use-voice-control 0.1.85 → 0.1.87

package/dist/client.js.map

@@ -1 +1 @@
-{"version":3,"file":"client.js","sources":["../speech/utils/semantic-split.js","../speech/client/read-aloud.ts","../speech/client/live-transcriber.ts"],"sourcesContent":["/**\n * @fileoverview Splits long text into TTS-sized chunks along paragraph, sentence, comma, and word boundaries so speech synthesis can process it incrementally.\n *\n * Exports `splitTextSmart` (primary chunker, falling back to `splitLongSentence` for\n * oversized sentences) and `splitLongSentence`. Also retains an older, unused\n * `splitTextSmartOld` variant that lacks the long-sentence fallback.\n */\nexport function splitTextSmart(text, maxChunkLength = 500) {\n  const paragraphChunks = text.split(/\\n\\s*\\n/);\n  const finalChunks = [];\n\n  for (let para of paragraphChunks) {\n    if (para.length <= maxChunkLength) {\n      finalChunks.push(para.trim());\n      continue;\n    }\n\n    const sentenceRegex = /(?<=[.?!])(?=\\s+[\"“”'a-z])/gi;\n    const sentences = para.split(sentenceRegex);\n\n    let chunk = '';\n    for (let sentence of sentences) {\n      sentence = sentence.trim();\n\n      if (sentence.length > maxChunkLength) {\n        // Sentence too long — fallback split\n        const subChunks = splitLongSentence(sentence, maxChunkLength);\n        for (let sub of subChunks) {\n          if ((chunk + ' ' + sub).length > maxChunkLength) {\n            if (chunk) finalChunks.push(chunk.trim());\n            chunk = sub;\n          } else {\n            chunk += (chunk ? ' ' : '') + sub;\n          }\n        }\n        continue;\n      }\n\n      if ((chunk + ' ' + sentence).length > maxChunkLength) {\n        if (chunk) finalChunks.push(chunk.trim());\n        chunk = sentence;\n      } else {\n        chunk += (chunk ? ' ' : '') + sentence;\n      }\n    }\n    if (chunk) finalChunks.push(chunk.trim());\n  }\n\n  return finalChunks;\n}\n\nexport function splitLongSentence(sentence, maxLen) {\n  const chunks = [];\n  let current = '';\n\n  const commaParts = sentence.split(/,\\s*/);\n  for (let part of commaParts) {\n    if ((current + ', ' + part).length > maxLen) {\n      if (current) chunks.push(current.trim());\n      if (part.length > maxLen) {\n        const words = part.split(/\\s+/);\n        let wordChunk = '';\n        for (let word of words) {\n          if ((wordChunk + ' ' + word).length > maxLen) {\n            if (wordChunk) chunks.push(wordChunk.trim());\n            wordChunk = word;\n          } else {\n            wordChunk += (wordChunk ? ' ' : '') + word;\n          }\n        }\n        if (wordChunk) chunks.push(wordChunk.trim());\n        current = '';\n      } else {\n        current = part;\n      }\n    } else {\n      current += (current ? ', ' : '') + part;\n    }\n  }\n  if (current) chunks.push(current.trim());\n\n  return chunks;\n}\n\n\nfunction splitTextSmartOld(text, maxChunkLength = 500) {\n  const paragraphChunks = text.split(/\\n\\s*\\n/); // Step 1: split on double returns\n  const finalChunks = [];\n\n  for (let para of paragraphChunks) {\n    if (para.length <= maxChunkLength) {\n      finalChunks.push(para.trim());\n      continue;\n    }\n\n    // Step 2: Further split on sentence boundaries if too long\n    const sentenceRegex = /(?<=[.?!])(?=\\s+[\"“”'a-z])/gi;\n    const sentences = para.split(sentenceRegex);\n\n    let chunk = '';\n    for (let sentence of sentences) {\n      sentence = sentence.trim();\n      if ((chunk + ' ' + sentence).length > maxChunkLength) {\n        if (chunk) finalChunks.push(chunk.trim());\n        chunk = sentence;\n      } else {\n        chunk += (chunk ? ' ' : '') + sentence;\n      }\n    }\n    if (chunk) finalChunks.push(chunk.trim());\n  }\n\n  return finalChunks;\n}\n","/**\n * @fileoverview Browser-side \"read aloud\" playback engine.\n *\n * `ReadAloudController` takes a block of text, chunks it along sentence/paragraph\n * boundaries with `splitTextSmart`, synthesizes each chunk (Kokoro by default, via\n * the package's `/api/speech/tts` route) and plays the chunks back-to-back while\n * pre-fetching the next one, so playback starts after the first short chunk rather\n * than after the whole document. Pause/resume/stop are supported throughout, and\n * each chunk is reported to `onChunk` so callers can highlight what is being spoken.\n *\n * When no TTS endpoint is reachable — a host app that has not mounted the route, or\n * an offline build — the controller falls back to the browser's built-in\n * `speechSynthesis` rather than failing, so the feature still works everywhere.\n */\nimport type { TTSProvider } from \"../types/types\";\nimport { splitTextSmart } from \"../utils/semantic-split.js\";\n\nexport type ReadAloudState = \"idle\" | \"loading\" | \"speaking\" | \"paused\";\n\nexport interface ReadAloudChunk {\n  /** The text of the chunk currently being spoken. */\n  text: string;\n  /** Zero-based position of this chunk in the queue. */\n  index: number;\n  /** Total number of chunks queued for this utterance. */\n  total: number;\n}\n\n/** Turns a chunk of text into playable audio. Return `null` to defer to `speechSynthesis`. */\nexport type SynthesizeFn = (\n  text: string,\n  signal: AbortSignal\n) => Promise<Blob | null>;\n\nexport interface ReadAloudOptions {\n  /** TTS provider passed to the endpoint. Default `kokoro`. */\n  provider?: TTSProvider;\n  /** Provider-specific voice id. Default `af_heart`. */\n  voice?: string;\n  /** HTTP route that synthesizes text. Default `/api/speech/tts`. */\n  endpoint?: string;\n  /**\n   * Target chunk size in characters. Smaller starts talking sooner but makes the\n   * seams between chunks more audible. Default 240.\n   */\n  maxChunkLength?: number;\n  /** Override synthesis entirely (tests, a bring-your-own-TTS host app). */\n  synthesize?: SynthesizeFn;\n  /** Called as each chunk starts playing. */\n  onChunk?: (chunk: ReadAloudChunk) => void;\n  /** Called on every state transition. */\n  onStateChange?: (state: ReadAloudState) => void;\n  /** Called when playback ends, either by running out of chunks or by `stop()`. */\n  onEnd?: (reason: \"finished\" | \"stopped\") => void;\n  /** Called when synthesis or playback fails irrecoverably. */\n  onError?: (error: Error) => void;\n}\n\nconst DEFAULT_ENDPOINT = \"/api/speech/tts\";\nconst DEFAULT_VOICE = \"af_heart\";\nconst DEFAULT_MAX_CHUNK = 240;\n\n/** Resolves once `signal` aborts. Used to race playback against `stop()`. */\nfunction abortPromise(signal: AbortSignal): Promise<void> {\n  if (signal.aborted) return Promise.resolve();\n  return new Promise((resolve) => {\n    signal.addEventListener(\"abort\", () => resolve(), { once: true });\n  });\n}\n\nexport class ReadAloudController {\n  private options: ReadAloudOptions;\n  private state: ReadAloudState = \"idle\";\n  private abortController: AbortController | null = null;\n  private audio: HTMLAudioElement | null = null;\n  private objectUrl: string | null = null;\n  /** Set once the endpoint has proved unreachable, so later chunks skip the retry. */\n  private endpointUnavailable = false;\n\n  constructor(options: ReadAloudOptions = {}) {\n    this.options = options;\n  }\n\n  /** Replace the options (voice, callbacks, …) without discarding playback state. */\n  setOptions(options: ReadAloudOptions): void {\n    this.options = options;\n  }\n\n  getState(): ReadAloudState {\n    return this.state;\n  }\n\n  isActive(): boolean {\n    return this.state === \"speaking\" || this.state === \"paused\" || this.state === \"loading\";\n  }\n\n  /**\n   * Speak `text`, cancelling anything already playing. Resolves when playback\n   * finishes or is stopped — it never rejects; failures go to `onError`.\n   */\n  async speak(text: string): Promise<void> {\n    this.stop();\n\n    const maxChunkLength = this.options.maxChunkLength ?? DEFAULT_MAX_CHUNK;\n    const chunks = splitTextSmart(text ?? \"\", maxChunkLength)\n      .map((chunk) => chunk.trim())\n      .filter((chunk) => chunk.length > 0);\n\n    if (chunks.length === 0) return;\n\n    const abortController = new AbortController();\n    this.abortController = abortController;\n    const { signal } = abortController;\n    this.setState(\"loading\");\n\n    try {\n      // Kick off the first chunk, then always keep exactly one chunk in flight\n      // ahead of the one playing.\n      let upcoming = this.synthesize(chunks[0], signal);\n\n      for (let index = 0; index < chunks.length; index += 1) {\n        const current = upcoming;\n        upcoming =\n          index + 1 < chunks.length\n            ? this.synthesize(chunks[index + 1], signal)\n            : Promise.resolve(null);\n\n        const blob = await current;\n        if (signal.aborted) return;\n\n        // Announce the state before the chunk so a listener reacting to\n        // `onChunk` sees the controller already speaking.\n        this.setState(\"speaking\");\n        this.options.onChunk?.({ text: chunks[index], index, total: chunks.length });\n        if (signal.aborted) return;\n        await this.playChunk(blob, chunks[index], signal);\n        if (signal.aborted) return;\n      }\n\n      this.cleanup();\n      this.setState(\"idle\");\n      this.options.onEnd?.(\"finished\");\n    } catch (error) {\n      if (signal.aborted) return;\n      this.cleanup();\n      this.setState(\"idle\");\n      this.options.onError?.(\n        error instanceof Error ? error : new Error(String(error))\n      );\n      this.options.onEnd?.(\"stopped\");\n    } finally {\n      if (this.abortController === abortController) {\n        this.abortController = null;\n      }\n    }\n  }\n\n  pause(): void {\n    if (this.state !== \"speaking\") return;\n    if (this.audio) {\n      this.audio.pause();\n    } else if (typeof speechSynthesis !== \"undefined\") {\n      speechSynthesis.pause();\n    }\n    this.setState(\"paused\");\n  }\n\n  resume(): void {\n    if (this.state !== \"paused\") return;\n    if (this.audio) {\n      void this.audio.play().catch(() => undefined);\n    } else if (typeof speechSynthesis !== \"undefined\") {\n      speechSynthesis.resume();\n    }\n    this.setState(\"speaking\");\n  }\n\n  /** Stop playback and drop any queued chunks. Safe to call when already idle. */\n  stop(): void {\n    const wasActive = this.isActive();\n    this.abortController?.abort();\n    this.abortController = null;\n    this.cleanup();\n    if (wasActive) {\n      this.setState(\"idle\");\n      this.options.onEnd?.(\"stopped\");\n    }\n  }\n\n  private setState(state: ReadAloudState): void {\n    if (this.state === state) return;\n    this.state = state;\n    this.options.onStateChange?.(state);\n  }\n\n  private cleanup(): void {\n    if (this.audio) {\n      this.audio.pause();\n      this.audio.src = \"\";\n      this.audio = null;\n    }\n    if (this.objectUrl) {\n      URL.revokeObjectURL(this.objectUrl);\n      this.objectUrl = null;\n    }\n    if (typeof speechSynthesis !== \"undefined\") {\n      speechSynthesis.cancel();\n    }\n  }\n\n  private synthesize(text: string, signal: AbortSignal): Promise<Blob | null> {\n    const promise = this.options.synthesize\n      ? this.options.synthesize(text, signal)\n      : this.fetchAudio(text, signal);\n    // Pre-fetched chunks are awaited a beat later; mark them handled now so a\n    // stop() mid-flight does not surface as an unhandled rejection.\n    promise.catch(() => undefined);\n    return promise;\n  }\n\n  private async fetchAudio(text: string, signal: AbortSignal): Promise<Blob | null> {\n    if (this.endpointUnavailable) return null;\n\n    const endpoint = this.options.endpoint ?? DEFAULT_ENDPOINT;\n    try {\n      const response = await fetch(endpoint, {\n        method: \"POST\",\n        headers: { \"Content-Type\": \"application/json\" },\n        body: JSON.stringify({\n          text,\n          provider: this.options.provider ?? \"kokoro\",\n          voice: this.options.voice ?? DEFAULT_VOICE,\n        }),\n        signal,\n      });\n\n      if (!response.ok) throw new Error(`TTS failed: ${response.status}`);\n\n      const contentType = response.headers.get(\"Content-Type\") || \"audio/wav\";\n      return new Blob([await response.arrayBuffer()], { type: contentType });\n    } catch (error) {\n      if (signal.aborted) throw error;\n      // No usable endpoint in this host app — remember it and let every\n      // remaining chunk go straight to the browser's own synthesizer.