;\n  /** Duration of the silence segment. */\n  silenceDuration: number;\n  /**\n   * List of audio frames associated with the speech.\n   *\n   * @remarks\n   * - For `start_of_speech` events, this contains the audio chunks that triggered the detection.\n   * - For `inference_done` events, this contains the audio chunks that were processed.\n   * - For `end_of_speech` events, this contains the complete user speech.\n   */\n  frames: AudioFrame[];\n  /** Probability that speech is present (only for `INFERENCE_DONE` events). */\n  probability: number;\n  /** Time taken to perform the inference, in seconds (only for `INFERENCE_DONE` events). */\n  inferenceDuration: number;\n  /** Indicates whether speech was detected in the frames. */\n  speaking: boolean;\n  /** Threshold used to detect silence. */\n  rawAccumulatedSilence: number;\n  /** Threshold used to detect speech. */\n  rawAccumulatedSpeech: number;\n}\n\nexport interface VADCapabilities {\n  updateInterval: number;\n}\n\nexport type VADCallbacks = {\n  [VADEventType.METRICS_COLLECTED]: (metrics: VADMetrics) => void;\n};\n\nexport abstract class VAD extends (EventEmitter as new () => TypedEmitter<VADCallbacks>) {\n  #capabilities: VADCapabilities;\n  abstract label: string;\n\n  constructor(capabilities: VADCapabilities) {\n    super();\n    this.#capabilities = capabilities;\n  }\n\n  get capabilities(): VADCapabilities {\n    return this.#capabilities;\n  }\n\n  /**\n   * Returns a {@link VADStream} that can be used to push audio frames and receive VAD events.\n   */\n  abstract stream(): VADStream;\n}\n\nexport abstract class VADStream implements AsyncIterableIterator<VADEvent> {\n  protected static readonly FLUSH_SENTINEL = Symbol('FLUSH_SENTINEL');\n  protected input = new AsyncIterableQueue<AudioFrame | typeof VADStream.FLUSH_SENTINEL>();\n  protected queue = new AsyncIterableQueue<VADEvent>();\n  protected output = new AsyncIterableQueue<VADEvent>();\n  protected closed = false;\n  #vad: VAD;\n  #lastActivityTime = BigInt(0);\n\n  constructor(vad: VAD) {\n    this.#vad = vad;\n    this.monitorMetrics();\n  }\n\n  protected async monitorMetrics() {\n    let inferenceDurationTotal = 0;\n    let inferenceCount = 0;\n\n    for await (const event of this.queue) {\n      this.output.put(event);\n      switch (event.type) {\n        case VADEventType.START_OF_SPEECH:\n          inferenceCount++;\n          if (inferenceCount >= 1 / this.#vad.capabilities.updateInterval) {\n            this.#vad.emit(VADEventType.METRICS_COLLECTED, {\n              timestamp: Date.now(),\n              idleTime: Math.trunc(\n                Number((process.hrtime.bigint() - this.#lastActivityTime) / BigInt(1000000)),\n              ),\n              inferenceDurationTotal,\n              inferenceCount,\n              label: this.