@tanstack/ai-elevenlabs 0.4.2 → 0.6.0

package/dist/esm/adapters/audio.d.ts

@@ -32,7 +32,7 @@ interface CommonAudioOptions {
     outputFormat?: ElevenLabsOutputFormat;
 }
 /**
- * Provider options for music generation (`music_v1`).
+ * Provider options for music generation (`music_v*`).
  */
 export interface ElevenLabsMusicProviderOptions extends CommonAudioOptions {
     /** Structured composition plan. Mutually exclusive with `prompt`/`duration`. */
@@ -65,7 +65,7 @@ export type ElevenLabsAudioProviderOptions = (ElevenLabsMusicProviderOptions & E
  *
  * @example
  * ```ts
- * const music = elevenlabsAudio('music_v1')
+ * const music = elevenlabsAudio('music_v2_5')
  * await generateAudio({ adapter: music, prompt: 'lo-fi beat', duration: 15 })
  *
  * const sfx = elevenlabsAudio('eleven_text_to_sound_v2')

package/dist/esm/adapters/audio.js

@@ -1,5 +1,5 @@
 import { arrayBufferToBase64, createElevenLabsClient, generateId, parseOutputFormat, readStreamToArrayBuffer } from "../utils/client.js";
-import { isElevenLabsMusicModel, isElevenLabsSoundEffectsModel } from "../model-meta.js";
+import { ELEVENLABS_AUDIO_MODELS, isElevenLabsMusicModel, isElevenLabsSoundEffectsModel } from "../model-meta.js";
 import { BaseAudioAdapter } from "@tanstack/ai/adapters";
 //#region src/adapters/audio.ts
 /**
@@ -9,7 +9,7 @@ import { BaseAudioAdapter } from "@tanstack/ai/adapters";
 *
 * @example
 * ```ts
-* const music = elevenlabsAudio('music_v1')
+* const music = elevenlabsAudio('music_v2_5')
 * await generateAudio({ adapter: music, prompt: 'lo-fi beat', duration: 15 })
 *
 * const sfx = elevenlabsAudio('eleven_text_to_sound_v2')
@@ -32,7 +32,7 @@ var ElevenLabsAudioAdapter = class extends BaseAudioAdapter {
 		try {
 			if (isElevenLabsMusicModel(this.model)) return await this.runMusic(options);
 			if (isElevenLabsSoundEffectsModel(this.model)) return await this.runSoundEffects(options);
-			throw new Error(`Unsupported ElevenLabs audio model "${this.model}". Expected one of: music_v1, eleven_text_to_sound_v2, eleven_text_to_sound_v1.`);
+			throw new Error(`Unsupported ElevenLabs audio model "${this.model}". Expected one of: ${ELEVENLABS_AUDIO_MODELS.join(", ")}.`);
 		} catch (error) {
 			logger.errors("elevenlabs.generateAudio fatal", {
 				error,

package/dist/esm/adapters/audio.js.map

@@ -1 +1 @@
-{"version":3,"file":"audio.js","names":[],"sources":["../../../src/adapters/audio.ts"],"sourcesContent":["import { BaseAudioAdapter } from '@tanstack/ai/adapters'\nimport {\n  arrayBufferToBase64,\n  createElevenLabsClient,\n  generateId,\n  parseOutputFormat,\n  readStreamToArrayBuffer,\n} from '../utils/client'\nimport {\n  isElevenLabsMusicModel,\n  isElevenLabsSoundEffectsModel,\n} from '../model-meta'\nimport type { ElevenLabsClient } from '@elevenlabs/elevenlabs-js'\nimport type {\n  AudioGenerationOptions,\n  AudioGenerationResult,\n} from '@tanstack/ai'\nimport type { ElevenLabsClientConfig } from '../utils/client'\nimport type {\n  ElevenLabsAudioModel,\n  ElevenLabsMusicModel,\n  ElevenLabsOutputFormat,\n  ElevenLabsSoundEffectsModel,\n} from '../model-meta'\n\n/**\n * Structured composition plan for ElevenLabs music generation. Mutually\n * exclusive with a free-form `prompt` on the `generateAudio()` call — when\n * supplied, `prompt` is ignored by ElevenLabs.\n *\n * We mirror the SDK's camelCase naming. Lengths are in milliseconds.