\n      this.endpointUnavailable = true;\n      return null;\n    }\n  }\n\n  private playChunk(\n    blob: Blob | null,\n    text: string,\n    signal: AbortSignal\n  ): Promise<void> {\n    return blob\n      ? this.playAudioBlob(blob, signal)\n      : this.playWithSpeechSynthesis(text, signal);\n  }\n\n  private async playAudioBlob(blob: Blob, signal: AbortSignal): Promise<void> {\n    const url = URL.createObjectURL(blob);\n    const audio = new Audio(url);\n    this.audio = audio;\n    this.objectUrl = url;\n\n    const finished = new Promise<void>((resolve, reject) => {\n      audio.addEventListener(\"ended\", () => resolve(), { once: true });\n      audio.addEventListener(\n        \"error\",\n        () => reject(new Error(\"Audio playback failed\")),\n        { once: true }\n      );\n    });\n\n    try {\n      await audio.play();\n      await Promise.race([finished, abortPromise(signal)]);\n    } finally {\n      if (this.audio === audio) this.audio = null;\n      if (this.objectUrl === url) this.objectUrl = null;\n      audio.pause();\n      URL.revokeObjectURL(url);\n    }\n  }\n\n  private async playWithSpeechSynthesis(\n    text: string,\n    signal: AbortSignal\n  ): Promise<void> {\n    if (typeof speechSynthesis === \"undefined\" || typeof SpeechSynthesisUtterance === \"undefined\") {\n      throw new Error(\"No speech synthesis available in this browser\");\n    }\n\n    const utterance = new SpeechSynthesisUtterance(text);\n    const finished = new Promise<void>((resolve, reject) => {\n      utterance.onend = () => resolve();\n      // `cancel()` fires an error event; a stop is not a failure.\n      utterance.onerror = () =>\n        signal.aborted ? resolve() : reject(new Error(\"Speech synthesis failed\"));\n    });\n\n    speechSynthesis.speak(utterance);\n    await Promise.race([finished, abortPromise(signal)]);\n  }\n}\n","/**\n * @fileoverview Browser-side live dictation engine.\n *\n * `LiveTranscriber` streams microphone audio to a recognizer and reports two kinds\n * of text: `onPartial`, the in-progress guess that keeps changing while a phrase is\n * being spoken, and `onCommit`, a phrase the recognizer has settled on. Callers use\n * the first to type text into an input as it is being said, and the second to\n * finalize it.\n *\n * Two engines are supported. The browser's own `SpeechRecognition` is preferred when\n * present because it needs no model download; otherwise the package's bundled\n * Moonshine model runs the recognition entirely on-device. Chromium's recognizer\n * stops itself after a pause, so a session that is still meant to be listening is\n * restarted automatically.\n */\n\nexport type TranscriberEngine = \"auto\" | \"webspeech\" | \"moonshine\";\n\nexport interface LiveTranscriberOptions {\n  /** Which recognizer to use. Default `auto` (browser first, Moonshine as fallback). */\n  engine?: TranscriberEngine;\n  /** BCP-47 language tag for the browser recognizer. Default `en-US`. */\n  language?: string;\n  /** Moonshine model name. Default `model/small`. */\n  model?: string;\n  /** Fires continuously with the current in-progress phrase. */\n  onPartial?: (text: string) => void;\n  /** Fires once per phrase the recognizer has settled on. */\n  onCommit?: (text: string) => void;\n  /** Fires whenever the microphone starts or stops. */\n  onStateChange?: (listening: boolean) => void;\n  onError?: (error: Error) => void;\n}\n\ntype SpeechRecognitionCtor = new () => any;\n\nfunction getSpeechRecognition(): SpeechRecognitionCtor | null {\n  if (typeof window === \"undefined\") return null;\n  return (\n    (window as any).SpeechRecognition ||\n    (window as any).webkitSpeechRecognition ||\n    null\n  );\n}\n\n/** True when this browser can dictate at all (either engine). */\nexport function isTranscriptionSupported(): boolean {\n  if (typeof window === \"undefined\") return false;\n  if (getSpeechRecognition()) return true;\n  return !!navigator.mediaDevices?.getUserMedia;\n}\n\nexport class LiveTranscriber {\n  private options: LiveTranscriberOptions;\n  private listening = false;\n  /** Distinguishes a deliberate `stop()` from Chromium's idle auto-stop. */\n  private stopRequested = false;\n  private recognition: any = null;\n  private moonshine: any = null;\n\n  constructor(options: LiveTranscriberOptions = {}) {\n    this.options = options;\n  }\n\n  setOptions(options: LiveTranscriberOptions): void {\n    this.options = options;\n  }\n\n  isListening(): boolean {\n    return this.listening;\n  }\n\n  async start(): Promise<void> {\n    if (this.listening) return;\n    this.stopRequested = false;\n\n    const engine = this.options.engine ?? \"auto\";\n    const Recognition = getSpeechRecognition();\n\n    try {\n      if (engine !== \"moonshine\" && Recognition) {\n        this.startWebSpeech(Recognition);\n        return;\n      }\n      if (engine === \"webspeech\") {\n        throw new Error(\"Speech recognition is not available in this browser\");\n      }\n      await this.startMoonshine();\n    } catch (error) {\n      this.setListening(false);\n      this.options.onError?.(\n        error instanceof Error ? error : new Error(String(error))\n      );\n    }\n  }\n\n  async stop(): Promise<void> {\n    this.stopRequested = true;\n\n    if (this.recognition) {\n      try {\n        this.recognition.stop();\n      } catch {\n        /* already stopped */\n      }\n      this.recognition = null;\n    }\n\n    if (this.moonshine) {\n      try {\n        await this.moonshine.stop?.();\n      } catch {\n        /* already stopped */\n      }\n      this.moonshine = null;\n    }\n\n    this.setListening(false);\n  }\n\n  async toggle(): Promise<void> {\n    if (this.listening) {\n      await this.stop();\n    } else {\n      await this.start();\n    }\n  }\n\n  private setListening(listening: boolean): void {\n    if (this.listening === listening) return;\n    this.listening = listening;\n    this.options.onStateChange?.(listening);\n  }\n\n  private startWebSpeech(Recognition: SpeechRecognitionCtor): void {\n    const recognition = new Recognition();\n    recognition.continuous = true;\n    recognition.interimResults = true;\n    recognition.lang = this.options.language ?? \"en-US\";\n\n    recognition.onstart = () => this.setListening(true);\n\n    recognition.onresult = (event: any) => {\n      let interim = \"\";\n      for (let i = event.resultIndex; i < event.results.length; i += 1) {\n        const result = event.results[i];\n        const text = String(result[0]?.transcript ?? \"\").trim();\n        if (!text) continue;\n        if (result.isFinal) {\n          // No `onPartial(\"\")` first: a commit supersedes the interim phrase, and\n          // clearing it separately would make consumers that write the interim\n          // into a document erase and re-insert the same words.\n          this.options.onCommit?.(text);\n        } else {\n          interim += `${interim ? \" \" : \"\"}${text}`;\n        }\n      }\n      if (interim) this.options.onPartial?.(interim);\n    };\n\n    recognition.onerror = (event: any) => {\n      const code = event?.error;\n      // `no-speech` and `aborted` are routine during a long dictation session;\n      // `onend` restarts the recognizer for us.\n      if (code === \"no-speech\" || code === \"aborted\") return;\n      this.stopRequested = true;\n      this.options.onError?.(new Error(`Speech recognition error: ${code}`));\n    };\n\n    recognition.onend = () => {\n      if (this.stopRequested || this.recognition !== recognition) {\n        this.recognition = null;\n        this.setListening(false);\n        return;\n      }\n      // Chromium ends the session after a pause — start a fresh one so the user\n      // can keep dictating without touching the button again.\n      try {\n        recognition.start();\n      } catch {\n        this.recognition = null;\n        this.setListening(false);\n      }\n    };\n\n    this.recognition = recognition;\n    recognition.start();\n  }\n\n  private async startMoonshine(): Promise<void> {\n    const Moonshine = await import(\"@moonshine-ai/moonshine-js\");\n\n    const transcriber = new Moonshine.MicrophoneTranscriber(\n      this.options.model ?? \"model/small\",\n      {\n        onTranscriptionUpdated: (text: string) => {\n          this.options.onPartial?.(String(text ?? \"\").trim());\n        },\n        onTranscriptionCommitted: (text: string) => {\n          const committed = String(text ?? \"\").trim();\n          if (committed) this.options.onCommit?.(committed);\n          else this.options.onPartial?.(\"\");\n        },\n      },\n      false // streaming mode\n    );\n\n    this.moonshine = transcriber;\n    await transcriber.start();\n    if (this.stopRequested) {\n      await this.stop();\n      return;\n    }\n    this.setListening(true);\n  }\n}\n"],"names":["splitTextSmart","text","maxChunkLength","paragraphChunks","finalChunks","para","sentenceRegex","sentences","chunk","sentence","subChunks","splitLongSentence","sub","maxLen","chunks","current","commaParts","part","words","wordChunk","word","DEFAULT_ENDPOINT","DEFAULT_VOICE","DEFAULT_MAX_CHUNK","abortPromise","signal","resolve","ReadAloudController","options","__publicField","_a","_b","_c","_d","_e","_f","_g","_h","abortController","upcoming","index","blob","error","wasActive","state","promise","endpoint","response","contentType","url","audio","finished","reject","utterance","getSpeechRecognition","isTranscriptionSupported","LiveTranscriber","engine","Recognition","listening","recognition","event","interim","i","result","code","Moonshine","transcriber","committed"],"mappings":";;;AAOO,SAASA,EAAeC,GAAMC,IAAiB,KAAK;AACzD,QAAMC,IAAkBF,EAAK,MAAM,SAAS,GACtCG,IAAc,CAAA;AAEpB,WAASC,KAAQF,GAAiB;AAChC,QAAIE,EAAK,UAAUH,GAAgB;AACjC,MAAAE,EAAY,KAAKC,EAAK,MAAM;AAC5B;AAAA,IACF;AAEA,UAAMC,IAAgB,WAAA,+BAAA,IAA8B,GAC9CC,IAAYF,EAAK,MAAMC,CAAa;AAE1C,QAAIE,IAAQ;AACZ,aAASC,KAAYF,GAAW;AAG9B,UAFAE,IAAWA,EAAS,KAAI,GAEpBA,EAAS,SAASP,GAAgB;AAEpC,cAAMQ,IAAYC,EAAkBF,GAAUP,CAAc;AAC5D,iBAASU,KAAOF;AACd,WAAKF,IAAQ,MAAMI,GAAK,SAASV,KAC3BM,KAAOJ,EAAY,KAAKI,EAAM,KAAI,CAAE,GACxCA,IAAQI,KAERJ,MAAUA,IAAQ,MAAM,MAAMI;AAGlC;AAAA,MACF;AAEA,OAAKJ,IAAQ,MAAMC,GAAU,SAASP,KAChCM,KAAOJ,EAAY,KAAKI,EAAM,KAAI,CAAE,GACxCA,IAAQC,KAERD,MAAUA,IAAQ,MAAM,MAAMC;AAAA,IAElC;AACA,IAAID,KAAOJ,EAAY,KAAKI,EAAM,KAAI,CAAE;AAAA,EAC1C;AAEA,SAAOJ;AACT;AAEO,SAASO,EAAkBF,GAAUI,GAAQ;AAClD,QAAMC,IAAS,CAAA;AACf,MAAIC,IAAU;AAEd,QAAMC,IAAaP,EAAS,MAAM,MAAM;AACxC,WAASQ,KAAQD;AACf,SAAKD,IAAU,OAAOE,GAAM,SAASJ;AAEnC,UADIE,KAASD,EAAO,KAAKC,EAAQ,KAAI,CAAE,GACnCE,EAAK,SAASJ,GAAQ;AACxB,cAAMK,IAAQD,EAAK,MAAM,KAAK;AAC9B,YAAIE,IAAY;AAChB,iBAASC,KAAQF;AACf,WAAKC,IAAY,MAAMC,GAAM,SAASP,KAChCM,KAAWL,EAAO,KAAKK,EAAU,KAAI,CAAE,GAC3CA,IAAYC,KAEZD,MAAcA,IAAY,MAAM,MAAMC;AAG1C,QAAID,KAAWL,EAAO,KAAKK,EAAU,KAAI,CAAE,GAC3CJ,IAAU;AAAA,MACZ;AACE,QAAAA,IAAUE;AAAA;AAGZ,MAAAF,MAAYA,IAAU,OAAO,MAAME;AAGvC,SAAIF,KAASD,EAAO,KAAKC,EAAQ,KAAI,CAAE,GAEhCD;AACT;ACxBA,MAAMO,IAAmB,mBACnBC,IAAgB,YAChBC,IAAoB;AAG1B,SAASC,EAAaC,GAAoC;AACxD,SAAIA,EAAO,UAAgB,QAAQ,QAAA,IAC5B,IAAI,QAAQ,CAACC,MAAY;AAC9B,IAAAD,EAAO,iBAAiB,SAAS,MAAMC,EAAA,GAAW,EAAE,MAAM,IAAM;AAAA,EAClE,CAAC;AACH;AAEO,MAAMC,EAAoB;AAAA,EAS/B,YAAYC,IAA4B,IAAI;AARpC,IAAAC,EAAA;AACA,IAAAA,EAAA,eAAwB;AACxB,IAAAA,EAAA,yBAA0C;AAC1C,IAAAA,EAAA,eAAiC;AACjC,IAAAA,EAAA,mBAA2B;AAE3B;AAAA,IAAAA,EAAA,6BAAsB;AAG5B,SAAK,UAAUD;AAAA,EACjB;AAAA;AAAA,EAGA,WAAWA,GAAiC;AAC1C,SAAK,UAAUA;AAAA,EACjB;AAAA,EAEA,WAA2B;AACzB,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,WAAoB;AAClB,WAAO,KAAK,UAAU,cAAc,KAAK,UAAU,YAAY,KAAK,UAAU;AAAA,EAChF;AAAA;AAAA;AAAA;AAAA;AAAA,EAMA,MAAM,MAAM3B,GAA6B;ADpG3C,QAAA6B,GAAAC,GAAAC,GAAAC,GAAAC,GAAAC,GAAAC,GAAAC;ACqGI,SAAK,KAAA;AAEL,UAAMnC,IAAiB,KAAK,QAAQ,kBAAkBqB,GAChDT,IAASd,EAAeC,KAAQ,IAAIC,CAAc,EACrD,IAAI,CAACM,MAAUA,EAAM,KAAA,CAAM,EAC3B,OAAO,CAACA,MAAUA,EAAM,SAAS,CAAC;AAErC,QAAIM,EAAO,WAAW,EAAG;AAEzB,UAAMwB,IAAkB,IAAI,gBAAA;AAC5B,SAAK,kBAAkBA;AACvB,UAAM,EAAE,QAAAb,MAAWa;AACnB,SAAK,SAAS,SAAS;AAEvB,QAAI;AAGF,UAAIC,IAAW,KAAK,WAAWzB,EAAO,CAAC,GAAGW,CAAM;AAEhD,eAASe,IAAQ,GAAGA,IAAQ1B,EAAO,QAAQ0B,KAAS,GAAG;AACrD,cAAMzB,IAAUwB;AAChB,QAAAA,IACEC,IAAQ,IAAI1B,EAAO,SACf,KAAK,WAAWA,EAAO0B,IAAQ,CAAC,GAAGf,CAAM,IACzC,QAAQ,QAAQ,IAAI;AAE1B,cAAMgB,IAAO,MAAM1B;AASnB,YARIU,EAAO,YAIX,KAAK,SAAS,UAAU,IACxBM,KAAAD,IAAA,KAAK,SAAQ,YAAb,QAAAC,EAAA,KAAAD,GAAuB,EAAE,MAAMhB,EAAO0B,CAAK,GAAG,OAAAA,GAAO,OAAO1B,EAAO,OAAA,IAC/DW,EAAO,aACX,MAAM,KAAK,UAAUgB,GAAM3B,EAAO0B,CAAK,GAAGf,CAAM,GAC5CA,EAAO,SAAS;AAAA,MACtB;AAEA,WAAK,QAAA,GACL,KAAK,SAAS,MAAM,IACpBQ,KAAAD,IAAA,KAAK,SAAQ,UAAb,QAAAC,EAAA,KAAAD,GAAqB;AAAA,IACvB,SAASU,GAAO;AACd,UAAIjB,EAAO,QAAS;AACpB,WAAK,QAAA,GACL,KAAK,SAAS,MAAM,IACpBU,KAAAD,IAAA,KAAK,SAAQ,YAAb,QAAAC,EAAA;AAAA,QAAAD;AAAA,QACEQ,aAAiB,QAAQA,IAAQ,IAAI,MAAM,OAAOA,CAAK,CAAC;AAAA,UAE1DL,KAAAD,IAAA,KAAK,SAAQ,UAAb,QAAAC,EAAA,KAAAD,GAAqB;AAAA,IACvB,UAAA;AACE,MAAI,KAAK,oBAAoBE,MAC3B,KAAK,kBAAkB;AAAA,IAE3B;AAAA,EACF;AAAA,EAEA,QAAc;AACZ,IAAI,KAAK,UAAU,eACf,KAAK,QACP,KAAK,MAAM,MAAA,IACF,OAAO,kBAAoB,OACpC,gBAAgB,MAAA,GAElB,KAAK,SAAS,QAAQ;AAAA,EACxB;AAAA,EAEA,SAAe;AACb,IAAI,KAAK,UAAU,aACf,KAAK,QACF,KAAK,MAAM,KAAA,EAAO,MAAM,MAAA;AAAA,KAAe,IACnC,OAAO,kBAAoB,OACpC,gBAAgB,OAAA,GAElB,KAAK,SAAS,UAAU;AAAA,EAC1B;AAAA;AAAA,EAGA,OAAa;ADlLf,QAAAR,GAAAC,GAAAC;ACmLI,UAAMW,IAAY,KAAK,SAAA;AACvB,KAAAb,IAAA,KAAK,oBAAL,QAAAA,EAAsB,SACtB,KAAK,kBAAkB,MACvB,KAAK,QAAA,GACDa,MACF,KAAK,SAAS,MAAM,IACpBX,KAAAD,IAAA,KAAK,SAAQ,UAAb,QAAAC,EAAA,KAAAD,GAAqB;AAAA,EAEzB;AAAA,EAEQ,SAASa,GAA6B;AD7LhD,QAAAd,GAAAC;AC8LI,IAAI,KAAK,UAAUa,MACnB,KAAK,QAAQA,IACbb,KAAAD,IAAA,KAAK,SAAQ,kBAAb,QAAAC,EAAA,KAAAD,GAA6Bc;AAAA,EAC/B;AAAA,EAEQ,UAAgB;AACtB,IAAI,KAAK,UACP,KAAK,MAAM,MAAA,GACX,KAAK,MAAM,MAAM,IACjB,KAAK,QAAQ,OAEX,KAAK,cACP,IAAI,gBAAgB,KAAK,SAAS,GAClC,KAAK,YAAY,OAEf,OAAO,kBAAoB,OAC7B,gBAAgB,OAAA;AAAA,EAEpB;AAAA,EAEQ,WAAW3C,GAAcwB,GAA2C;AAC1E,UAAMoB,IAAU,KAAK,QAAQ,aACzB,KAAK,QAAQ,WAAW5C,GAAMwB,CAAM,IACpC,KAAK,WAAWxB,GAAMwB,CAAM;AAGhC,WAAAoB,EAAQ,MAAM,MAAA;AAAA,KAAe,GACtBA;AAAA,EACT;AAAA,EAEA,MAAc,WAAW5C,GAAcwB,GAA2C;AAChF,QAAI,KAAK,oBAAqB,QAAO;AAErC,UAAMqB,IAAW,KAAK,QAAQ,YAAYzB;AAC1C,QAAI;AACF,YAAM0B,IAAW,MAAM,MAAMD,GAAU;AAAA,QACrC,QAAQ;AAAA,QACR,SAAS,EAAE,gBAAgB,mBAAA;AAAA,QAC3B,MAAM,KAAK,UAAU;AAAA,UACnB,MAAA7C;AAAA,UACA,UAAU,KAAK,QAAQ,YAAY;AAAA,UACnC,OAAO,KAAK,QAAQ,SAASqB;AAAA,QAAA,CAC9B;AAAA,QACD,QAAAG;AAAA,MAAA,CACD;AAED,UAAI,CAACsB,EAAS,GAAI,OAAM,IAAI,MAAM,eAAeA,EAAS,MAAM,EAAE;AAElE,YAAMC,IAAcD,EAAS,QAAQ,IAAI,cAAc,KAAK;AAC5D,aAAO,IAAI,KAAK,CAAC,MAAMA,EAAS,YAAA,CAAa,GAAG,EAAE,MAAMC,GAAa;AAAA,IACvE,SAASN,GAAO;AACd,UAAIjB,EAAO,QAAS,OAAMiB;AAG1B,kBAAK,sBAAsB,IACpB;AAAA,IACT;AAAA,EACF;AAAA,EAEQ,UACND,GACAxC,GACAwB,GACe;AACf,WAAOgB,IACH,KAAK,cAAcA,GAAMhB,CAAM,IAC/B,KAAK,wBAAwBxB,GAAMwB,CAAM;AAAA,EAC/C;AAAA,EAEA,MAAc,cAAcgB,GAAYhB,GAAoC;AAC1E,UAAMwB,IAAM,IAAI,gBAAgBR,CAAI,GAC9BS,IAAQ,IAAI,MAAMD,CAAG;AAC3B,SAAK,QAAQC,GACb,KAAK,YAAYD;AAEjB,UAAME,IAAW,IAAI,QAAc,CAACzB,GAAS0B,MAAW;AACtD,MAAAF,EAAM,iBAAiB,SAAS,MAAMxB,EAAA,GAAW,EAAE,MAAM,IAAM,GAC/DwB,EAAM;AAAA,QACJ;AAAA,QACA,MAAME,EAAO,IAAI,MAAM,uBAAuB,CAAC;AAAA,QAC/C,EAAE,MAAM,GAAA;AAAA,MAAK;AAAA,IAEjB,CAAC;AAED,QAAI;AACF,YAAMF,EAAM,KAAA,GACZ,MAAM,QAAQ,KAAK,CAACC,GAAU3B,EAAaC,CAAM,CAAC,CAAC;AAAA,IACrD,UAAA;AACE,MAAI,KAAK,UAAUyB,MAAO,KAAK,QAAQ,OACnC,KAAK,cAAcD,MAAK,KAAK,YAAY,OAC7CC,EAAM,MAAA,GACN,IAAI,gBAAgBD,CAAG;AAAA,IACzB;AAAA,EACF;AAAA,EAEA,MAAc,wBACZhD,GACAwB,GACe;AACf,QAAI,OAAO,kBAAoB,OAAe,OAAO,2BAA6B;AAChF,YAAM,IAAI,MAAM,+CAA+C;AAGjE,UAAM4B,IAAY,IAAI,yBAAyBpD,CAAI,GAC7CkD,IAAW,IAAI,QAAc,CAACzB,GAAS0B,MAAW;AACtD,MAAAC,EAAU,QAAQ,MAAM3B,EAAA,GAExB2B,EAAU,UAAU,MAClB5B,EAAO,UAAUC,EAAA,IAAY0B,EAAO,IAAI,MAAM,yBAAyB,CAAC;AAAA,IAC5E,CAAC;AAED,oBAAgB,MAAMC,CAAS,GAC/B,MAAM,QAAQ,KAAK,CAACF,GAAU3B,EAAaC,CAAM,CAAC,CAAC;AAAA,EACrD;AACF;AC5QA,SAAS6B,IAAqD;AAC5D,SAAI,OAAO,SAAW,MAAoB,OAEvC,OAAe,qBACf,OAAe,2BAChB;AAEJ;AAGO,SAASC,IAAoC;AF9CpD,MAAAzB;AE+CE,SAAI,OAAO,SAAW,MAAoB,KACtCwB,EAAA,IAA+B,KAC5B,CAAC,GAACxB,IAAA,UAAU,iBAAV,QAAAA,EAAwB;AACnC;AAEO,MAAM0B,EAAgB;AAAA,EAQ3B,YAAY5B,IAAkC,IAAI;AAP1C,IAAAC,EAAA;AACA,IAAAA,EAAA,mBAAY;AAEZ;AAAA,IAAAA,EAAA,uBAAgB;AAChB,IAAAA,EAAA,qBAAmB;AACnB,IAAAA,EAAA,mBAAiB;AAGvB,SAAK,UAAUD;AAAA,EACjB;AAAA,EAEA,WAAWA,GAAuC;AAChD,SAAK,UAAUA;AAAA,EACjB;AAAA,EAEA,cAAuB;AACrB,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,MAAM,QAAuB;AFxE/B,QAAAE,GAAAC;AEyEI,QAAI,KAAK,UAAW;AACpB,SAAK,gBAAgB;AAErB,UAAM0B,IAAS,KAAK,QAAQ,UAAU,QAChCC,IAAcJ,EAAA;AAEpB,QAAI;AACF,UAAIG,MAAW,eAAeC,GAAa;AACzC,aAAK,eAAeA,CAAW;AAC/B;AAAA,MACF;AACA,UAAID,MAAW;AACb,cAAM,IAAI,MAAM,qDAAqD;AAEvE,YAAM,KAAK,eAAA;AAAA,IACb,SAASf,GAAO;AACd,WAAK,aAAa,EAAK,IACvBX,KAAAD,IAAA,KAAK,SAAQ,YAAb,QAAAC,EAAA;AAAA,QAAAD;AAAA,QACEY,aAAiB,QAAQA,IAAQ,IAAI,MAAM,OAAOA,CAAK,CAAC;AAAA;AAAA,IAE5D;AAAA,EACF;AAAA,EAEA,MAAM,OAAsB;AFhG9B,QAAAZ,GAAAC;AEmGI,QAFA,KAAK,gBAAgB,IAEjB,KAAK,aAAa;AACpB,UAAI;AACF,aAAK,YAAY,KAAA;AAAA,MACnB,QAAQ;AAAA,MAER;AACA,WAAK,cAAc;AAAA,IACrB;AAEA,QAAI,KAAK,WAAW;AAClB,UAAI;AACF,gBAAMA,KAAAD,IAAA,KAAK,WAAU,SAAf,gBAAAC,EAAA,KAAAD;AAAA,MACR,QAAQ;AAAA,MAER;AACA,WAAK,YAAY;AAAA,IACnB;AAEA,SAAK,aAAa,EAAK;AAAA,EACzB;AAAA,EAEA,MAAM,SAAwB;AAC5B,IAAI,KAAK,YACP,MAAM,KAAK,KAAA,IAEX,MAAM,KAAK,MAAA;AAAA,EAEf;AAAA,EAEQ,aAAa6B,GAA0B;AFhIjD,QAAA7B,GAAAC;AEiII,IAAI,KAAK,cAAc4B,MACvB,KAAK,YAAYA,IACjB5B,KAAAD,IAAA,KAAK,SAAQ,kBAAb,QAAAC,EAAA,KAAAD,GAA6B6B;AAAA,EAC/B;AAAA,EAEQ,eAAeD,GAA0C;AAC/D,UAAME,IAAc,IAAIF,EAAA;AACxB,IAAAE,EAAY,aAAa,IACzBA,EAAY,iBAAiB,IAC7BA,EAAY,OAAO,KAAK,QAAQ,YAAY,SAE5CA,EAAY,UAAU,MAAM,KAAK,aAAa,EAAI,GAElDA,EAAY,WAAW,CAACC,MAAe;AF9I3C,UAAA/B,GAAAC,GAAAC,GAAAC,GAAAC;AE+IM,UAAI4B,IAAU;AACd,eAASC,IAAIF,EAAM,aAAaE,IAAIF,EAAM,QAAQ,QAAQE,KAAK,GAAG;AAChE,cAAMC,IAASH,EAAM,QAAQE,CAAC,GACxB9D,IAAO,SAAO6B,IAAAkC,EAAO,CAAC,MAAR,gBAAAlC,EAAW,eAAc,EAAE,EAAE,KAAA;AACjD,QAAK7B,MACD+D,EAAO,WAIThC,KAAAD,IAAA,KAAK,SAAQ,aAAb,QAAAC,EAAA,KAAAD,GAAwB9B,KAExB6D,KAAW,GAAGA,IAAU,MAAM,EAAE,GAAG7D,CAAI;AAAA,MAE3C;AACA,MAAI6D,OAAS5B,KAAAD,IAAA,KAAK,SAAQ,cAAb,QAAAC,EAAA,KAAAD,GAAyB6B;AAAA,IACxC,GAEAF,EAAY,UAAU,CAACC,MAAe;AFhK1C,UAAA/B,GAAAC;AEiKM,YAAMkC,IAAOJ,KAAA,gBAAAA,EAAO;AAGpB,MAAII,MAAS,eAAeA,MAAS,cACrC,KAAK,gBAAgB,KACrBlC,KAAAD,IAAA,KAAK,SAAQ,YAAb,QAAAC,EAAA,KAAAD,GAAuB,IAAI,MAAM,6BAA6BmC,CAAI,EAAE;AAAA,IACtE,GAEAL,EAAY,QAAQ,MAAM;AACxB,UAAI,KAAK,iBAAiB,KAAK,gBAAgBA,GAAa;AAC1D,aAAK,cAAc,MACnB,KAAK,aAAa,EAAK;AACvB;AAAA,MACF;AAGA,UAAI;AACF,QAAAA,EAAY,MAAA;AAAA,MACd,QAAQ;AACN,aAAK,cAAc,MACnB,KAAK,aAAa,EAAK;AAAA,MACzB;AAAA,IACF,GAEA,KAAK,cAAcA,GACnBA,EAAY,MAAA;AAAA,EACd;AAAA,EAEA,MAAc,iBAAgC;AAC5C,UAAMM,IAAY,MAAM,OAAO,4BAA4B,GAErDC,IAAc,IAAID,EAAU;AAAA,MAChC,KAAK,QAAQ,SAAS;AAAA,MACtB;AAAA,QACE,wBAAwB,CAACjE,MAAiB;AFnMlD,cAAA6B,GAAAC;AEoMU,WAAAA,KAAAD,IAAA,KAAK,SAAQ,cAAb,QAAAC,EAAA,KAAAD,GAAyB,OAAO7B,KAAQ,EAAE,EAAE;QAC9C;AAAA,QACA,0BAA0B,CAACA,MAAiB;AFtMpD,cAAA6B,GAAAC,GAAAC,GAAAC;AEuMU,gBAAMmC,IAAY,OAAOnE,KAAQ,EAAE,EAAE,KAAA;AACrC,UAAImE,KAAWrC,KAAAD,IAAA,KAAK,SAAQ,aAAb,QAAAC,EAAA,KAAAD,GAAwBsC,MAClCnC,KAAAD,IAAA,KAAK,SAAQ,cAAb,QAAAC,EAAA,KAAAD,GAAyB;AAAA,QAChC;AAAA,MAAA;AAAA,MAEF;AAAA;AAAA,IAAA;AAKF,QAFA,KAAK,YAAYmC,GACjB,MAAMA,EAAY,MAAA,GACd,KAAK,eAAe;AACtB,YAAM,KAAK,KAAA;AACX;AAAA,IACF;AACA,SAAK,aAAa,EAAI;AAAA,EACxB;AACF;"}
+{"version":3,"file":"client.js","sources":["../speech/client/read-aloud.ts","../speech/client/live-transcriber.ts"],"sourcesContent":["/**\n * @fileoverview Browser-side \"read aloud\" playback engine.\n *\n * `ReadAloudController` takes a block of text, chunks it along sentence/paragraph\n * boundaries with `splitTextSmart`, synthesizes each chunk (Kokoro by default, via\n * the package's `/api/speech/tts` route) and plays the chunks back-to-back while\n * pre-fetching the next one, so playback starts after the first short chunk rather\n * than after the whole document. Pause/resume/stop are supported throughout, and\n * each chunk is reported to `onChunk` so callers can highlight what is being spoken.\n *\n * When no TTS endpoint is reachable — a host app that has not mounted the route, or\n * an offline build — the controller falls back to the browser's built-in\n * `speechSynthesis` rather than failing, so the feature still works everywhere.\n */\nimport type { TTSProvider } from \"../types/types\";\nimport {\n  looksLikeMarkdown,\n  markdownToSpeech,\n  type MarkdownToSpeechOptions,\n} from \"../utils/markdown-to-speech\";\nimport { splitTextSmart } from \"../utils/semantic-split.js\";\n\nexport type ReadAloudState = \"idle\" | \"loading\" | \"speaking\" | \"paused\";\n\nexport interface ReadAloudChunk {\n  /** The text of the chunk currently being spoken. */\n  text: string;\n  /** Zero-based position of this chunk in the queue. */\n  index: number;\n  /** Total number of chunks queued for this utterance. */\n  total: number;\n}\n\n/** Turns a chunk of text into playable audio. Return `null` to defer to `speechSynthesis`. */\nexport type SynthesizeFn = (\n  text: string,\n  signal: AbortSignal\n) => Promise<Blob | null>;\n\nexport interface ReadAloudOptions {\n  /** TTS provider passed to the endpoint. Default `kokoro`. */\n  provider?: TTSProvider;\n  /** Provider-specific voice id. Default `af_heart`. */\n  voice?: string;\n  /** HTTP route that synthesizes text. Default `/api/speech/tts`. */\n  endpoint?: string;\n  /**\n   * Target chunk size in characters. Smaller starts talking sooner but makes the\n   * seams between chunks more audible. Default 240.\n   */\n  maxChunkLength?: number;\n  /**\n   * How to read the text handed to `speak()`. `auto` (default) converts text\n   * that looks like Markdown so \"##\" and \"**\" are not read out; `markdown`\n   * always converts; `text` never does.\n   */\n  format?: \"auto\" | \"markdown\" | \"text\";\n  /** Markdown conversion options, when the text is read as Markdown. */\n  markdown?: MarkdownToSpeechOptions;\n  /** Override synthesis entirely (tests, a bring-your-own-TTS host app). */\n  synthesize?: SynthesizeFn;\n  /** Called as each chunk starts playing. */\n  onChunk?: (chunk: ReadAloudChunk) => void;\n  /** Called on every state transition. */\n  onStateChange?: (state: ReadAloudState) => void;\n  /** Called when playback ends, either by running out of chunks or by `stop()`. */\n  onEnd?: (reason: \"finished\" | \"stopped\") => void;\n  /** Called when synthesis or playback fails irrecoverably. */\n  onError?: (error: Error) => void;\n}\n\nconst DEFAULT_ENDPOINT = \"/api/speech/tts\";\nconst DEFAULT_VOICE = \"af_heart\";\nconst DEFAULT_MAX_CHUNK = 240;\n\n/** Resolves once `signal` aborts. Used to race playback against `stop()`. */\nfunction abortPromise(signal: AbortSignal): Promise<void> {\n  if (signal.aborted) return Promise.resolve();\n  return new Promise((resolve) => {\n    signal.addEventListener(\"abort\", () => resolve(), { once: true });\n  });\n}\n\nexport class ReadAloudController {\n  private options: ReadAloudOptions;\n  private state: ReadAloudState = \"idle\";\n  private abortController: AbortController | null = null;\n  private audio: HTMLAudioElement | null = null;\n  private objectUrl: string | null = null;\n  /** Set once the endpoint has proved unreachable, so later chunks skip the retry. */\n  private endpointUnavailable = false;\n\n  constructor(options: ReadAloudOptions = {}) {\n    this.options = options;\n  }\n\n  /** Replace the options (voice, callbacks, …) without discarding playback state. */\n  setOptions(options: ReadAloudOptions): void {\n    this.options = options;\n  }\n\n  getState(): ReadAloudState {\n    return this.state;\n  }\n\n  isActive(): boolean {\n    return this.state === \"speaking\" || this.state === \"paused\" || this.state === \"loading\";\n  }\n\n  /**\n   * Speak `text`, cancelling anything already playing. Resolves when playback\n   * finishes or is stopped — it never rejects; failures go to `onError`.\n   */\n  async speak(text: string): Promise<void> {\n    this.stop();\n\n    const maxChunkLength = this.options.maxChunkLength ?? DEFAULT_MAX_CHUNK;\n    const chunks = splitTextSmart(this.toSpeakableText(text ?? \"\"), maxChunkLength)\n      .map((chunk) => chunk.trim())\n      .filter((chunk) => chunk.length > 0);\n\n    if (chunks.length === 0) return;\n\n    const abortController = new AbortController();\n    this.abortController = abortController;\n    const { signal } = abortController;\n    this.setState(\"loading\");\n\n    try {\n      // Kick off the first chunk, then always keep exactly one chunk in flight\n      // ahead of the one playing.