#vad.label,\n            });\n\n            inferenceCount = 0;\n            inferenceDurationTotal = 0;\n          }\n          break;\n        case VADEventType.INFERENCE_DONE:\n        case VADEventType.END_OF_SPEECH:\n          this.#lastActivityTime = process.hrtime.bigint();\n          break;\n      }\n    }\n    this.output.close();\n  }\n\n  pushFrame(frame: AudioFrame) {\n    if (this.input.closed) {\n      throw new Error('Input is closed');\n    }\n    if (this.closed) {\n      throw new Error('Stream is closed');\n    }\n    this.input.put(frame);\n  }\n\n  flush() {\n    if (this.input.closed) {\n      throw new Error('Input is closed');\n    }\n    if (this.closed) {\n      throw new Error('Stream is closed');\n    }\n    this.input.put(VADStream.FLUSH_SENTINEL);\n  }\n\n  endInput() {\n    if (this.input.closed) {\n      throw new Error('Input is closed');\n    }\n    if (this.closed) {\n      throw new Error('Stream is closed');\n    }\n    this.input.close();\n  }\n\n  next(): Promise<IteratorResult<VADEvent>> {\n    return this.output.next();\n  }\n\n  close() {\n    this.input.close();\n    this.queue.close();\n    this.output.close();\n    this.closed = true;\n  }\n\n  [Symbol.asyncIterator](): VADStream {\n    return this;\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAKA,yBAA6B;AAE7B,mBAAmC;AAE5B,IAAK,eAAL,kBAAKA,kBAAL;AACL,EAAAA,4BAAA;AACA,EAAAA,4BAAA;AACA,EAAAA,4BAAA;AACA,EAAAA,4BAAA;AAJU,SAAAA;AAAA,GAAA;AAiDL,MAAe,YAAa,gCAAsD;AAAA,EACvF;AAAA,EAGA,YAAY,cAA+B;AACzC,UAAM;AACN,SAAK,gBAAgB;AAAA,EACvB;AAAA,EAEA,IAAI,eAAgC;AAClC,WAAO,KAAK;AAAA,EACd;AAMF;AAEO,MAAe,UAAqD;AAAA,EACzE,OAA0B,iBAAiB,OAAO,gBAAgB;AAAA,EACxD,QAAQ,IAAI,gCAAiE;AAAA,EAC7E,QAAQ,IAAI,gCAA6B;AAAA,EACzC,SAAS,IAAI,gCAA6B;AAAA,EAC1C,SAAS;AAAA,EACnB;AAAA,EACA,oBAAoB,OAAO,CAAC;AAAA,EAE5B,YAAY,KAAU;AACpB,SAAK,OAAO;AACZ,SAAK,eAAe;AAAA,EACtB;AAAA,EAEA,MAAgB,iBAAiB;AAC/B,QAAI,yBAAyB;AAC7B,QAAI,iBAAiB;AAErB,qBAAiB,SAAS,KAAK,OAAO;AACpC,WAAK,OAAO,IAAI,KAAK;AACrB,cAAQ,MAAM,MAAM;AAAA,QAClB,KAAK;AACH;AACA,cAAI,kBAAkB,IAAI,KAAK,KAAK,aAAa,gBAAgB;AAC/D,iBAAK,KAAK,KAAK,2BAAgC;AAAA,cAC7C,WAAW,KAAK,IAAI;AAAA,cACpB,UAAU,KAAK;AAAA,gBACb,QAAQ,QAAQ,OAAO,OAAO,IAAI,KAAK,qBAAqB,OAAO,GAAO,CAAC;AAAA,cAC7E;AAAA,cACA;AAAA,cACA;AAAA,cACA,OAAO,KAAK,KAAK;AAAA,YACnB,CAAC;AAED,6BAAiB;AACjB,qCAAyB;AAAA,UAC3B;AACA;AAAA,QACF,KAAK;AAAA,QACL,KAAK;AACH,eAAK,oBAAoB,QAAQ,OAAO,OAAO;AAC/C;AAAA,MACJ;AAAA,IACF;AACA,SAAK,OAAO,MAAM;AAAA,EACpB;AAAA,EAEA,UAAU,OAAmB;AAC3B,QAAI,KAAK,MAAM,QAAQ;AACrB,YAAM,IAAI,MAAM,iBAAiB;AAAA,IACnC;AACA,QAAI,KAAK,QAAQ;AACf,YAAM,IAAI,MAAM,kBAAkB;AAAA,IACpC;AACA,SAAK,MAAM,IAAI,KAAK;AAAA,EACtB;AAAA,EAEA,QAAQ;AACN,QAAI,KAAK,MAAM,QAAQ;AACrB,YAAM,IAAI,MAAM,iBAAiB;AAAA,IACnC;AACA,QAAI,KAAK,QAAQ;AACf,YAAM,IAAI,MAAM,kBAAkB;AAAA,IACpC;AACA,SAAK,MAAM,IAAI,UAAU,cAAc;AAAA,EACzC;AAAA,EAEA,WAAW;AACT,QAAI,KAAK,MAAM,QAAQ;AACrB,YAAM,IAAI,MAAM,iBAAiB;AAAA,IACnC;AACA,QAAI,KAAK,QAAQ;AACf,YAAM,IAAI,MAAM,kBAAkB;AAAA,IACpC;AACA,SAAK,MAAM,MAAM;AAAA,EACnB;AAAA,EAEA,OAA0C;AACxC,WAAO,KAAK,OAAO,KAAK;AAAA,EAC1B;AAAA,EAEA,QAAQ;AACN,SAAK,MAAM,MAAM;AACjB,SAAK,MAAM,MAAM;AACjB,SAAK,OAAO,MAAM;AAClB,SAAK,SAAS;AAAA,EAChB;AAAA,EAEA,CAAC,OAAO,aAAa,IAAe;AAClC,WAAO;AAAA,EACT;AACF;","names":["VADEventType"]}