\n * @see https://elevenlabs.io/docs/api-reference/music/compose\n */\nexport interface ElevenLabsMusicCompositionPlan {\n  /** Positive global style descriptors (mood, instruments, tempo, …). */\n  positiveGlobalStyles?: Array<string>\n  /** Negative global style descriptors — styles to avoid. */\n  negativeGlobalStyles?: Array<string>\n  /** Section definitions (verse/chorus/bridge/…) with local style hints. */\n  sections?: Array<{\n    sectionName: string\n    positiveLocalStyles?: Array<string>\n    negativeLocalStyles?: Array<string>\n    durationMs?: number\n    lines?: Array<string>\n  }>\n}\n\n/**\n * Provider options common to all ElevenLabs audio endpoints.\n */\ninterface CommonAudioOptions {\n  /** Output audio format. Defaults to `mp3_44100_128`. */\n  outputFormat?: ElevenLabsOutputFormat\n}\n\n/**\n * Provider options for music generation (`music_v1`).\n */\nexport interface ElevenLabsMusicProviderOptions extends CommonAudioOptions {\n  /** Structured composition plan. Mutually exclusive with `prompt`/`duration`. */\n  compositionPlan?: ElevenLabsMusicCompositionPlan\n  /** Deterministic sampling seed (incompatible with `prompt`). */\n  seed?: number\n  /** Force the output to be purely instrumental (prompt-mode only). */\n  forceInstrumental?: boolean\n  /** Strictly respect section durations in `compositionPlan`. */\n  respectSectionsDurations?: boolean\n}\n\n/**\n * Provider options for sound-effect generation (`eleven_text_to_sound_v*`).\n */\nexport interface ElevenLabsSoundEffectsProviderOptions extends CommonAudioOptions {\n  /** Prompt influence, 0..1. Default 0.3. Higher = more prompt adherence. */\n  promptInfluence?: number\n  /** Generate a loopable SFX (v2 only). */\n  loop?: boolean\n}\n\n/**\n * Union of per-model provider options. We keep both branches on one type so\n * the adapter stays tree-shakeable; callers narrow by model at the factory.\n */\nexport type ElevenLabsAudioProviderOptions =\n  | (ElevenLabsMusicProviderOptions & ElevenLabsSoundEffectsProviderOptions)\n  | ElevenLabsMusicProviderOptions\n  | ElevenLabsSoundEffectsProviderOptions\n\n/**\n * ElevenLabs audio generation adapter. Dispatches to music or SFX endpoints\n * based on the model id. Music → `client.music.compose`, SFX →\n * `client.textToSoundEffects.convert`.\n *\n * @example\n * ```ts\n * const music = elevenlabsAudio('music_v1')\n * await generateAudio({ adapter: music, prompt: 'lo-fi beat', duration: 15 })\n *\n * const sfx = elevenlabsAudio('eleven_text_to_sound_v2')\n * await generateAudio({ adapter: sfx, prompt: 'glass shattering', duration: 3 })\n * ```\n */\nexport class ElevenLabsAudioAdapter<\n  TModel extends ElevenLabsAudioModel,\n> extends BaseAudioAdapter<TModel, ElevenLabsAudioProviderOptions> {\n  readonly name = 'elevenlabs' as const\n\n  private readonly client: ElevenLabsClient\n\n  constructor(model: TModel, config?: ElevenLabsClientConfig) {\n    super(model, config ?? {})\n    this.client = createElevenLabsClient(config)\n  }\n\n  async generateAudio(\n    options: AudioGenerationOptions<ElevenLabsAudioProviderOptions>,\n  ): Promise<AudioGenerationResult> {\n    const { logger } = options\n    logger.request(\n      `activity=generateAudio provider=elevenlabs model=${this.model}`,\n      { provider: 'elevenlabs', model: this.model },\n    )\n    try {\n      if (isElevenLabsMusicModel(this.model)) {\n        return await this.runMusic(options)\n      }\n      if (isElevenLabsSoundEffectsModel(this.model)) {\n        return await this.runSoundEffects(options)\n      }\n      throw new Error(\n        `Unsupported ElevenLabs audio model \"${this.model}\". Expected one of: music_v1, eleven_text_to_sound_v2, eleven_text_to_sound_v1.