\n      let upcoming = this.synthesize(chunks[0], signal);\n\n      for (let index = 0; index < chunks.length; index += 1) {\n        const current = upcoming;\n        upcoming =\n          index + 1 < chunks.length\n            ? this.synthesize(chunks[index + 1], signal)\n            : Promise.resolve(null);\n\n        const blob = await current;\n        if (signal.aborted) return;\n\n        // Announce the state before the chunk so a listener reacting to\n        // `onChunk` sees the controller already speaking.\n        this.setState(\"speaking\");\n        this.options.onChunk?.({ text: chunks[index], index, total: chunks.length });\n        if (signal.aborted) return;\n        await this.playChunk(blob, chunks[index], signal);\n        if (signal.aborted) return;\n      }\n\n      this.cleanup();\n      this.setState(\"idle\");\n      this.options.onEnd?.(\"finished\");\n    } catch (error) {\n      if (signal.aborted) return;\n      this.cleanup();\n      this.setState(\"idle\");\n      this.options.onError?.(\n        error instanceof Error ? error : new Error(String(error))\n      );\n      this.options.onEnd?.(\"stopped\");\n    } finally {\n      if (this.abortController === abortController) {\n        this.abortController = null;\n      }\n    }\n  }\n\n  pause(): void {\n    if (this.state !== \"speaking\") return;\n    if (this.audio) {\n      this.audio.pause();\n    } else if (typeof speechSynthesis !== \"undefined\") {\n      speechSynthesis.pause();\n    }\n    this.setState(\"paused\");\n  }\n\n  resume(): void {\n    if (this.state !== \"paused\") return;\n    if (this.audio) {\n      void this.audio.play().catch(() => undefined);\n    } else if (typeof speechSynthesis !== \"undefined\") {\n      speechSynthesis.resume();\n    }\n    this.setState(\"speaking\");\n  }\n\n  /** Stop playback and drop any queued chunks. Safe to call when already idle. */\n  stop(): void {\n    const wasActive = this.isActive();\n    this.abortController?.abort();\n    this.abortController = null;\n    this.cleanup();\n    if (wasActive) {\n      this.setState(\"idle\");\n      this.options.onEnd?.(\"stopped\");\n    }\n  }\n\n  /**\n   * Converts Markdown to spoken words before chunking, so a document read out\n   * of an editor does not have its syntax read back to the listener.\n   */\n  private toSpeakableText(text: string): string {\n    const format = this.options.format ?? \"auto\";\n    if (format === \"text\") return text;\n    if (format === \"markdown\" || looksLikeMarkdown(text)) {\n      return markdownToSpeech(text, this.options.markdown);\n    }\n    return text;\n  }\n\n  private setState(state: ReadAloudState): void {\n    if (this.state === state) return;\n    this.state = state;\n    this.options.onStateChange?.(state);\n  }\n\n  private cleanup(): void {\n    if (this.audio) {\n      this.audio.pause();\n      this.audio.src = \"\";\n      this.audio = null;\n    }\n    if (this.objectUrl) {\n      URL.revokeObjectURL(this.objectUrl);\n      this.objectUrl = null;\n    }\n    if (typeof speechSynthesis !== \"undefined\") {\n      speechSynthesis.cancel();\n    }\n  }\n\n  private synthesize(text: string, signal: AbortSignal): Promise<Blob | null> {\n    const promise = this.options.synthesize\n      ? this.options.synthesize(text, signal)\n      : this.fetchAudio(text, signal);\n    // Pre-fetched chunks are awaited a beat later; mark them handled now so a\n    // stop() mid-flight does not surface as an unhandled rejection.\n    promise.catch(() => undefined);\n    return promise;\n  }\n\n  private async fetchAudio(text: string, signal: AbortSignal): Promise<Blob | null> {\n    if (this.endpointUnavailable) return null;\n\n    const endpoint = this.options.endpoint ?? DEFAULT_ENDPOINT;\n    try {\n      const response = await fetch(endpoint, {\n        method: \"POST\",\n        headers: { \"Content-Type\": \"application/json\" },\n        body: JSON.stringify({\n          text,\n          provider: this.options.provider ?? \"kokoro\",\n          voice: this.options.voice ?? DEFAULT_VOICE,\n        }),\n        signal,\n      });\n\n      if (!response.ok) throw new Error(`TTS failed: ${response.status}`);\n\n      const contentType = response.headers.get(\"Content-Type\") || \"audio/wav\";\n      return new Blob([await response.arrayBuffer()], { type: contentType });\n    } catch (error) {\n      if (signal.aborted) throw error;\n      // No usable endpoint in this host app — remember it and let every\n      // remaining chunk go straight to the browser's own synthesizer.\n      this.endpointUnavailable = true;\n      return null;\n    }\n  }\n\n  private playChunk(\n    blob: Blob | null,\n    text: string,\n    signal: AbortSignal\n  ): Promise<void> {\n    return blob\n      ? this.playAudioBlob(blob, signal)\n      : this.playWithSpeechSynthesis(text, signal);\n  }\n\n  private async playAudioBlob(blob: Blob, signal: AbortSignal): Promise<void> {\n    const url = URL.createObjectURL(blob);\n    const audio = new Audio(url);\n    this.audio = audio;\n    this.objectUrl = url;\n\n    const finished = new Promise<void>((resolve, reject) => {\n      audio.addEventListener(\"ended\", () => resolve(), { once: true });\n      audio.addEventListener(\n        \"error\",\n        () => reject(new Error(\"Audio playback failed\")),\n        { once: true }\n      );\n    });\n\n    try {\n      await audio.play();\n      await Promise.race([finished, abortPromise(signal)]);\n    } finally {\n      if (this.audio === audio) this.audio = null;\n      if (this.objectUrl === url) this.objectUrl = null;\n      audio.pause();\n      URL.revokeObjectURL(url);\n    }\n  }\n\n  private async playWithSpeechSynthesis(\n    text: string,\n    signal: AbortSignal\n  ): Promise<void> {\n    if (typeof speechSynthesis === \"undefined\" || typeof SpeechSynthesisUtterance === \"undefined\") {\n      throw new Error(\"No speech synthesis available in this browser\");\n    }\n\n    const utterance = new SpeechSynthesisUtterance(text);\n    const finished = new Promise<void>((resolve, reject) => {\n      utterance.onend = () => resolve();\n      // `cancel()` fires an error event; a stop is not a failure.\n      utterance.onerror = () =>\n        signal.aborted ? resolve() : reject(new Error(\"Speech synthesis failed\"));\n    });\n\n    speechSynthesis.speak(utterance);\n    await Promise.race([finished, abortPromise(signal)]);\n  }\n}\n","/**\n * @fileoverview Browser-side live dictation engine.\n *\n * `LiveTranscriber` streams microphone audio to a recognizer and reports two kinds\n * of text: `onPartial`, the in-progress guess that keeps changing while a phrase is\n * being spoken, and `onCommit`, a phrase the recognizer has settled on. Callers use\n * the first to type text into an input as it is being said, and the second to\n * finalize it.\n *\n * Two engines are supported. The browser's own `SpeechRecognition` is preferred when\n * present because it needs no model download; otherwise the package's bundled\n * Moonshine model runs the recognition entirely on-device. Chromium's recognizer\n * stops itself after a pause, so a session that is still meant to be listening is\n * restarted automatically.\n */\n\nexport type TranscriberEngine = \"auto\" | \"webspeech\" | \"moonshine\";\n\nexport interface LiveTranscriberOptions {\n  /** Which recognizer to use. Default `auto` (browser first, Moonshine as fallback). */\n  engine?: TranscriberEngine;\n  /** BCP-47 language tag for the browser recognizer. Default `en-US`. */\n  language?: string;\n  /** Moonshine model name. Default `model/small`. */\n  model?: string;\n  /** Fires continuously with the current in-progress phrase. */\n  onPartial?: (text: string) => void;\n  /** Fires once per phrase the recognizer has settled on. */\n  onCommit?: (text: string) => void;\n  /** Fires whenever the microphone starts or stops. */\n  onStateChange?: (listening: boolean) => void;\n  onError?: (error: Error) => void;\n}\n\ntype SpeechRecognitionCtor = new () => any;\n\nfunction getSpeechRecognition(): SpeechRecognitionCtor | null {\n  if (typeof window === \"undefined\") return null;\n  return (\n    (window as any).SpeechRecognition ||\n    (window as any).webkitSpeechRecognition ||\n    null\n  );\n}\n\n/** True when this browser can dictate at all (either engine). */\nexport function isTranscriptionSupported(): boolean {\n  if (typeof window === \"undefined\") return false;\n  if (getSpeechRecognition()) return true;\n  return !!navigator.mediaDevices?.getUserMedia;\n}\n\nexport class LiveTranscriber {\n  private options: LiveTranscriberOptions;\n  private listening = false;\n  /** Distinguishes a deliberate `stop()` from Chromium's idle auto-stop. */\n  private stopRequested = false;\n  private recognition: any = null;\n  private moonshine: any = null;\n\n  constructor(options: LiveTranscriberOptions = {}) {\n    this.options = options;\n  }\n\n  setOptions(options: LiveTranscriberOptions): void {\n    this.options = options;\n  }\n\n  isListening(): boolean {\n    return this.listening;\n  }\n\n  async start(): Promise<void> {\n    if (this.listening) return;\n    this.stopRequested = false;\n\n    const engine = this.options.engine ?? \"auto\";\n    const Recognition = getSpeechRecognition();\n\n    try {\n      if (engine !== \"moonshine\" && Recognition) {\n        this.startWebSpeech(Recognition);\n        return;\n      }\n      if (engine === \"webspeech\") {\n        throw new Error(\"Speech recognition is not available in this browser\");\n      }\n      await this.startMoonshine();\n    } catch (error) {\n      this.setListening(false);\n      this.options.onError?.(\n        error instanceof Error ? error : new Error(String(error))\n      );\n    }\n  }\n\n  async stop(): Promise<void> {\n    this.stopRequested = true;\n\n    if (this.recognition) {\n      try {\n        this.recognition.stop();\n      } catch {\n        /* already stopped */\n      }\n      this.recognition = null;\n    }\n\n    if (this.moonshine) {\n      try {\n        await this.moonshine.stop?.();\n      } catch {\n        /* already stopped */\n      }\n      this.moonshine = null;\n    }\n\n    this.setListening(false);\n  }\n\n  async toggle(): Promise<void> {\n    if (this.listening) {\n      await this.stop();\n    } else {\n      await this.start();\n    }\n  }\n\n  private setListening(listening: boolean): void {\n    if (this.listening === listening) return;\n    this.listening = listening;\n    this.options.onStateChange?.(listening);\n  }\n\n  private startWebSpeech(Recognition: SpeechRecognitionCtor): void {\n    const recognition = new Recognition();\n    recognition.continuous = true;\n    recognition.interimResults = true;\n    recognition.lang = this.options.language ?? \"en-US\";\n\n    recognition.onstart = () => this.setListening(true);\n\n    recognition.onresult = (event: any) => {\n      let interim = \"\";\n      for (let i = event.resultIndex; i < event.results.length; i += 1) {\n        const result = event.results[i];\n        const text = String(result[0]?.transcript ?? \"\").trim();\n        if (!text) continue;\n        if (result.isFinal) {\n          // No `onPartial(\"\")` first: a commit supersedes the interim phrase, and\n          // clearing it separately would make consumers that write the interim\n          // into a document erase and re-insert the same words.\n          this.options.onCommit?.(text);\n        } else {\n          interim += `${interim ? \" \" : \"\"}${text}`;\n        }\n      }\n      if (interim) this.options.onPartial?.(interim);\n    };\n\n    recognition.onerror = (event: any) => {\n      const code = event?.error;\n      // `no-speech` and `aborted` are routine during a long dictation session;\n      // `onend` restarts the recognizer for us.\n      if (code === \"no-speech\" || code === \"aborted\") return;\n      this.stopRequested = true;\n      this.options.onError?.(new Error(`Speech recognition error: ${code}`));\n    };\n\n    recognition.onend = () => {\n      if (this.stopRequested || this.recognition !== recognition) {\n        this.recognition = null;\n        this.setListening(false);\n        return;\n      }\n      // Chromium ends the session after a pause — start a fresh one so the user\n      // can keep dictating without touching the button again.