package/dist/vad.d.ts

@@ -1,9 +1,12 @@
 import type { AudioFrame } from '@livekit/rtc-node';
+import type { TypedEventEmitter as TypedEmitter } from '@livekit/typed-emitter';
+import type { VADMetrics } from './metrics/base.js';
 import { AsyncIterableQueue } from './utils.js';
 export declare enum VADEventType {
     START_OF_SPEECH = 0,
     INFERENCE_DONE = 1,
-    END_OF_SPEECH = 2
+    END_OF_SPEECH = 2,
+    METRICS_COLLECTED = 3
 }
 export interface VADEvent {
     /** Type of the VAD event (e.g., start of speech, end of speech, inference done). */
@@ -14,9 +17,9 @@ export interface VADEvent {
     samplesIndex: number;
     /** Timestamp when the event was fired. */
     timestamp: number;
-    /** Duration of the detected speech segment in seconds. */
+    /** Duration of the speech segment. */
     speechDuration: number;
-    /** Duration of the silence segment preceding or following the speech, in seconds. */
+    /** Duration of the silence segment. */
     silenceDuration: number;
     /**
      * List of audio frames associated with the speech.
@@ -33,12 +36,21 @@ export interface VADEvent {
     inferenceDuration: number;
     /** Indicates whether speech was detected in the frames. */
     speaking: boolean;
+    /** Threshold used to detect silence. */
+    rawAccumulatedSilence: number;
+    /** Threshold used to detect speech. */
+    rawAccumulatedSpeech: number;
 }
 export interface VADCapabilities {
     updateInterval: number;
 }
-export declare abstract class VAD {
+export type VADCallbacks = {
+    [VADEventType.METRICS_COLLECTED]: (metrics: VADMetrics) => void;
+};
+declare const VAD_base: new () => TypedEmitter<VADCallbacks>;
+export declare abstract class VAD extends VAD_base {
     #private;
+    abstract label: string;
     constructor(capabilities: VADCapabilities);
     get capabilities(): VADCapabilities;
     /**
@@ -47,10 +59,14 @@ export declare abstract class VAD {
     abstract stream(): VADStream;
 }
 export declare abstract class VADStream implements AsyncIterableIterator<VADEvent> {
+    #private;
     protected static readonly FLUSH_SENTINEL: unique symbol;
     protected input: AsyncIterableQueue<AudioFrame | typeof VADStream.FLUSH_SENTINEL>;
     protected queue: AsyncIterableQueue<VADEvent>;
+    protected output: AsyncIterableQueue<VADEvent>;
     protected closed: boolean;
+    constructor(vad: VAD);