`,\n      )\n    } catch (error) {\n      logger.errors('elevenlabs.generateAudio fatal', {\n        error,\n        source: 'elevenlabs.generateAudio',\n      })\n      throw error\n    }\n  }\n\n  private async runMusic(\n    options: AudioGenerationOptions<ElevenLabsAudioProviderOptions>,\n  ): Promise<AudioGenerationResult> {\n    // Gated by isElevenLabsMusicModel() in generateAudio().\n    const modelId = this.model as ElevenLabsMusicModel\n    const music = (options.modelOptions ?? {}) as ElevenLabsMusicProviderOptions\n    const outputFormat = music.outputFormat\n\n    const stream = await this.client.music.compose({\n      modelId,\n      ...(options.prompt && !music.compositionPlan\n        ? { prompt: options.prompt }\n        : {}),\n      ...(music.compositionPlan\n        ? { compositionPlan: toMusicPrompt(music.compositionPlan) }\n        : {}),\n      ...(options.duration != null && !music.compositionPlan\n        ? { musicLengthMs: Math.round(options.duration * 1000) }\n        : {}),\n      ...(outputFormat ? { outputFormat } : {}),\n      ...(music.seed != null ? { seed: music.seed } : {}),\n      ...(music.forceInstrumental != null\n        ? { forceInstrumental: music.forceInstrumental }\n        : {}),\n      ...(music.respectSectionsDurations != null\n        ? { respectSectionsDurations: music.respectSectionsDurations }\n        : {}),\n    })\n\n    return this.finalize(stream, outputFormat, options.duration)\n  }\n\n  private async runSoundEffects(\n    options: AudioGenerationOptions<ElevenLabsAudioProviderOptions>,\n  ): Promise<AudioGenerationResult> {\n    // Gated by isElevenLabsSoundEffectsModel() in generateAudio().\n    const modelId = this.model as ElevenLabsSoundEffectsModel\n    const sfx = (options.modelOptions ??\n      {}) as ElevenLabsSoundEffectsProviderOptions\n    const outputFormat = sfx.outputFormat\n\n    const stream = await this.client.textToSoundEffects.convert({\n      text: options.prompt,\n      modelId,\n      ...(options.duration != null\n        ? { durationSeconds: options.duration }\n        : {}),\n      ...(outputFormat ? { outputFormat } : {}),\n      ...(sfx.promptInfluence != null\n        ? { promptInfluence: sfx.promptInfluence }\n        : {}),\n      ...(sfx.loop != null ? { loop: sfx.loop } : {}),\n    })\n\n    return this.finalize(stream, outputFormat, options.duration)\n  }\n\n  private async finalize(\n    stream: ReadableStream<Uint8Array>,\n    outputFormat: ElevenLabsOutputFormat | undefined,\n    duration: number | undefined,\n  ): Promise<AudioGenerationResult> {\n    const buffer = await readStreamToArrayBuffer(stream)\n    const base64 = arrayBufferToBase64(buffer)\n    const { contentType } = parseOutputFormat(outputFormat)\n    return {\n      id: generateId(this.name),\n      model: this.model,\n      audio: {\n        b64Json: base64,\n        contentType,\n        ...