\n      try {\n        recognition.start();\n      } catch {\n        this.recognition = null;\n        this.setListening(false);\n      }\n    };\n\n    this.recognition = recognition;\n    recognition.start();\n  }\n\n  private async startMoonshine(): Promise<void> {\n    const Moonshine = await import(\"@moonshine-ai/moonshine-js\");\n\n    const transcriber = new Moonshine.MicrophoneTranscriber(\n      this.options.model ?? \"model/small\",\n      {\n        onTranscriptionUpdated: (text: string) => {\n          this.options.onPartial?.(String(text ?? \"\").trim());\n        },\n        onTranscriptionCommitted: (text: string) => {\n          const committed = String(text ?? \"\").trim();\n          if (committed) this.options.onCommit?.(committed);\n          else this.options.onPartial?.(\"\");\n        },\n      },\n      false // streaming mode\n    );\n\n    this.moonshine = transcriber;\n    await transcriber.start();\n    if (this.stopRequested) {\n      await this.stop();\n      return;\n    }\n    this.setListening(true);\n  }\n}\n"],"names":["DEFAULT_ENDPOINT","DEFAULT_VOICE","DEFAULT_MAX_CHUNK","abortPromise","signal","resolve","ReadAloudController","options","__publicField","text","maxChunkLength","chunks","splitTextSmart","chunk","abortController","upcoming","index","current","blob","_b","_a","_d","_c","error","_f","_e","_h","_g","wasActive","format","looksLikeMarkdown","markdownToSpeech","state","promise","endpoint","response","contentType","url","audio","finished","reject","utterance","getSpeechRecognition","isTranscriptionSupported","LiveTranscriber","engine","Recognition","listening","recognition","event","interim","i","result","code","Moonshine","transcriber","committed"],"mappings":";;;;;;AAuEA,MAAMA,IAAmB,mBACnBC,IAAgB,YAChBC,IAAoB;AAG1B,SAASC,EAAaC,GAAoC;AACxD,SAAIA,EAAO,UAAgB,QAAQ,QAAA,IAC5B,IAAI,QAAQ,CAACC,MAAY;AAC9B,IAAAD,EAAO,iBAAiB,SAAS,MAAMC,EAAA,GAAW,EAAE,MAAM,IAAM;AAAA,EAClE,CAAC;AACH;AAEO,MAAMC,EAAoB;AAAA,EAS/B,YAAYC,IAA4B,IAAI;AARpC,IAAAC,EAAA;AACA,IAAAA,EAAA,eAAwB;AACxB,IAAAA,EAAA,yBAA0C;AAC1C,IAAAA,EAAA,eAAiC;AACjC,IAAAA,EAAA,mBAA2B;AAE3B;AAAA,IAAAA,EAAA,6BAAsB;AAG5B,SAAK,UAAUD;AAAA,EACjB;AAAA;AAAA,EAGA,WAAWA,GAAiC;AAC1C,SAAK,UAAUA;AAAA,EACjB;AAAA,EAEA,WAA2B;AACzB,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,WAAoB;AAClB,WAAO,KAAK,UAAU,cAAc,KAAK,UAAU,YAAY,KAAK,UAAU;AAAA,EAChF;AAAA;AAAA;AAAA;AAAA;AAAA,EAMA,MAAM,MAAME,GAA6B;;AACvC,SAAK,KAAA;AAEL,UAAMC,IAAiB,KAAK,QAAQ,kBAAkBR,GAChDS,IAASC,EAAe,KAAK,gBAAgBH,KAAQ,EAAE,GAAGC,CAAc,EAC3E,IAAI,CAACG,MAAUA,EAAM,MAAM,EAC3B,OAAO,CAACA,MAAUA,EAAM,SAAS,CAAC;AAErC,QAAIF,EAAO,WAAW,EAAG;AAEzB,UAAMG,IAAkB,IAAI,gBAAA;AAC5B,SAAK,kBAAkBA;AACvB,UAAM,EAAE,QAAAV,MAAWU;AACnB,SAAK,SAAS,SAAS;AAEvB,QAAI;AAGF,UAAIC,IAAW,KAAK,WAAWJ,EAAO,CAAC,GAAGP,CAAM;AAEhD,eAASY,IAAQ,GAAGA,IAAQL,EAAO,QAAQK,KAAS,GAAG;AACrD,cAAMC,IAAUF;AAChB,QAAAA,IACEC,IAAQ,IAAIL,EAAO,SACf,KAAK,WAAWA,EAAOK,IAAQ,CAAC,GAAGZ,CAAM,IACzC,QAAQ,QAAQ,IAAI;AAE1B,cAAMc,IAAO,MAAMD;AASnB,YARIb,EAAO,YAIX,KAAK,SAAS,UAAU,IACxBe,KAAAC,IAAA,KAAK,SAAQ,YAAb,QAAAD,EAAA,KAAAC,GAAuB,EAAE,MAAMT,EAAOK,CAAK,GAAG,OAAAA,GAAO,OAAOL,EAAO,OAAA,IAC/DP,EAAO,aACX,MAAM,KAAK,UAAUc,GAAMP,EAAOK,CAAK,GAAGZ,CAAM,GAC5CA,EAAO,SAAS;AAAA,MACtB;AAEA,WAAK,QAAA,GACL,KAAK,SAAS,MAAM,IACpBiB,KAAAC,IAAA,KAAK,SAAQ,UAAb,QAAAD,EAAA,KAAAC,GAAqB;AAAA,IACvB,SAASC,GAAO;AACd,UAAInB,EAAO,QAAS;AACpB,WAAK,QAAA,GACL,KAAK,SAAS,MAAM,IACpBoB,KAAAC,IAAA,KAAK,SAAQ,YAAb,QAAAD,EAAA;AAAA,QAAAC;AAAA,QACEF,aAAiB,QAAQA,IAAQ,IAAI,MAAM,OAAOA,CAAK,CAAC;AAAA,UAE1DG,KAAAC,IAAA,KAAK,SAAQ,UAAb,QAAAD,EAAA,KAAAC,GAAqB;AAAA,IACvB,UAAA;AACE,MAAI,KAAK,oBAAoBb,MAC3B,KAAK,kBAAkB;AAAA,IAE3B;AAAA,EACF;AAAA,EAEA,QAAc;AACZ,IAAI,KAAK,UAAU,eACf,KAAK,QACP,KAAK,MAAM,MAAA,IACF,OAAO,kBAAoB,OACpC,gBAAgB,MAAA,GAElB,KAAK,SAAS,QAAQ;AAAA,EACxB;AAAA,EAEA,SAAe;AACb,IAAI,KAAK,UAAU,aACf,KAAK,QACF,KAAK,MAAM,KAAA,EAAO,MAAM,MAAA;AAAA,KAAe,IACnC,OAAO,kBAAoB,OACpC,gBAAgB,OAAA,GAElB,KAAK,SAAS,UAAU;AAAA,EAC1B;AAAA;AAAA,EAGA,OAAa;;AACX,UAAMc,IAAY,KAAK,SAAA;AACvB,KAAAR,IAAA,KAAK,oBAAL,QAAAA,EAAsB,SACtB,KAAK,kBAAkB,MACvB,KAAK,QAAA,GACDQ,MACF,KAAK,SAAS,MAAM,IACpBN,KAAAH,IAAA,KAAK,SAAQ,UAAb,QAAAG,EAAA,KAAAH,GAAqB;AAAA,EAEzB;AAAA;AAAA;AAAA;AAAA;AAAA,EAMQ,gBAAgBV,GAAsB;AAC5C,UAAMoB,IAAS,KAAK,QAAQ,UAAU;AACtC,WAAIA,MAAW,SAAepB,IAC1BoB,MAAW,cAAcC,EAAkBrB,CAAI,IAC1CsB,EAAiBtB,GAAM,KAAK,QAAQ,QAAQ,IAE9CA;AAAA,EACT;AAAA,EAEQ,SAASuB,GAA6B;;AAC5C,IAAI,KAAK,UAAUA,MACnB,KAAK,QAAQA,IACbb,KAAAC,IAAA,KAAK,SAAQ,kBAAb,QAAAD,EAAA,KAAAC,GAA6BY;AAAA,EAC/B;AAAA,EAEQ,UAAgB;AACtB,IAAI,KAAK,UACP,KAAK,MAAM,MAAA,GACX,KAAK,MAAM,MAAM,IACjB,KAAK,QAAQ,OAEX,KAAK,cACP,IAAI,gBAAgB,KAAK,SAAS,GAClC,KAAK,YAAY,OAEf,OAAO,kBAAoB,OAC7B,gBAAgB,OAAA;AAAA,EAEpB;AAAA,EAEQ,WAAWvB,GAAcL,GAA2C;AAC1E,UAAM6B,IAAU,KAAK,QAAQ,aACzB,KAAK,QAAQ,WAAWxB,GAAML,CAAM,IACpC,KAAK,WAAWK,GAAML,CAAM;AAGhC,WAAA6B,EAAQ,MAAM,MAAA;AAAA,KAAe,GACtBA;AAAA,EACT;AAAA,EAEA,MAAc,WAAWxB,GAAcL,GAA2C;AAChF,QAAI,KAAK,oBAAqB,QAAO;AAErC,UAAM8B,IAAW,KAAK,QAAQ,YAAYlC;AAC1C,QAAI;AACF,YAAMmC,IAAW,MAAM,MAAMD,GAAU;AAAA,QACrC,QAAQ;AAAA,QACR,SAAS,EAAE,gBAAgB,mBAAA;AAAA,QAC3B,MAAM,KAAK,UAAU;AAAA,UACnB,MAAAzB;AAAA,UACA,UAAU,KAAK,QAAQ,YAAY;AAAA,UACnC,OAAO,KAAK,QAAQ,SAASR;AAAA,QAAA,CAC9B;AAAA,QACD,QAAAG;AAAA,MAAA,CACD;AAED,UAAI,CAAC+B,EAAS,GAAI,OAAM,IAAI,MAAM,eAAeA,EAAS,MAAM,EAAE;AAElE,YAAMC,IAAcD,EAAS,QAAQ,IAAI,cAAc,KAAK;AAC5D,aAAO,IAAI,KAAK,CAAC,MAAMA,EAAS,YAAA,CAAa,GAAG,EAAE,MAAMC,GAAa;AAAA,IACvE,SAASb,GAAO;AACd,UAAInB,EAAO,QAAS,OAAMmB;AAG1B,kBAAK,sBAAsB,IACpB;AAAA,IACT;AAAA,EACF;AAAA,EAEQ,UACNL,GACAT,GACAL,GACe;AACf,WAAOc,IACH,KAAK,cAAcA,GAAMd,CAAM,IAC/B,KAAK,wBAAwBK,GAAML,CAAM;AAAA,EAC/C;AAAA,EAEA,MAAc,cAAcc,GAAYd,GAAoC;AAC1E,UAAMiC,IAAM,IAAI,gBAAgBnB,CAAI,GAC9BoB,IAAQ,IAAI,MAAMD,CAAG;AAC3B,SAAK,QAAQC,GACb,KAAK,YAAYD;AAEjB,UAAME,IAAW,IAAI,QAAc,CAAClC,GAASmC,MAAW;AACtD,MAAAF,EAAM,iBAAiB,SAAS,MAAMjC,EAAA,GAAW,EAAE,MAAM,IAAM,GAC/DiC,EAAM;AAAA,QACJ;AAAA,QACA,MAAME,EAAO,IAAI,MAAM,uBAAuB,CAAC;AAAA,QAC/C,EAAE,MAAM,GAAA;AAAA,MAAK;AAAA,IAEjB,CAAC;AAED,QAAI;AACF,YAAMF,EAAM,KAAA,GACZ,MAAM,QAAQ,KAAK,CAACC,GAAUpC,EAAaC,CAAM,CAAC,CAAC;AAAA,IACrD,UAAA;AACE,MAAI,KAAK,UAAUkC,MAAO,KAAK,QAAQ,OACnC,KAAK,cAAcD,MAAK,KAAK,YAAY,OAC7CC,EAAM,MAAA,GACN,IAAI,gBAAgBD,CAAG;AAAA,IACzB;AAAA,EACF;AAAA,EAEA,MAAc,wBACZ5B,GACAL,GACe;AACf,QAAI,OAAO,kBAAoB,OAAe,OAAO,2BAA6B;AAChF,YAAM,IAAI,MAAM,+CAA+C;AAGjE,UAAMqC,IAAY,IAAI,yBAAyBhC,CAAI,GAC7C8B,IAAW,IAAI,QAAc,CAAClC,GAASmC,MAAW;AACtD,MAAAC,EAAU,QAAQ,MAAMpC,EAAA,GAExBoC,EAAU,UAAU,MAClBrC,EAAO,UAAUC,EAAA,IAAYmC,EAAO,IAAI,MAAM,yBAAyB,CAAC;AAAA,IAC5E,CAAC;AAED,oBAAgB,MAAMC,CAAS,GAC/B,MAAM,QAAQ,KAAK,CAACF,GAAUpC,EAAaC,CAAM,CAAC,CAAC;AAAA,EACrD;AACF;ACtSA,SAASsC,IAAqD;AAC5D,SAAI,OAAO,SAAW,MAAoB,OAEvC,OAAe,qBACf,OAAe,2BAChB;AAEJ;AAGO,SAASC,IAAoC;;AAClD,SAAI,OAAO,SAAW,MAAoB,KACtCD,EAAA,IAA+B,KAC5B,CAAC,GAACtB,IAAA,UAAU,iBAAV,QAAAA,EAAwB;AACnC;AAEO,MAAMwB,EAAgB;AAAA,EAQ3B,YAAYrC,IAAkC,IAAI;AAP1C,IAAAC,EAAA;AACA,IAAAA,EAAA,mBAAY;AAEZ;AAAA,IAAAA,EAAA,uBAAgB;AAChB,IAAAA,EAAA,qBAAmB;AACnB,IAAAA,EAAA,mBAAiB;AAGvB,SAAK,UAAUD;AAAA,EACjB;AAAA,EAEA,WAAWA,GAAuC;AAChD,SAAK,UAAUA;AAAA,EACjB;AAAA,EAEA,cAAuB;AACrB,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,MAAM,QAAuB;;AAC3B,QAAI,KAAK,UAAW;AACpB,SAAK,gBAAgB;AAErB,UAAMsC,IAAS,KAAK,QAAQ,UAAU,QAChCC,IAAcJ,EAAA;AAEpB,QAAI;AACF,UAAIG,MAAW,eAAeC,GAAa;AACzC,aAAK,eAAeA,CAAW;AAC/B;AAAA,MACF;AACA,UAAID,MAAW;AACb,cAAM,IAAI,MAAM,qDAAqD;AAEvE,YAAM,KAAK,eAAA;AAAA,IACb,SAAStB,GAAO;AACd,WAAK,aAAa,EAAK,IACvBJ,KAAAC,IAAA,KAAK,SAAQ,YAAb,QAAAD,EAAA;AAAA,QAAAC;AAAA,QACEG,aAAiB,QAAQA,IAAQ,IAAI,MAAM,OAAOA,CAAK,CAAC;AAAA;AAAA,IAE5D;AAAA,EACF;AAAA,EAEA,MAAM,OAAsB;;AAG1B,QAFA,KAAK,gBAAgB,IAEjB,KAAK,aAAa;AACpB,UAAI;AACF,aAAK,YAAY,KAAA;AAAA,MACnB,QAAQ;AAAA,MAER;AACA,WAAK,cAAc;AAAA,IACrB;AAEA,QAAI,KAAK,WAAW;AAClB,UAAI;AACF,gBAAMJ,KAAAC,IAAA,KAAK,WAAU,SAAf,gBAAAD,EAAA,KAAAC;AAAA,MACR,QAAQ;AAAA,MAER;AACA,WAAK,YAAY;AAAA,IACnB;AAEA,SAAK,aAAa,EAAK;AAAA,EACzB;AAAA,EAEA,MAAM,SAAwB;AAC5B,IAAI,KAAK,YACP,MAAM,KAAK,KAAA,IAEX,MAAM,KAAK,MAAA;AAAA,EAEf;AAAA,EAEQ,aAAa2B,GAA0B;;AAC7C,IAAI,KAAK,cAAcA,MACvB,KAAK,YAAYA,IACjB5B,KAAAC,IAAA,KAAK,SAAQ,kBAAb,QAAAD,EAAA,KAAAC,GAA6B2B;AAAA,EAC/B;AAAA,EAEQ,eAAeD,GAA0C;AAC/D,UAAME,IAAc,IAAIF,EAAA;AACxB,IAAAE,EAAY,aAAa,IACzBA,EAAY,iBAAiB,IAC7BA,EAAY,OAAO,KAAK,QAAQ,YAAY,SAE5CA,EAAY,UAAU,MAAM,KAAK,aAAa,EAAI,GAElDA,EAAY,WAAW,CAACC,MAAe;;AACrC,UAAIC,IAAU;AACd,eAASC,IAAIF,EAAM,aAAaE,IAAIF,EAAM,QAAQ,QAAQE,KAAK,GAAG;AAChE,cAAMC,IAASH,EAAM,QAAQE,CAAC,GACxB1C,IAAO,SAAOW,IAAAgC,EAAO,CAAC,MAAR,gBAAAhC,EAAW,eAAc,EAAE,EAAE,KAAA;AACjD,QAAKX,MACD2C,EAAO,WAIT9B,KAAAH,IAAA,KAAK,SAAQ,aAAb,QAAAG,EAAA,KAAAH,GAAwBV,KAExByC,KAAW,GAAGA,IAAU,MAAM,EAAE,GAAGzC,CAAI;AAAA,MAE3C;AACA,MAAIyC,OAASzB,KAAAJ,IAAA,KAAK,SAAQ,cAAb,QAAAI,EAAA,KAAAJ,GAAyB6B;AAAA,IACxC,GAEAF,EAAY,UAAU,CAACC,MAAe;;AACpC,YAAMI,IAAOJ,KAAA,gBAAAA,EAAO;AAGpB,MAAII,MAAS,eAAeA,MAAS,cACrC,KAAK,gBAAgB,KACrBlC,KAAAC,IAAA,KAAK,SAAQ,YAAb,QAAAD,EAAA,KAAAC,GAAuB,IAAI,MAAM,6BAA6BiC,CAAI,EAAE;AAAA,IACtE,GAEAL,EAAY,QAAQ,MAAM;AACxB,UAAI,KAAK,iBAAiB,KAAK,gBAAgBA,GAAa;AAC1D,aAAK,cAAc,MACnB,KAAK,aAAa,EAAK;AACvB;AAAA,MACF;AAGA,UAAI;AACF,QAAAA,EAAY,MAAA;AAAA,MACd,QAAQ;AACN,aAAK,cAAc,MACnB,KAAK,aAAa,EAAK;AAAA,MACzB;AAAA,IACF,GAEA,KAAK,cAAcA,GACnBA,EAAY,MAAA;AAAA,EACd;AAAA,EAEA,MAAc,iBAAgC;AAC5C,UAAMM,IAAY,MAAM,OAAO,4BAA4B,GAErDC,IAAc,IAAID,EAAU;AAAA,MAChC,KAAK,QAAQ,SAAS;AAAA,MACtB;AAAA,QACE,wBAAwB,CAAC7C,MAAiB;;AACxC,WAAAU,KAAAC,IAAA,KAAK,SAAQ,cAAb,QAAAD,EAAA,KAAAC,GAAyB,OAAOX,KAAQ,EAAE,EAAE;QAC9C;AAAA,QACA,0BAA0B,CAACA,MAAiB;;AAC1C,gBAAM+C,IAAY,OAAO/C,KAAQ,EAAE,EAAE,KAAA;AACrC,UAAI+C,KAAWrC,KAAAC,IAAA,KAAK,SAAQ,aAAb,QAAAD,EAAA,KAAAC,GAAwBoC,MAClCnC,KAAAC,IAAA,KAAK,SAAQ,cAAb,QAAAD,EAAA,KAAAC,GAAyB;AAAA,QAChC;AAAA,MAAA;AAAA,MAEF;AAAA;AAAA,IAAA;AAKF,QAFA,KAAK,YAAYiC,GACjB,MAAMA,EAAY,MAAA,GACd,KAAK,eAAe;AACtB,YAAM,KAAK,KAAA;AACX;AAAA,IACF;AACA,SAAK,aAAa,EAAI;AAAA,EACxB;AACF;"}

package/dist/core/kokoro-node.d.ts

@@ -0,0 +1,51 @@
+import { TTSResult } from '../types/types';
+export type KokoroDtype = "fp32" | "fp16" | "q8" | "q4" | "q4f16";
+export type KokoroDevice = "wasm" | "webgpu" | "cpu";
+export interface KokoroNodeOptions {
+    /** Voice id, e.g. `af_heart`. Default `af_heart`. */
+    voice?: string;
+    /** Speaking rate; 1 is the model's natural pace. Default 1. */
+    speed?: number;
+    /** Hugging Face model id. Default `onnx-community/Kokoro-82M-v1.0-ONNX`. */
+    model?: string;
+    /** Weight quantization. `q8` (default) is the best size/quality trade on CPU. */
+    dtype?: KokoroDtype;
+    /** Execution device. Default `cpu`. */
+    device?: KokoroDevice;
+    /**
+     * Target chunk size in characters. Kokoro's context is limited, so long text
+     * is split before synthesis; 400 leaves headroom for phoneme expansion.
+     */
+    maxChunkLength?: number;
+    /** Silence inserted between chunks, in milliseconds. Default 120. */
+    gapMs?: number;
+    /** Model download/loading progress, forwarded from transformers.js. */
+    onModelProgress?: (progress: unknown) => void;
+    /** Called before each chunk is synthesized, for CLI progress output. */
+    onChunk?: (info: {
+        index: number;
+        total: number;
+        text: string;
+    }) => void;
+}
+export interface KokoroAudio {
+    samples: Float32Array;
+    sampleRate: number;
+}
+/**
+ * Loads (and caches) a `kokoro-js` instance.
+ *
+ * `kokoro-js` is an optional dependency: a browser-only consumer should not have
+ * to download onnxruntime-node, so a missing install is reported as an
+ * actionable error rather than a module-resolution stack trace.
+ */
+export declare function loadKokoroTTS(options?: KokoroNodeOptions): Promise<any>;
+/** Drops cached model instances. Mainly useful in tests and long-lived workers. */
+export declare function resetKokoroCache(): void;
+/**
+ * Synthesizes `text` to raw samples, chunking anything longer than the model's
+ * comfortable context.
+ */
+export declare function synthesizeSamples(text: string, options?: KokoroNodeOptions): Promise<KokoroAudio>;
+/** Synthesizes `text` and encodes it as a 16-bit PCM WAV file. */
+export declare function synthesizeWav(text: string, options?: KokoroNodeOptions): Promise<TTSResult>;

package/dist/core/kokoro.d.ts

@@ -1,5 +1,9 @@
 import { TTSResult } from '../types/types';
+import { KokoroNodeOptions } from './kokoro-node';
 /**
- * Generate speech from text using Kokoro
+ * Generate speech from text using Kokoro.
+ *
+ * Long text is chunked and joined automatically. An unrecognised voice falls
+ * back to `af_heart` rather than failing the request.