+    protected monitorMetrics(): Promise<void>;
     pushFrame(frame: AudioFrame): void;
     flush(): void;
     endInput(): void;
@@ -58,4 +74,5 @@ export declare abstract class VADStream implements AsyncIterableIterator<VADEven
     close(): void;
     [Symbol.asyncIterator](): VADStream;
 }
+export {};
 //# sourceMappingURL=vad.d.ts.map

package/dist/vad.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"vad.d.ts","sourceRoot":"","sources":["../src/vad.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,EAAE,kBAAkB,EAAE,MAAM,YAAY,CAAC;AAEhD,oBAAY,YAAY;IACtB,eAAe,IAAA;IACf,cAAc,IAAA;IACd,aAAa,IAAA;CACd;AAED,MAAM,WAAW,QAAQ;IACvB,oFAAoF;IACpF,IAAI,EAAE,YAAY,CAAC;IACnB;;OAEG;IACH,YAAY,EAAE,MAAM,CAAC;IACrB,0CAA0C;IAC1C,SAAS,EAAE,MAAM,CAAC;IAClB,0DAA0D;IAC1D,cAAc,EAAE,MAAM,CAAC;IACvB,qFAAqF;IACrF,eAAe,EAAE,MAAM,CAAC;IACxB;;;;;;;OAOG;IACH,MAAM,EAAE,UAAU,EAAE,CAAC;IACrB,6EAA6E;IAC7E,WAAW,EAAE,MAAM,CAAC;IACpB,0FAA0F;IAC1F,iBAAiB,EAAE,MAAM,CAAC;IAC1B,2DAA2D;IAC3D,QAAQ,EAAE,OAAO,CAAC;CACnB;AAED,MAAM,WAAW,eAAe;IAC9B,cAAc,EAAE,MAAM,CAAC;CACxB;AAED,8BAAsB,GAAG;;gBAEX,YAAY,EAAE,eAAe;IAIzC,IAAI,YAAY,IAAI,eAAe,CAElC;IAED;;OAEG;IACH,QAAQ,CAAC,MAAM,IAAI,SAAS;CAC7B;AAED,8BAAsB,SAAU,YAAW,qBAAqB,CAAC,QAAQ,CAAC;IACxE,SAAS,CAAC,MAAM,CAAC,QAAQ,CAAC,cAAc,gBAA4B;IACpE,SAAS,CAAC,KAAK,mEAA0E;IACzF,SAAS,CAAC,KAAK,+BAAsC;IACrD,SAAS,CAAC,MAAM,UAAS;IAEzB,SAAS,CAAC,KAAK,EAAE,UAAU;IAU3B,KAAK;IAUL,QAAQ;IAUR,IAAI,IAAI,OAAO,CAAC,cAAc,CAAC,QAAQ,CAAC,CAAC;IAIzC,KAAK;IAML,CAAC,MAAM,CAAC,aAAa,CAAC,IAAI,SAAS;CAGpC"}
+{"version":3,"file":"vad.d.ts","sourceRoot":"","sources":["../src/vad.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,KAAK,EAAE,iBAAiB,IAAI,YAAY,EAAE,MAAM,wBAAwB,CAAC;AAEhF,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,EAAE,kBAAkB,EAAE,MAAM,YAAY,CAAC;AAEhD,oBAAY,YAAY;IACtB,eAAe,IAAA;IACf,cAAc,IAAA;IACd,aAAa,IAAA;IACb,iBAAiB,IAAA;CAClB;AAED,MAAM,WAAW,QAAQ;IACvB,oFAAoF;IACpF,IAAI,EAAE,YAAY,CAAC;IACnB;;OAEG;IACH,YAAY,EAAE,MAAM,CAAC;IACrB,0CAA0C;IAC1C,SAAS,EAAE,MAAM,CAAC;IAClB,sCAAsC;IACtC,cAAc,EAAE,MAAM,CAAC;IACvB,uCAAuC;IACvC,eAAe,EAAE,MAAM,CAAC;IACxB;;;;;;;OAOG;IACH,MAAM,EAAE,UAAU,EAAE,CAAC;IACrB,6EAA6E;IAC7E,WAAW,EAAE,MAAM,CAAC;IACpB,0FAA0F;IAC1F,iBAAiB,EAAE,MAAM,CAAC;IAC1B,2DAA2D;IAC3D,QAAQ,EAAE,OAAO,CAAC;IAClB,wCAAwC;IACxC,qBAAqB,EAAE,MAAM,CAAC;IAC9B,uCAAuC;IACvC,oBAAoB,EAAE,MAAM,CAAC;CAC9B;AAED,MAAM,WAAW,eAAe;IAC9B,cAAc,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,MAAM,YAAY,GAAG;IACzB,CAAC,YAAY,CAAC,iBAAiB,CAAC,EAAE,CAAC,OAAO,EAAE,UAAU,KAAK,IAAI,CAAC;CACjE,CAAC;kCAE2D,aAAa,YAAY,CAAC;AAAvF,8BAAsB,GAAI,SAAQ,QAAsD;;IAEtF,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;gBAEX,YAAY,EAAE,eAAe;IAKzC,IAAI,YAAY,IAAI,eAAe,CAElC;IAED;;OAEG;IACH,QAAQ,CAAC,MAAM,IAAI,SAAS;CAC7B;AAED,8BAAsB,SAAU,YAAW,qBAAqB,CAAC,QAAQ,CAAC;;IACxE,SAAS,CAAC,MAAM,CAAC,QAAQ,CAAC,cAAc,gBAA4B;IACpE,SAAS,CAAC,KAAK,mEAA0E;IACzF,SAAS,CAAC,KAAK,+BAAsC;IACrD,SAAS,CAAC,MAAM,+BAAsC;IACtD,SAAS,CAAC,MAAM,UAAS;gBAIb,GAAG,EAAE,GAAG;cAKJ,cAAc;IAiC9B,SAAS,CAAC,KAAK,EAAE,UAAU;IAU3B,KAAK;IAUL,QAAQ;IAUR,IAAI,IAAI,OAAO,CAAC,cAAc,CAAC,QAAQ,CAAC,CAAC;IAIzC,KAAK;IAOL,CAAC,MAAM,CAAC,aAAa,CAAC,IAAI,SAAS;CAGpC"}