  */
-export declare function generateKokoroSpeech(text: string, voice?: string): Promise<TTSResult>;
+export declare function generateKokoroSpeech(text: string, voice?: string, options?: Omit<KokoroNodeOptions, "voice">): Promise<TTSResult>;

package/dist/index.d.ts

@@ -1,5 +1,7 @@
 import { TTSOptions, TTSResult } from './types/types';
 export * from './types/types';
+export { looksLikeMarkdown, markdownToSpeech, markdownToSpeechSegments, stripInlineMarkdown, type MarkdownToSpeechOptions, type SpeechSegment, type SpeechSegmentType, } from './utils/markdown-to-speech';
+export { encodeWav, concatSamples, wavDurationSeconds } from './utils/wav';
 /**
  * Generate speech from text using the specified provider
  *

package/dist/index.js

@@ -1,102 +1,59 @@
-const c = [
-  "af_heart",
-  "af_alloy",
-  "af_aoede",
-  "af_bella",
-  "af_jessica",
-  "af_nicole",
-  "af_river",
-  "af_sarah",
-  "af_sky",
-  "am_adam",
-  "am_echo",
-  "am_fable",
-  "am_fenrir",
-  "am_liam",
-  "am_michael",
-  "am_onyx"
-], u = [
-  "angus",
-  "asteria",
-  "arcas",
-  "orion",
-  "orpheus",
-  "athena",
-  "luna",
-  "zeus",
-  "perseus",
-  "helios",
-  "hera",
-  "stella"
-];
-let t = null, n = null;
-async function f() {
-  return t || (n ? (await n, t) : (n = (async () => {
-    try {
-      const e = await import("@huggingface/transformers"), { StyleTextToSpeech2Model: o, AutoTokenizer: a } = e, s = "hexgrad/Kokoro-82M", [i, l] = await Promise.all([
-        o.from_pretrained(s, {
-          device: "cpu",
-          dtype: "q8"
-        }),
-        a.from_pretrained(s)
-      ]);
-      t = { model: i, tokenizer: l }, console.log("[Kokoro] Model loaded successfully");
-    } catch (r) {
-      throw console.error("[Kokoro] Failed to load model:", r), r;
-    } finally {
-      n = null;
-    }
-  })(), await n, t));
-}
-async function d(r, e = "af_heart") {
-  c.includes(e), await f();
-  try {
-    throw new Error("Server-side Kokoro TTS requires additional setup. Use the browser-based implementation via Web Workers.");
-  } catch (o) {
-    throw console.error("[Kokoro] Error generating speech:", o), o;
-  }
+import { K as t, s as a, D as c } from "./kokoro-node-DJ_Rxp_N.js";
+import { c as h, d as k, e as w, w as m } from "./kokoro-node-DJ_Rxp_N.js";
+import { looksLikeMarkdown as v, markdownToSpeech as _, markdownToSpeechSegments as y, stripInlineMarkdown as A } from "./markdown.js";
+async function s(r, e = "af_heart", n = {}) {
+  const o = t.includes(e) ? e : "af_heart";
+  return a(r, { ...n, voice: o });
 }
-function h() {
-  var e, o;
+function i() {
+  var e, n;
   const r = globalThis;
   try {
     if (typeof r.getCloudflareContext == "function") {
-      const a = r.getCloudflareContext();
-      if ((e = a == null ? void 0 : a.env) != null && e.AI) return a.env.AI;
+      const o = r.getCloudflareContext();
+      if ((e = o == null ? void 0 : o.env) != null && e.AI) return o.env.AI;
     }
   } catch {
   }
-  return ((o = r.__env__) == null ? void 0 : o.AI) ?? r.AI;
+  return ((n = r.__env__) == null ? void 0 : n.AI) ?? r.AI;
 }
-async function m(r, e = "angus") {
-  const o = u.includes(e) ? e : "angus", a = h();
-  if (!a)
+async function u(r, e = "angus") {
+  const n = c.includes(e) ? e : "angus", o = i();
+  if (!o)
     throw new Error("Cloudflare AI binding not available");
   return {
-    audio: await a.run("@cf/deepgram/aura-1", {
+    audio: await o.run("@cf/deepgram/aura-1", {
       text: r.slice(0, 5e3),
-      speaker: o,
+      speaker: n,
       encoding: "mp3"
     }),
     contentType: "audio/mpeg"
   };
 }
-async function _(r) {
-  const { text: e, provider: o = "kokoro", voice: a = "af_heart" } = r;
+async function l(r) {
+  const { text: e, provider: n = "kokoro", voice: o = "af_heart" } = r;
   if (!e || typeof e != "string" || e.trim().length === 0)
     throw new Error("Text is required");
-  switch (o) {
+  switch (n) {
     case "kokoro":
-      return d(e, a);
+      return s(e, o);
     case "deepgram":
-      return m(e, a);
+      return u(e, o);
     default:
-      throw new Error(`Unknown TTS provider: ${o}`);
+      throw new Error(`Unknown TTS provider: ${n}`);
   }
 }
 export {
-  u as DEEPGRAM_SPEAKERS,
-  c as KOKORO_VOICES,
-  _ as generateSpeech
+  c as DEEPGRAM_SPEAKERS,
+  t as KOKORO_VOICES,
+  h as concatSamples,
+  k as describeKokoroVoice,
+  w as encodeWav,
+  l as generateSpeech,
+  v as looksLikeMarkdown,
+  _ as markdownToSpeech,
+  y as markdownToSpeechSegments,
+  A as stripInlineMarkdown,
+  m as wavDurationSeconds
 };
 //# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sources":["../speech/types/types.ts","../speech/core/kokoro.ts","../speech/core/deepgram.ts","../speech/index.ts"],"sourcesContent":["/**\n * @fileoverview Type definitions for text-to-speech providers\n */\n\nexport type TTSProvider = \"kokoro\" | \"deepgram\";\n\nexport interface TTSOptions {\n  text: string;\n  provider?: TTSProvider;\n  voice?: string;\n}\n\nexport interface TTSResult {\n  audio: ArrayBuffer;\n  contentType: string;\n}\n\n// Kokoro voices from the model\nexport const KOKORO_VOICES = [\n  \"af_heart\", \"af_alloy\", \"af_aoede\", \"af_bella\",\n  \"af_jessica\", \"af_nicole\", \"af_river\", \"af_sarah\", \"af_sky\",\n  \"am_adam\", \"am_echo\", \"am_fable\", \"am_fenrir\",\n  \"am_liam\", \"am_michael\", \"am_onyx\"\n] as const;\n\nexport type KokoroVoice = (typeof KOKORO_VOICES)[number];\n\n// Deepgram Aura speakers\nexport const DEEPGRAM_SPEAKERS = [\n  \"angus\", \"asteria\", \"arcas\", \"orion\", \"orpheus\", \"athena\",\n  \"luna\", \"zeus\", \"perseus\", \"helios\", \"hera\", \"stella\",\n] as const;\n\nexport type DeepgramSpeaker = (typeof DEEPGRAM_SPEAKERS)[number];\n","/**\n * @fileoverview Kokoro TTS provider implementation using Hugging Face transformers\n * Runs on Node.js CPU via transformers library\n */\nimport type { TTSResult } from \"../types/types\";\nimport { KOKORO_VOICES, type KokoroVoice } from \"../types/types\";\n\nlet ttsInstance: any = null;\nlet modelLoading: Promise<any> | null = null;\n\n/**\n * Lazy-load Kokoro model using Hugging Face transformers (happens once per server instance)\n */\nasync function getKokoroTTS() {\n  if (ttsInstance) return ttsInstance;\n\n  if (modelLoading) {\n    await modelLoading;\n    return ttsInstance;\n  }\n\n  modelLoading = (async () => {\n    try {\n      // Dynamic import to load transformers library. The specifier is kept in\n      // a variable so the type-checker/bundler treats it as an optional runtime\n      // dependency (see optionalDependencies) rather than a build-time one.\n      const transformersModule = \"@huggingface/transformers\";\n      const transformers: any = await import(/* @vite-ignore */ transformersModule);\n      const { StyleTextToSpeech2Model, AutoTokenizer } = transformers;\n\n      const model_id = \"hexgrad/Kokoro-82M\";\n\n      // Load model and tokenizer in parallel\n      const [model, tokenizer] = await Promise.all([\n        StyleTextToSpeech2Model.from_pretrained(model_id, {\n          device: \"cpu\",\n          dtype: \"q8\"\n        }),\n        AutoTokenizer.from_pretrained(model_id)\n      ]);\n\n      ttsInstance = { model, tokenizer };\n\n      console.log(\"[Kokoro] Model loaded successfully\");\n    } catch (error) {\n      console.error(\"[Kokoro] Failed to load model:\", error);\n      throw error;\n    } finally {\n      modelLoading = null;\n    }\n  })();\n\n  await modelLoading;\n  return ttsInstance;\n}\n\n/**\n * Generate speech from text using Kokoro\n */\nexport async function generateKokoroSpeech(\n  text: string,\n  voice: string = \"af_heart\"\n): Promise<TTSResult> {\n  // Validate voice\n  const kokoroVoice = KOKORO_VOICES.includes(voice as KokoroVoice)\n    ? (voice as KokoroVoice)\n    : \"af_heart\";\n\n  const tts = await getKokoroTTS();\n\n  try {\n    // For Node.js server-side TTS, we'd need the complete transformers implementation\n    // This is a placeholder showing the expected interface\n    // Consider using the browser-based implementation via Web Workers for production\n\n    throw new Error(\"Server-side Kokoro TTS requires additional setup. Use the browser-based implementation via Web Workers.\");\n  } catch (error) {\n    console.error(\"[Kokoro] Error generating speech:\", error);\n    throw error;\n  }\n}\n","/**\n * @fileoverview Deepgram TTS provider implementation using Cloudflare Workers AI\n * Requires Cloudflare AI binding\n */\nimport type { TTSResult } from \"../types/types\";\nimport { DEEPGRAM_SPEAKERS, type DeepgramSpeaker } from \"../types/types\";\n\n/**\n * Resolve the Cloudflare Workers AI binding at runtime without a hard\n * dependency on the host application. Consumers running on Cloudflare can\n * expose the context by setting `globalThis.getCloudflareContext` (as the\n * `@opennextjs/cloudflare` / `@cloudflare/next-on-pages` helpers do) or by\n * placing the bound `env` on `globalThis.__env__`.\n */\nfunction resolveCloudflareAI(): any {\n  const g = globalThis as any;\n  try {\n    if (typeof g.getCloudflareContext === \"function\") {\n      const ctx = g.getCloudflareContext();\n      if (ctx?.env?.AI) return ctx.env.AI;\n    }\n  } catch {\n    // CF context helper threw (e.g. called outside a request scope)\n  }\n  return g.__env__?.AI ?? g.AI;\n}\n\n/**\n * Generate speech from text using Deepgram Aura via Cloudflare Workers AI\n */\nexport async function generateDeepgramSpeech(\n  text: string,\n  speaker: string = \"angus\"\n): Promise<TTSResult> {\n  // Validate speaker\n  const auraVoice = DEEPGRAM_SPEAKERS.includes(speaker as DeepgramSpeaker)\n    ? (speaker as DeepgramSpeaker)\n    : \"angus\";\n\n  const ai = resolveCloudflareAI();\n\n  if (!ai) {\n    throw new Error(\"Cloudflare AI binding not available\");\n  }\n\n  const result = await ai.run(\"@cf/deepgram/aura-1\", {\n    text: text.slice(0, 5000),\n    speaker: auraVoice,\n    encoding: \"mp3\",\n  });\n\n  return {\n    audio: result,\n    contentType: \"audio/mpeg\",\n  };\n}\n","/**\n * @fileoverview Unified text-to-speech API supporting Kokoro (default) and Deepgram\n *\n * Kokoro: Faster, more natural, runs on Node CPU\n * Deepgram: Requires Cloudflare AI binding, MP3 output\n */\nimport type { TTSOptions, TTSResult } from \"./types/types\";\nimport { generateKokoroSpeech } from \"./core/kokoro\";\nimport { generateDeepgramSpeech } from \"./core/deepgram\";\n\nexport * from \"./types/types\";\n\n/**\n * Generate speech from text using the specified provider\n *\n * @param options - TTS configuration\n * @returns Audio buffer and content type\n *\n * @example\n * ```ts\n * // Use Kokoro (default)\n * const audio = await generateSpeech({\n *   text: \"Hello world\",\n *   voice: \"af_heart\"\n * });\n *\n * // Use Deepgram\n * const audio = await generateSpeech({\n *   text: \"Hello world\",\n *   provider: \"deepgram\",\n *   voice: \"angus\"\n * });\n * ```\n */\nexport async function generateSpeech(\n  options: TTSOptions\n): Promise<TTSResult> {\n  const { text, provider = \"kokoro\", voice = \"af_heart\" } = options;\n\n  if (!text || typeof text !== \"string\" || text.trim().length === 0) {\n    throw new Error(\"Text is required\");\n  }\n\n  switch (provider) {\n    case \"kokoro\":\n      return generateKokoroSpeech(text, voice);\n\n    case \"deepgram\":\n      return generateDeepgramSpeech(text, voice);\n\n    default:\n      throw new Error(`Unknown TTS provider: ${provider}`);\n  }\n}\n"],"names":["KOKORO_VOICES","DEEPGRAM_SPEAKERS","ttsInstance","modelLoading","getKokoroTTS","transformers","StyleTextToSpeech2Model","AutoTokenizer","model_id","model","tokenizer","error","generateKokoroSpeech","text","voice","resolveCloudflareAI","_a","_b","g","ctx","generateDeepgramSpeech","speaker","auraVoice","ai","generateSpeech","options","provider"],"mappings":"AAkBO,MAAMA,IAAgB;AAAA,EAC3B;AAAA,EAAY;AAAA,EAAY;AAAA,EAAY;AAAA,EACpC;AAAA,EAAc;AAAA,EAAa;AAAA,EAAY;AAAA,EAAY;AAAA,EACnD;AAAA,EAAW;AAAA,EAAW;AAAA,EAAY;AAAA,EAClC;AAAA,EAAW;AAAA,EAAc;AAC3B,GAKaC,IAAoB;AAAA,EAC/B;AAAA,EAAS;AAAA,EAAW;AAAA,EAAS;AAAA,EAAS;AAAA,EAAW;AAAA,EACjD;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAW;AAAA,EAAU;AAAA,EAAQ;AAC/C;ACxBA,IAAIC,IAAmB,MACnBC,IAAoC;AAKxC,eAAeC,IAAe;AAC5B,SAAIF,MAEAC,KACF,MAAMA,GACCD,MAGTC,KAAgB,YAAY;AAC1B,QAAI;AAKF,YAAME,IAAoB,MAAM,OADL,8BAErB,EAAE,yBAAAC,GAAyB,eAAAC,EAAA,IAAkBF,GAE7CG,IAAW,sBAGX,CAACC,GAAOC,CAAS,IAAI,MAAM,QAAQ,IAAI;AAAA,QAC3CJ,EAAwB,gBAAgBE,GAAU;AAAA,UAChD,QAAQ;AAAA,UACR,OAAO;AAAA,QAAA,CACR;AAAA,QACDD,EAAc,gBAAgBC,CAAQ;AAAA,MAAA,CACvC;AAED,MAAAN,IAAc,EAAE,OAAAO,GAAO,WAAAC,EAAA,GAEvB,QAAQ,IAAI,oCAAoC;AAAA,IAClD,SAASC,GAAO;AACd,oBAAQ,MAAM,kCAAkCA,CAAK,GAC/CA;AAAA,IACR,UAAA;AACE,MAAAR,IAAe;AAAA,IACjB;AAAA,EACF,GAAA,GAEA,MAAMA,GACCD;AACT;AAKA,eAAsBU,EACpBC,GACAC,IAAgB,YACI;AAEA,EAAAd,EAAc,SAASc,CAAoB,GAInD,MAAMV,EAAA;AAElB,MAAI;AAKF,UAAM,IAAI,MAAM,yGAAyG;AAAA,EAC3H,SAASO,GAAO;AACd,kBAAQ,MAAM,qCAAqCA,CAAK,GAClDA;AAAA,EACR;AACF;AClEA,SAASI,IAA2B;AFI7B,MAAAC,GAAAC;AEHL,QAAMC,IAAI;AACV,MAAI;AACF,QAAI,OAAOA,EAAE,wBAAyB,YAAY;AAChD,YAAMC,IAAMD,EAAE,qBAAA;AACd,WAAIF,IAAAG,KAAA,gBAAAA,EAAK,QAAL,QAAAH,EAAU,GAAI,QAAOG,EAAI,IAAI;AAAA,IACnC;AAAA,EACF,QAAQ;AAAA,EAER;AACA,WAAOF,IAAAC,EAAE,YAAF,gBAAAD,EAAW,OAAMC,EAAE;AAC5B;AAKA,eAAsBE,EACpBP,GACAQ,IAAkB,SACE;AAEpB,QAAMC,IAAYrB,EAAkB,SAASoB,CAA0B,IAClEA,IACD,SAEEE,IAAKR,EAAA;AAEX,MAAI,CAACQ;AACH,UAAM,IAAI,MAAM,qCAAqC;AASvD,SAAO;AAAA,IACL,OAPa,MAAMA,EAAG,IAAI,uBAAuB;AAAA,MACjD,MAAMV,EAAK,MAAM,GAAG,GAAI;AAAA,MACxB,SAASS;AAAA,MACT,UAAU;AAAA,IAAA,CACX;AAAA,IAIC,aAAa;AAAA,EAAA;AAEjB;ACrBA,eAAsBE,EACpBC,GACoB;AACpB,QAAM,EAAE,MAAAZ,GAAM,UAAAa,IAAW,UAAU,OAAAZ,IAAQ,eAAeW;AAE1D,MAAI,CAACZ,KAAQ,OAAOA,KAAS,YAAYA,EAAK,KAAA,EAAO,WAAW;AAC9D,UAAM,IAAI,MAAM,kBAAkB;AAGpC,UAAQa,GAAA;AAAA,IACN,KAAK;AACH,aAAOd,EAAqBC,GAAMC,CAAK;AAAA,IAEzC,KAAK;AACH,aAAOM,EAAuBP,GAAMC,CAAK;AAAA,IAE3C;AACE,YAAM,IAAI,MAAM,yBAAyBY,CAAQ,EAAE;AAAA,EAAA;AAEzD;"}
+{"version":3,"file":"index.js","sources":["../speech/core/kokoro.ts","../speech/core/deepgram.ts","../speech/index.ts"],"sourcesContent":["/**\n * @fileoverview Kokoro TTS provider for `generateSpeech`.\n *\n * The work happens in `kokoro-node.ts`, which runs the Kokoro model locally\n * through `kokoro-js`; this file is the thin provider adapter that validates the\n * voice and returns the package's `TTSResult` shape.\n */\nimport type { TTSResult } from \"../types/types\";\nimport { KOKORO_VOICES, type KokoroVoice } from \"../types/types\";\nimport { synthesizeWav, type KokoroNodeOptions } from \"./kokoro-node\";\n\n/**\n * Generate speech from text using Kokoro.\n *\n * Long text is chunked and joined automatically. An unrecognised voice falls\n * back to `af_heart` rather than failing the request.\n */\nexport async function generateKokoroSpeech(\n  text: string,\n  voice: string = \"af_heart\",\n  options: Omit<KokoroNodeOptions, \"voice\"> = {}\n): Promise<TTSResult> {\n  const kokoroVoice = KOKORO_VOICES.includes(voice as KokoroVoice)\n    ? (voice as KokoroVoice)\n    : \"af_heart\";\n\n  return synthesizeWav(text, { ...options, voice: kokoroVoice });\n}\n","/**\n * @fileoverview Deepgram TTS provider implementation using Cloudflare Workers AI\n * Requires Cloudflare AI binding\n */\nimport type { TTSResult } from \"../types/types\";\nimport { DEEPGRAM_SPEAKERS, type DeepgramSpeaker } from \"../types/types\";\n\n/**\n * Resolve the Cloudflare Workers AI binding at runtime without a hard\n * dependency on the host application. Consumers running on Cloudflare can\n * expose the context by setting `globalThis.getCloudflareContext` (as the\n * `@opennextjs/cloudflare` / `@cloudflare/next-on-pages` helpers do) or by\n * placing the bound `env` on `globalThis.__env__`.\n */\nfunction resolveCloudflareAI(): any {\n  const g = globalThis as any;\n  try {\n    if (typeof g.getCloudflareContext === \"function\") {\n      const ctx = g.getCloudflareContext();\n      if (ctx?.env?.AI) return ctx.env.AI;\n    }\n  } catch {\n    // CF context helper threw (e.g. called outside a request scope)\n  }\n  return g.__env__?.AI ?? g.AI;\n}\n\n/**\n * Generate speech from text using Deepgram Aura via Cloudflare Workers AI\n */\nexport async function generateDeepgramSpeech(\n  text: string,\n  speaker: string = \"angus\"\n): Promise<TTSResult> {\n  // Validate speaker\n  const auraVoice = DEEPGRAM_SPEAKERS.includes(speaker as DeepgramSpeaker)\n    ? (speaker as DeepgramSpeaker)\n    : \"angus\";\n\n  const ai = resolveCloudflareAI();\n\n  if (!ai) {\n    throw new Error(\"Cloudflare AI binding not available\");\n  }\n\n  const result = await ai.run(\"@cf/deepgram/aura-1\", {\n    text: text.slice(0, 5000),\n    speaker: auraVoice,\n    encoding: \"mp3\",\n  });\n\n  return {\n    audio: result,\n    contentType: \"audio/mpeg\",\n  };\n}\n","/**\n * @fileoverview Unified text-to-speech API supporting Kokoro (default) and Deepgram\n *\n * Kokoro: Faster, more natural, runs on Node CPU\n * Deepgram: Requires Cloudflare AI binding, MP3 output\n */\nimport type { TTSOptions, TTSResult } from \"./types/types\";\nimport { generateKokoroSpeech } from \"./core/kokoro\";\nimport { generateDeepgramSpeech } from \"./core/deepgram\";\n\nexport * from \"./types/types\";\n\n// Markdown handling is exported from the root entry so a server route can turn a\n// document into speakable text with the same rules the CLI and the browser use.\nexport {\n  looksLikeMarkdown,\n  markdownToSpeech,\n  markdownToSpeechSegments,\n  stripInlineMarkdown,\n  type MarkdownToSpeechOptions,\n  type SpeechSegment,\n  type SpeechSegmentType,\n} from \"./utils/markdown-to-speech\";\n\nexport { encodeWav, concatSamples, wavDurationSeconds } from \"./utils/wav\";\n\n/**\n * Generate speech from text using the specified provider\n *\n * @param options - TTS configuration\n * @returns Audio buffer and content type\n *\n * @example\n * ```ts\n * // Use Kokoro (default)\n * const audio = await generateSpeech({\n *   text: \"Hello world\",\n *   voice: \"af_heart\"\n * });\n *\n * // Use Deepgram\n * const audio = await generateSpeech({\n *   text: \"Hello world\",\n *   provider: \"deepgram\",\n *   voice: \"angus\"\n * });\n * ```\n */\nexport async function generateSpeech(\n  options: TTSOptions\n): Promise<TTSResult> {\n  const { text, provider = \"kokoro\", voice = \"af_heart\" } = options;\n\n  if (!text || typeof text !== \"string\" || text.trim().length === 0) {\n    throw new Error(\"Text is required\");\n  }\n\n  switch (provider) {\n    case \"kokoro\":\n      return generateKokoroSpeech(text, voice);\n\n    case \"deepgram\":\n      return generateDeepgramSpeech(text, voice);\n\n    default:\n      throw new Error(`Unknown TTS provider: ${provider}`);\n  }\n}\n"],"names":["generateKokoroSpeech","text","voice","options","kokoroVoice","KOKORO_VOICES","synthesizeWav","resolveCloudflareAI","g","ctx","_a","_b","generateDeepgramSpeech","speaker","auraVoice","DEEPGRAM_SPEAKERS","ai","generateSpeech","provider"],"mappings":";;;AAiBA,eAAsBA,EACpBC,GACAC,IAAgB,YAChBC,IAA4C,CAAA,GACxB;AACpB,QAAMC,IAAcC,EAAc,SAASH,CAAoB,IAC1DA,IACD;AAEJ,SAAOI,EAAcL,GAAM,EAAE,GAAGE,GAAS,OAAOC,GAAa;AAC/D;ACbA,SAASG,IAA2B;;AAClC,QAAMC,IAAI;AACV,MAAI;AACF,QAAI,OAAOA,EAAE,wBAAyB,YAAY;AAChD,YAAMC,IAAMD,EAAE,qBAAA;AACd,WAAIE,IAAAD,KAAA,gBAAAA,EAAK,QAAL,QAAAC,EAAU,GAAI,QAAOD,EAAI,IAAI;AAAA,IACnC;AAAA,EACF,QAAQ;AAAA,EAER;AACA,WAAOE,IAAAH,EAAE,YAAF,gBAAAG,EAAW,OAAMH,EAAE;AAC5B;AAKA,eAAsBI,EACpBX,GACAY,IAAkB,SACE;AAEpB,QAAMC,IAAYC,EAAkB,SAASF,CAA0B,IAClEA,IACD,SAEEG,IAAKT,EAAA;AAEX,MAAI,CAACS;AACH,UAAM,IAAI,MAAM,qCAAqC;AASvD,SAAO;AAAA,IACL,OAPa,MAAMA,EAAG,IAAI,uBAAuB;AAAA,MACjD,MAAMf,EAAK,MAAM,GAAG,GAAI;AAAA,MACxB,SAASa;AAAA,MACT,UAAU;AAAA,IAAA,CACX;AAAA,IAIC,aAAa;AAAA,EAAA;AAEjB;ACPA,eAAsBG,EACpBd,GACoB;AACpB,QAAM,EAAE,MAAAF,GAAM,UAAAiB,IAAW,UAAU,OAAAhB,IAAQ,eAAeC;AAE1D,MAAI,CAACF,KAAQ,OAAOA,KAAS,YAAYA,EAAK,KAAA,EAAO,WAAW;AAC9D,UAAM,IAAI,MAAM,kBAAkB;AAGpC,UAAQiB,GAAA;AAAA,IACN,KAAK;AACH,aAAOlB,EAAqBC,GAAMC,CAAK;AAAA,IAEzC,KAAK;AACH,aAAOU,EAAuBX,GAAMC,CAAK;AAAA,IAE3C;AACE,YAAM,IAAI,MAAM,yBAAyBgB,CAAQ,EAAE;AAAA,EAAA;AAEzD;"}

package/dist/kokoro-node-DJ_Rxp_N.js

@@ -0,0 +1,144 @@
+import { s as d } from "./semantic-split-CXhk-k1F.js";
+const v = [
+  "af_heart",
+  "af_alloy",
+  "af_aoede",
+  "af_bella",
+  "af_jessica",
+  "af_kore",
+  "af_nicole",
+  "af_nova",
+  "af_river",
+  "af_sarah",
+  "af_sky",
+  "am_adam",
+  "am_echo",
+  "am_eric",
+  "am_fenrir",
+  "am_liam",
+  "am_michael",
+  "am_onyx",
+  "am_puck",
+  "am_santa",
+  "bf_alice",
+  "bf_emma",
+  "bf_isabella",
+  "bf_lily",
+  "bm_daniel",
+  "bm_fable",
+  "bm_george",
+  "bm_lewis"
+];
+function x(e) {
+  const [t, n] = e, o = e.slice(3) || e;
+  return {
+    id: e,
+    name: o.charAt(0).toUpperCase() + o.slice(1),
+    accent: t === "b" ? "British English" : "American English",
+    gender: n === "m" ? "Male" : "Female"
+  };
+}
+const S = [
+  "angus",
+  "asteria",
+  "arcas",
+  "orion",
+  "orpheus",
+  "athena",
+  "luna",
+  "zeus",
+  "perseus",
+  "helios",
+  "hera",
+  "stella"
+], _ = 44;
+function g(e, t = 0) {
+  const n = Math.max(e.length - 1, 0) * Math.max(t, 0), o = e.reduce((i, r) => i + r.length, 0) + n, a = new Float32Array(o);
+  let c = 0;
+  return e.forEach((i, r) => {
+    a.set(i, c), c += i.length, r < e.length - 1 && (c += Math.max(t, 0));
+  }), a;
+}
+function E(e, t = 24e3, n = 1) {
+  const o = new ArrayBuffer(_ + e.length * 2), a = new DataView(o), c = (r, s) => {
+    for (let f = 0; f < s.length; f += 1)
+      a.setUint8(r + f, s.charCodeAt(f));
+  }, i = t * n * 2;
+  c(0, "RIFF"), a.setUint32(4, 36 + e.length * 2, !0), c(8, "WAVE"), c(12, "fmt "), a.setUint32(16, 16, !0), a.setUint16(20, 1, !0), a.setUint16(22, n, !0), a.setUint32(24, t, !0), a.setUint32(28, i, !0), a.setUint16(32, n * 2, !0), a.setUint16(34, 16, !0), c(36, "data"), a.setUint32(40, e.length * 2, !0);
+  for (let r = 0; r < e.length; r += 1) {
+    const s = Math.max(-1, Math.min(1, e[r]));
+    a.setInt16(_ + r * 2, s < 0 ? s * 32768 : s * 32767, !0);
+  }
+  return o;
+}
+function C(e, t) {
+  return t > 0 ? e.length / t : 0;
+}
+const w = "onnx-community/Kokoro-82M-v1.0-ONNX", p = "af_heart", y = "q8", U = "cpu", b = 400, k = 120, u = /* @__PURE__ */ new Map();
+function A() {
+  var t, n;
+  const e = globalThis;
+  return !!((n = (t = e.process) == null ? void 0 : t.versions) != null && n.node || e.Bun || e.Deno);
+}
+async function M(e = {}) {
+  const t = e.model ?? w, n = e.dtype ?? y, o = e.device ?? (A() ? U : "wasm"), a = `${t}|${n}|${o}`, c = u.get(a);
+  if (c) return c;
+  const i = (async () => {
+    let r;
+    try {
+      ({ KokoroTTS: r } = await import("kokoro-js"));
+    } catch (s) {
+      throw new Error(
+        `Local Kokoro speech needs the optional \`kokoro-js\` package. Install it with \`npm install kokoro-js\`. (import failed: ${s instanceof Error ? s.message : String(s)})`
+      );
+    }
+    return r.from_pretrained(t, {
+      dtype: n,
+      device: o,
+      progress_callback: e.onModelProgress
+    });
+  })();
+  u.set(a, i);
+  try {
+    return await i;
+  } catch (r) {
+    throw u.delete(a), r;
+  }
+}
+function K() {
+  u.clear();
+}
+async function T(e, t = {}) {
+  var m;
+  const n = (e ?? "").trim();
+  if (!n) throw new Error("Text is required");
+  const o = d(n, t.maxChunkLength ?? b).map((l) => l.trim()).filter((l) => l.length > 0), a = await M(t), c = t.voice ?? p, i = t.speed ?? 1, r = [];
+  let s = 24e3;
+  for (let l = 0; l < o.length; l += 1) {
+    (m = t.onChunk) == null || m.call(t, { index: l, total: o.length, text: o[l] });
+    const h = await a.generate(o[l], { voice: c, speed: i });
+    r.push(h.audio), s = h.sampling_rate ?? s;
+  }
+  const f = t.gapMs ?? k;
+  return {
+    samples: g(r, Math.round(f / 1e3 * s)),
+    sampleRate: s
+  };
+}
+async function F(e, t = {}) {
+  const { samples: n, sampleRate: o } = await T(e, t);
+  return { audio: E(n, o), contentType: "audio/wav" };
+}
+export {
+  S as D,
+  v as K,
+  T as a,
+  g as c,
+  x as d,
+  E as e,
+  M as l,
+  K as r,
+  F as s,
+  C as w
+};
+//# sourceMappingURL=kokoro-node-DJ_Rxp_N.js.map

package/dist/kokoro-node-DJ_Rxp_N.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"kokoro-node-DJ_Rxp_N.js","sources":["../speech/types/types.ts","../speech/utils/wav.ts","../speech/core/kokoro-node.ts"],"sourcesContent":["/**\n * @fileoverview Type definitions for text-to-speech providers\n */\n\nexport type TTSProvider = \"kokoro\" | \"deepgram\";\n\nexport interface TTSOptions {\n  text: string;\n  provider?: TTSProvider;\n  voice?: string;\n}\n\nexport interface TTSResult {\n  audio: ArrayBuffer;\n  contentType: string;\n}\n\n// Kokoro voices from the model. The `a`/`b` prefix is the accent (American /\n// British English) and the `f`/`m` that follows it is the voice's gender, which\n// is how `describeKokoroVoice` derives a listing without a second catalog.\nexport const KOKORO_VOICES = [\n  \"af_heart\", \"af_alloy\", \"af_aoede\", \"af_bella\", \"af_jessica\",\n  \"af_kore\", \"af_nicole\", \"af_nova\", \"af_river\", \"af_sarah\", \"af_sky\",\n  \"am_adam\", \"am_echo\", \"am_eric\", \"am_fenrir\", \"am_liam\",\n  \"am_michael\", \"am_onyx\", \"am_puck\", \"am_santa\",\n  \"bf_alice\", \"bf_emma\", \"bf_isabella\", \"bf_lily\",\n  \"bm_daniel\", \"bm_fable\", \"bm_george\", \"bm_lewis\",\n] as const;\n\nexport type KokoroVoice = (typeof KOKORO_VOICES)[number];\n\nexport interface KokoroVoiceDescription {\n  id: string;\n  /** Display name, e.g. `Heart` for `af_heart`. */\n  name: string;\n  /** `American English` or `British English`. */\n  accent: string;\n  gender: \"Female\" | \"Male\";\n}\n\n/**\n * Describes a voice from its id, so listings (`--list-voices`, a voice picker)\n * do not need a hand-maintained table that can drift from `KOKORO_VOICES`.\n */\nexport function describeKokoroVoice(id: string): KokoroVoiceDescription {\n  const [accentCode, genderCode] = id;\n  const suffix = id.slice(3) || id;\n\n  return {\n    id,\n    name: suffix.charAt(0).toUpperCase() + suffix.slice(1),\n    accent: accentCode === \"b\" ? \"British English\" : \"American English\",\n    gender: genderCode === \"m\" ? \"Male\" : \"Female\",\n  };\n}\n\n// Deepgram Aura speakers\nexport const DEEPGRAM_SPEAKERS = [\n  \"angus\", \"asteria\", \"arcas\", \"orion\", \"orpheus\", \"athena\",\n  \"luna\", \"zeus\", \"perseus\", \"helios\", \"hera\", \"stella\",\n] as const;\n\nexport type DeepgramSpeaker = (typeof DEEPGRAM_SPEAKERS)[number];\n","/**\n * @fileoverview Minimal WAV (RIFF) encoder for raw Float32 PCM.\n *\n * The browser path gets its WAV from `convertAudioBufferToWav`, which needs a Web\n * Audio `AudioBuffer`. Node has no such thing: the Kokoro model hands back a bare\n * `Float32Array`, and long documents are synthesized one chunk at a time and\n * joined before encoding. Both of those are pure buffer work, so they live here\n * with no platform dependency.\n */\n\n/** Number of bytes in the fixed RIFF/fmt/data header this encoder writes. */\nconst HEADER_BYTES = 44;\n\n/**\n * Joins per-chunk sample buffers into one contiguous track.