package/dist/vad.js

@@ -1,13 +1,16 @@
+import { EventEmitter } from "node:events";
 import { AsyncIterableQueue } from "./utils.js";
 var VADEventType = /* @__PURE__ */ ((VADEventType2) => {
   VADEventType2[VADEventType2["START_OF_SPEECH"] = 0] = "START_OF_SPEECH";
   VADEventType2[VADEventType2["INFERENCE_DONE"] = 1] = "INFERENCE_DONE";
   VADEventType2[VADEventType2["END_OF_SPEECH"] = 2] = "END_OF_SPEECH";
+  VADEventType2[VADEventType2["METRICS_COLLECTED"] = 3] = "METRICS_COLLECTED";
   return VADEventType2;
 })(VADEventType || {});
-class VAD {
+class VAD extends EventEmitter {
   #capabilities;
   constructor(capabilities) {
+    super();
     this.#capabilities = capabilities;
   }
   get capabilities() {
@@ -18,7 +21,44 @@ class VADStream {
   static FLUSH_SENTINEL = Symbol("FLUSH_SENTINEL");
   input = new AsyncIterableQueue();
   queue = new AsyncIterableQueue();
+  output = new AsyncIterableQueue();
   closed = false;
+  #vad;
+  #lastActivityTime = BigInt(0);
+  constructor(vad) {
+    this.#vad = vad;
+    this.monitorMetrics();
+  }
+  async monitorMetrics() {
+    let inferenceDurationTotal = 0;
+    let inferenceCount = 0;
+    for await (const event of this.queue) {
+      this.output.put(event);
+      switch (event.type) {
+        case 0 /* START_OF_SPEECH */:
+          inferenceCount++;
+          if (inferenceCount >= 1 / this.#vad.capabilities.updateInterval) {
+            this.#vad.emit(3 /* METRICS_COLLECTED */, {
+              timestamp: Date.now(),
+              idleTime: Math.trunc(
+                Number((process.hrtime.bigint() - this.#lastActivityTime) / BigInt(1e6))
+              ),
+              inferenceDurationTotal,
+              inferenceCount,
+              label: this.#vad.label
+            });
+            inferenceCount = 0;
+            inferenceDurationTotal = 0;
+          }
+          break;
+        case 1 /* INFERENCE_DONE */:
+        case 2 /* END_OF_SPEECH */:
+          this.#lastActivityTime = process.hrtime.bigint();
+          break;
+      }
+    }
+    this.output.close();
+  }
   pushFrame(frame) {
     if (this.input.closed) {
       throw new Error("Input is closed");
@@ -47,11 +87,12 @@ class VADStream {
     this.input.close();
   }
   next() {
-    return this.queue.next();
+    return this.output.next();
   }
   close() {
     this.input.close();
     this.queue.close();
+    this.output.close();
     this.closed = true;
   }
   [Symbol.asyncIterator]() {