\n *\n * @param chunks Sample buffers in playback order.\n * @param silenceSamples Samples of silence to insert between chunks, so the\n *   seams between separately synthesized chunks do not sound clipped.\n */\nexport function concatSamples(\n  chunks: Float32Array[],\n  silenceSamples = 0\n): Float32Array {\n  const gaps = Math.max(chunks.length - 1, 0) * Math.max(silenceSamples, 0);\n  const total = chunks.reduce((sum, chunk) => sum + chunk.length, 0) + gaps;\n  const out = new Float32Array(total);\n\n  let offset = 0;\n  chunks.forEach((chunk, i) => {\n    out.set(chunk, offset);\n    offset += chunk.length;\n    if (i < chunks.length - 1) offset += Math.max(silenceSamples, 0);\n  });\n\n  return out;\n}\n\n/**\n * Encodes Float32 samples (nominally -1..1) as a 16-bit PCM WAV file.\n *\n * @param samples Interleaved samples when `channels` is greater than 1.\n * @param sampleRate Sample rate of the audio, e.g. 24000 for Kokoro.\n * @param channels Channel count. Kokoro is mono.\n */\nexport function encodeWav(\n  samples: Float32Array,\n  sampleRate = 24000,\n  channels = 1\n): ArrayBuffer {\n  const buffer = new ArrayBuffer(HEADER_BYTES + samples.length * 2);\n  const view = new DataView(buffer);\n\n  const writeString = (offset: number, value: string): void => {\n    for (let i = 0; i < value.length; i += 1) {\n      view.setUint8(offset + i, value.charCodeAt(i));\n    }\n  };\n\n  const byteRate = sampleRate * channels * 2;\n\n  writeString(0, \"RIFF\");\n  view.setUint32(4, 36 + samples.length * 2, true);\n  writeString(8, \"WAVE\");\n  writeString(12, \"fmt \");\n  view.setUint32(16, 16, true); // PCM header size\n  view.setUint16(20, 1, true); // format: PCM\n  view.setUint16(22, channels, true);\n  view.setUint32(24, sampleRate, true);\n  view.setUint32(28, byteRate, true);\n  view.setUint16(32, channels * 2, true); // block align\n  view.setUint16(34, 16, true); // bits per sample\n  writeString(36, \"data\");\n  view.setUint32(40, samples.length * 2, true);\n\n  for (let i = 0; i < samples.length; i += 1) {\n    // Clamp before scaling: the model occasionally overshoots 1.0 and wrapping\n    // a 16-bit sample turns that into a loud click.\n    const sample = Math.max(-1, Math.min(1, samples[i]));\n    view.setInt16(HEADER_BYTES + i * 2, sample < 0 ? sample * 0x8000 : sample * 0x7fff, true);\n  }\n\n  return buffer;\n}\n\n/** Duration of a Float32 track in seconds, for progress and log lines. */\nexport function wavDurationSeconds(samples: Float32Array, sampleRate: number): number {\n  return sampleRate > 0 ? samples.length / sampleRate : 0;\n}\n","/**\n * @fileoverview Kokoro TTS running locally on Node (or Bun/Deno) via `kokoro-js`.\n *\n * This is the engine behind `generateSpeech({ provider: \"kokoro\" })` on a server\n * and behind the `use-voice-control` CLI. The model runs on the CPU through\n * onnxruntime, so nothing is sent to a third-party service; the first call\n * downloads the weights (~90 MB at the default `q8`) into the Hugging Face cache\n * and every later call reuses the instance held in this module.\n *\n * Kokoro's context is a few hundred phonemes, so anything longer than a\n * paragraph has to be synthesized in pieces. `synthesizeSamples` chunks on\n * sentence and paragraph boundaries with `splitTextSmart`, runs the chunks in\n * order, and joins the audio with a short silence at each seam.\n */\nimport type { TTSResult } from \"../types/types\";\nimport { splitTextSmart } from \"../utils/semantic-split.js\";\nimport { concatSamples, encodeWav } from \"../utils/wav\";\n\nexport type KokoroDtype = \"fp32\" | \"fp16\" | \"q8\" | \"q4\" | \"q4f16\";\nexport type KokoroDevice = \"wasm\" | \"webgpu\" | \"cpu\";\n\nexport interface KokoroNodeOptions {\n  /** Voice id, e.g. `af_heart`. Default `af_heart`. */\n  voice?: string;\n  /** Speaking rate; 1 is the model's natural pace. Default 1. */\n  speed?: number;\n  /** Hugging Face model id. Default `onnx-community/Kokoro-82M-v1.0-ONNX`. */\n  model?: string;\n  /** Weight quantization. `q8` (default) is the best size/quality trade on CPU. */\n  dtype?: KokoroDtype;\n  /** Execution device. Default `cpu`. */\n  device?: KokoroDevice;\n  /**\n   * Target chunk size in characters. Kokoro's context is limited, so long text\n   * is split before synthesis; 400 leaves headroom for phoneme expansion.\n   */\n  maxChunkLength?: number;\n  /** Silence inserted between chunks, in milliseconds. Default 120. */\n  gapMs?: number;\n  /** Model download/loading progress, forwarded from transformers.js. */\n  onModelProgress?: (progress: unknown) => void;\n  /** Called before each chunk is synthesized, for CLI progress output. */\n  onChunk?: (info: { index: number; total: number; text: string }) => void;\n}\n\nexport interface KokoroAudio {\n  samples: Float32Array;\n  sampleRate: number;\n}\n\nconst DEFAULT_MODEL = \"onnx-community/Kokoro-82M-v1.0-ONNX\";\nconst DEFAULT_VOICE = \"af_heart\";\nconst DEFAULT_DTYPE: KokoroDtype = \"q8\";\nconst DEFAULT_DEVICE: KokoroDevice = \"cpu\";\nconst DEFAULT_MAX_CHUNK = 400;\nconst DEFAULT_GAP_MS = 120;\n\n/** One loaded model per model/dtype/device combination, keyed by those three. */\nconst instances = new Map<string, Promise<any>>();\n\n/** True on Node, Bun and Deno — anywhere `kokoro-js` can load onnxruntime-node. */\nfunction isServerRuntime(): boolean {\n  const g = globalThis as any;\n  return Boolean(g.process?.versions?.node || g.Bun || g.Deno);\n}\n\n/**\n * Loads (and caches) a `kokoro-js` instance.\n *\n * `kokoro-js` is an optional dependency: a browser-only consumer should not have\n * to download onnxruntime-node, so a missing install is reported as an\n * actionable error rather than a module-resolution stack trace.\n */\nexport async function loadKokoroTTS(options: KokoroNodeOptions = {}): Promise<any> {\n  const model = options.model ?? DEFAULT_MODEL;\n  const dtype = options.dtype ?? DEFAULT_DTYPE;\n  const device = options.device ?? (isServerRuntime() ? DEFAULT_DEVICE : \"wasm\");\n  const key = `${model}|${dtype}|${device}`;\n\n  const cached = instances.get(key);\n  if (cached) return cached;\n\n  const loading = (async () => {\n    let KokoroTTS: any;\n    try {\n      // Kept in a variable so bundlers treat this as an optional runtime\n      // dependency rather than something to resolve at build time.\n      const specifier = \"kokoro-js\";\n      ({ KokoroTTS } = await import(/* @vite-ignore */ specifier));\n    } catch (error) {\n      throw new Error(\n        \"Local Kokoro speech needs the optional `kokoro-js` package. \" +\n          \"Install it with `npm install kokoro-js`. \" +\n          `(import failed: ${error instanceof Error ? error.message : String(error)})`\n      );\n    }\n\n    return KokoroTTS.from_pretrained(model, {\n      dtype,\n      device,\n      progress_callback: options.onModelProgress,\n    });\n  })();\n\n  instances.set(key, loading);\n\n  try {\n    return await loading;\n  } catch (error) {\n    // A failed download must not poison every later attempt.\n    instances.delete(key);\n    throw error;\n  }\n}\n\n/** Drops cached model instances. Mainly useful in tests and long-lived workers. */\nexport function resetKokoroCache(): void {\n  instances.clear();\n}\n\n/**\n * Synthesizes `text` to raw samples, chunking anything longer than the model's\n * comfortable context.\n */\nexport async function synthesizeSamples(\n  text: string,\n  options: KokoroNodeOptions = {}\n): Promise<KokoroAudio> {\n  const trimmed = (text ?? \"\").trim();\n  if (!trimmed) throw new Error(\"Text is required\");\n\n  const chunks = splitTextSmart(trimmed, options.maxChunkLength ?? DEFAULT_MAX_CHUNK)\n    .map((chunk: string) => chunk.trim())\n    .filter((chunk: string) => chunk.length > 0);\n\n  const tts = await loadKokoroTTS(options);\n  const voice = options.voice ?? DEFAULT_VOICE;\n  const speed = options.speed ?? 1;\n\n  const rendered: Float32Array[] = [];\n  let sampleRate = 24000;\n\n  for (let index = 0; index < chunks.length; index += 1) {\n    options.onChunk?.({ index, total: chunks.length, text: chunks[index] });\n    const audio = await tts.generate(chunks[index], { voice, speed });\n    rendered.push(audio.audio);\n    sampleRate = audio.sampling_rate ?? sampleRate;\n  }\n\n  const gapMs = options.gapMs ?? DEFAULT_GAP_MS;\n  return {\n    samples: concatSamples(rendered, Math.round((gapMs / 1000) * sampleRate)),\n    sampleRate,\n  };\n}\n\n/** Synthesizes `text` and encodes it as a 16-bit PCM WAV file. */\nexport async function synthesizeWav(\n  text: string,\n  options: KokoroNodeOptions = {}\n): Promise<TTSResult> {\n  const { samples, sampleRate } = await synthesizeSamples(text, options);\n  return { audio: encodeWav(samples, sampleRate), contentType: \"audio/wav\" };\n}\n"],"names":["KOKORO_VOICES","describeKokoroVoice","id","accentCode","genderCode","suffix","DEEPGRAM_SPEAKERS","HEADER_BYTES","concatSamples","chunks","silenceSamples","gaps","total","sum","chunk","out","offset","i","encodeWav","samples","sampleRate","channels","buffer","view","writeString","value","byteRate","sample","wavDurationSeconds","DEFAULT_MODEL","DEFAULT_VOICE","DEFAULT_DTYPE","DEFAULT_DEVICE","DEFAULT_MAX_CHUNK","DEFAULT_GAP_MS","instances","isServerRuntime","g","_b","_a","loadKokoroTTS","options","model","dtype","device","key","cached","loading","KokoroTTS","error","resetKokoroCache","synthesizeSamples","text","trimmed","splitTextSmart","tts","voice","speed","rendered","index","audio","gapMs","synthesizeWav"],"mappings":";AAoBO,MAAMA,IAAgB;AAAA,EAC3B;AAAA,EAAY;AAAA,EAAY;AAAA,EAAY;AAAA,EAAY;AAAA,EAChD;AAAA,EAAW;AAAA,EAAa;AAAA,EAAW;AAAA,EAAY;AAAA,EAAY;AAAA,EAC3D;AAAA,EAAW;AAAA,EAAW;AAAA,EAAW;AAAA,EAAa;AAAA,EAC9C;AAAA,EAAc;AAAA,EAAW;AAAA,EAAW;AAAA,EACpC;AAAA,EAAY;AAAA,EAAW;AAAA,EAAe;AAAA,EACtC;AAAA,EAAa;AAAA,EAAY;AAAA,EAAa;AACxC;AAiBO,SAASC,EAAoBC,GAAoC;AACtE,QAAM,CAACC,GAAYC,CAAU,IAAIF,GAC3BG,IAASH,EAAG,MAAM,CAAC,KAAKA;AAE9B,SAAO;AAAA,IACL,IAAAA;AAAA,IACA,MAAMG,EAAO,OAAO,CAAC,EAAE,gBAAgBA,EAAO,MAAM,CAAC;AAAA,IACrD,QAAQF,MAAe,MAAM,oBAAoB;AAAA,IACjD,QAAQC,MAAe,MAAM,SAAS;AAAA,EAAA;AAE1C;AAGO,MAAME,IAAoB;AAAA,EAC/B;AAAA,EAAS;AAAA,EAAW;AAAA,EAAS;AAAA,EAAS;AAAA,EAAW;AAAA,EACjD;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAW;AAAA,EAAU;AAAA,EAAQ;AAC/C,GCjDMC,IAAe;AASd,SAASC,EACdC,GACAC,IAAiB,GACH;AACd,QAAMC,IAAO,KAAK,IAAIF,EAAO,SAAS,GAAG,CAAC,IAAI,KAAK,IAAIC,GAAgB,CAAC,GAClEE,IAAQH,EAAO,OAAO,CAACI,GAAKC,MAAUD,IAAMC,EAAM,QAAQ,CAAC,IAAIH,GAC/DI,IAAM,IAAI,aAAaH,CAAK;AAElC,MAAII,IAAS;AACb,SAAAP,EAAO,QAAQ,CAACK,GAAOG,MAAM;AAC3B,IAAAF,EAAI,IAAID,GAAOE,CAAM,GACrBA,KAAUF,EAAM,QACZG,IAAIR,EAAO,SAAS,WAAa,KAAK,IAAIC,GAAgB,CAAC;AAAA,EACjE,CAAC,GAEMK;AACT;AASO,SAASG,EACdC,GACAC,IAAa,MACbC,IAAW,GACE;AACb,QAAMC,IAAS,IAAI,YAAYf,IAAeY,EAAQ,SAAS,CAAC,GAC1DI,IAAO,IAAI,SAASD,CAAM,GAE1BE,IAAc,CAACR,GAAgBS,MAAwB;AAC3D,aAASR,IAAI,GAAGA,IAAIQ,EAAM,QAAQR,KAAK;AACrC,MAAAM,EAAK,SAASP,IAASC,GAAGQ,EAAM,WAAWR,CAAC,CAAC;AAAA,EAEjD,GAEMS,IAAWN,IAAaC,IAAW;AAEzC,EAAAG,EAAY,GAAG,MAAM,GACrBD,EAAK,UAAU,GAAG,KAAKJ,EAAQ,SAAS,GAAG,EAAI,GAC/CK,EAAY,GAAG,MAAM,GACrBA,EAAY,IAAI,MAAM,GACtBD,EAAK,UAAU,IAAI,IAAI,EAAI,GAC3BA,EAAK,UAAU,IAAI,GAAG,EAAI,GAC1BA,EAAK,UAAU,IAAIF,GAAU,EAAI,GACjCE,EAAK,UAAU,IAAIH,GAAY,EAAI,GACnCG,EAAK,UAAU,IAAIG,GAAU,EAAI,GACjCH,EAAK,UAAU,IAAIF,IAAW,GAAG,EAAI,GACrCE,EAAK,UAAU,IAAI,IAAI,EAAI,GAC3BC,EAAY,IAAI,MAAM,GACtBD,EAAK,UAAU,IAAIJ,EAAQ,SAAS,GAAG,EAAI;AAE3C,WAASF,IAAI,GAAGA,IAAIE,EAAQ,QAAQF,KAAK,GAAG;AAG1C,UAAMU,IAAS,KAAK,IAAI,IAAI,KAAK,IAAI,GAAGR,EAAQF,CAAC,CAAC,CAAC;AACnD,IAAAM,EAAK,SAAShB,IAAeU,IAAI,GAAGU,IAAS,IAAIA,IAAS,QAASA,IAAS,OAAQ,EAAI;AAAA,EAC1F;AAEA,SAAOL;AACT;AAGO,SAASM,EAAmBT,GAAuBC,GAA4B;AACpF,SAAOA,IAAa,IAAID,EAAQ,SAASC,IAAa;AACxD;ACtCA,MAAMS,IAAgB,uCAChBC,IAAgB,YAChBC,IAA6B,MAC7BC,IAA+B,OAC/BC,IAAoB,KACpBC,IAAiB,KAGjBC,wBAAgB,IAAA;AAGtB,SAASC,IAA2B;;AAClC,QAAMC,IAAI;AACV,SAAO,IAAQC,KAAAC,IAAAF,EAAE,YAAF,gBAAAE,EAAW,aAAX,QAAAD,EAAqB,QAAQD,EAAE,OAAOA,EAAE;AACzD;AASA,eAAsBG,EAAcC,IAA6B,IAAkB;AACjF,QAAMC,IAAQD,EAAQ,SAASZ,GACzBc,IAAQF,EAAQ,SAASV,GACzBa,IAASH,EAAQ,WAAWL,EAAA,IAAoBJ,IAAiB,SACjEa,IAAM,GAAGH,CAAK,IAAIC,CAAK,IAAIC,CAAM,IAEjCE,IAASX,EAAU,IAAIU,CAAG;AAChC,MAAIC,EAAQ,QAAOA;AAEnB,QAAMC,KAAW,YAAY;AAC3B,QAAIC;AACJ,QAAI;AAIF,OAAC,EAAE,WAAAA,EAAA,IAAc,MAAM,OADL;AAAA,IAEpB,SAASC,GAAO;AACd,YAAM,IAAI;AAAA,QACR,4HAEqBA,aAAiB,QAAQA,EAAM,UAAU,OAAOA,CAAK,CAAC;AAAA,MAAA;AAAA,IAE/E;AAEA,WAAOD,EAAU,gBAAgBN,GAAO;AAAA,MACtC,OAAAC;AAAA,MACA,QAAAC;AAAA,MACA,mBAAmBH,EAAQ;AAAA,IAAA,CAC5B;AAAA,EACH,GAAA;AAEA,EAAAN,EAAU,IAAIU,GAAKE,CAAO;AAE1B,MAAI;AACF,WAAO,MAAMA;AAAA,EACf,SAASE,GAAO;AAEd,UAAAd,EAAU,OAAOU,CAAG,GACdI;AAAA,EACR;AACF;AAGO,SAASC,IAAyB;AACvC,EAAAf,EAAU,MAAA;AACZ;AAMA,eAAsBgB,EACpBC,GACAX,IAA6B,IACP;;AACtB,QAAMY,KAAWD,KAAQ,IAAI,KAAA;AAC7B,MAAI,CAACC,EAAS,OAAM,IAAI,MAAM,kBAAkB;AAEhD,QAAM5C,IAAS6C,EAAeD,GAASZ,EAAQ,kBAAkBR,CAAiB,EAC/E,IAAI,CAACnB,MAAkBA,EAAM,KAAA,CAAM,EACnC,OAAO,CAACA,MAAkBA,EAAM,SAAS,CAAC,GAEvCyC,IAAM,MAAMf,EAAcC,CAAO,GACjCe,IAAQf,EAAQ,SAASX,GACzB2B,IAAQhB,EAAQ,SAAS,GAEzBiB,IAA2B,CAAA;AACjC,MAAItC,IAAa;AAEjB,WAASuC,IAAQ,GAAGA,IAAQlD,EAAO,QAAQkD,KAAS,GAAG;AACrD,KAAApB,IAAAE,EAAQ,YAAR,QAAAF,EAAA,KAAAE,GAAkB,EAAE,OAAAkB,GAAO,OAAOlD,EAAO,QAAQ,MAAMA,EAAOkD,CAAK;AACnE,UAAMC,IAAQ,MAAML,EAAI,SAAS9C,EAAOkD,CAAK,GAAG,EAAE,OAAAH,GAAO,OAAAC,GAAO;AAChE,IAAAC,EAAS,KAAKE,EAAM,KAAK,GACzBxC,IAAawC,EAAM,iBAAiBxC;AAAA,EACtC;AAEA,QAAMyC,IAAQpB,EAAQ,SAASP;AAC/B,SAAO;AAAA,IACL,SAAS1B,EAAckD,GAAU,KAAK,MAAOG,IAAQ,MAAQzC,CAAU,CAAC;AAAA,IACxE,YAAAA;AAAA,EAAA;AAEJ;AAGA,eAAsB0C,EACpBV,GACAX,IAA6B,IACT;AACpB,QAAM,EAAE,SAAAtB,GAAS,YAAAC,EAAA,IAAe,MAAM+B,EAAkBC,GAAMX,CAAO;AACrE,SAAO,EAAE,OAAOvB,EAAUC,GAASC,CAAU,GAAG,aAAa,YAAA;AAC/D;"}