(duration != null ? { duration } : {}),\n      },\n    }\n  }\n\n  protected override generateId(): string {\n    return generateId(this.name)\n  }\n}\n\nfunction toMusicPrompt(plan: ElevenLabsMusicCompositionPlan) {\n  return {\n    positiveGlobalStyles: plan.positiveGlobalStyles ?? [],\n    negativeGlobalStyles: plan.negativeGlobalStyles ?? [],\n    sections: (plan.sections ?? []).map((section) => ({\n      sectionName: section.sectionName,\n      positiveLocalStyles: section.positiveLocalStyles ?? [],\n      negativeLocalStyles: section.negativeLocalStyles ?? [],\n      durationMs: section.durationMs ?? 10000,\n      lines: section.lines ?? [],\n    })),\n  }\n}\n\n/**\n * Create an ElevenLabs audio adapter using `ELEVENLABS_API_KEY` from env.\n */\nexport function elevenlabsAudio<TModel extends ElevenLabsAudioModel>(\n  model: TModel,\n  config?: ElevenLabsClientConfig,\n): ElevenLabsAudioAdapter<TModel> {\n  return new ElevenLabsAudioAdapter(model, config)\n}\n\n/**\n * Create an ElevenLabs audio adapter with an explicit API key.\n */\nexport function createElevenLabsAudio<TModel extends ElevenLabsAudioModel>(\n  model: TModel,\n  apiKey: string,\n  config?: Omit<ElevenLabsClientConfig, 'apiKey'>,\n): ElevenLabsAudioAdapter<TModel> {\n  return new ElevenLabsAudioAdapter(model, { apiKey, ...config })\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;AAuGA,IAAa,yBAAb,cAEU,iBAAyD;CACjE,OAAgB;CAEhB;CAEA,YAAY,OAAe,QAAiC;EAC1D,MAAM,OAAO,UAAU,CAAC,CAAC;EACzB,KAAK,SAAS,uBAAuB,MAAM;CAC7C;CAEA,MAAM,cACJ,SACgC;EAChC,MAAM,EAAE,WAAW;EACnB,OAAO,QACL,oDAAoD,KAAK,SACzD;GAAE,UAAU;GAAc,OAAO,KAAK;EAAM,CAC9C;EACA,IAAI;GACF,IAAI,uBAAuB,KAAK,KAAK,GACnC,OAAO,MAAM,KAAK,SAAS,OAAO;GAEpC,IAAI,8BAA8B,KAAK,KAAK,GAC1C,OAAO,MAAM,KAAK,gBAAgB,OAAO;GAE3C,MAAM,IAAI,MACR,uCAAuC,KAAK,MAAM,gFACpD;EACF,SAAS,OAAO;GACd,OAAO,OAAO,kCAAkC;IAC9C;IACA,QAAQ;GACV,CAAC;GACD,MAAM;EACR;CACF;CAEA,MAAc,SACZ,SACgC;EAEhC,MAAM,UAAU,KAAK;EACrB,MAAM,QAAS,QAAQ,gBAAgB,CAAC;EACxC,MAAM,eAAe,MAAM;EAE3B,MAAM,SAAS,MAAM,KAAK,OAAO,MAAM,QAAQ;GAC7C;GACA,GAAI,QAAQ,UAAU,CAAC,MAAM,kBACzB,EAAE,QAAQ,QAAQ,OAAO,IACzB,CAAC;GACL,GAAI,MAAM,kBACN,EAAE,iBAAiB,cAAc,MAAM,eAAe,EAAE,IACxD,CAAC;GACL,GAAI,QAAQ,YAAY,QAAQ,CAAC,MAAM,kBACnC,EAAE,eAAe,KAAK,MAAM,QAAQ,WAAW,GAAI,EAAE,IACrD,CAAC;GACL,GAAI,eAAe,EAAE,aAAa,IAAI,CAAC;GACvC,GAAI,MAAM,QAAQ,OAAO,EAAE,MAAM,MAAM,KAAK,IAAI,CAAC;GACjD,GAAI,MAAM,qBAAqB,OAC3B,EAAE,mBAAmB,MAAM,kBAAkB,IAC7C,CAAC;GACL,GAAI,MAAM,4BAA4B,OAClC,EAAE,0BAA0B,MAAM,yBAAyB,IAC3D,CAAC;EACP,CAAC;EAED,OAAO,KAAK,SAAS,QAAQ,cAAc,QAAQ,QAAQ;CAC7D;CAEA,MAAc,gBACZ,SACgC;EAEhC,MAAM,UAAU,KAAK;EACrB,MAAM,MAAO,QAAQ,gBACnB,CAAC;EACH,MAAM,eAAe,IAAI;EAEzB,MAAM,SAAS,MAAM,KAAK,OAAO,mBAAmB,QAAQ;GAC1D,MAAM,QAAQ;GACd;GACA,GAAI,QAAQ,YAAY,OACpB,EAAE,iBAAiB,QAAQ,SAAS,IACpC,CAAC;GACL,GAAI,eAAe,EAAE,aAAa,IAAI,CAAC;GACvC,GAAI,IAAI,mBAAmB,OACvB,EAAE,iBAAiB,IAAI,gBAAgB,IACvC,CAAC;GACL,GAAI,IAAI,QAAQ,OAAO,EAAE,MAAM,IAAI,KAAK,IAAI,CAAC;EAC/C,CAAC;EAED,OAAO,KAAK,SAAS,QAAQ,cAAc,QAAQ,QAAQ;CAC7D;CAEA,MAAc,SACZ,QACA,cACA,UACgC;EAChC,MAAM,SAAS,MAAM,wBAAwB,MAAM;EACnD,MAAM,SAAS,oBAAoB,MAAM;EACzC,MAAM,EAAE,gBAAgB,kBAAkB,YAAY;EACtD,OAAO;GACL,IAAI,WAAW,KAAK,IAAI;GACxB,OAAO,KAAK;GACZ,OAAO;IACL,SAAS;IACT;IACA,GAAI,YAAY,OAAO,EAAE,SAAS,IAAI,CAAC;GACzC;EACF;CACF;CAEA,aAAwC;EACtC,OAAO,WAAW,KAAK,IAAI;CAC7B;AACF;AAEA,SAAS,cAAc,MAAsC;CAC3D,OAAO;EACL,sBAAsB,KAAK,wBAAwB,CAAC;EACpD,sBAAsB,KAAK,wBAAwB,CAAC;EACpD,WAAW,KAAK,YAAY,CAAC,EAAA,CAAG,KAAK,aAAa;GAChD,aAAa,QAAQ;GACrB,qBAAqB,QAAQ,uBAAuB,CAAC;GACrD,qBAAqB,QAAQ,uBAAuB,CAAC;GACrD,YAAY,QAAQ,cAAc;GAClC,OAAO,QAAQ,SAAS,CAAC;EAC3B,EAAE;CACJ;AACF;;;;AAKA,SAAgB,gBACd,OACA,QACgC;CAChC,OAAO,IAAI,uBAAuB,OAAO,MAAM;AACjD;;;;AAKA,SAAgB,sBACd,OACA,QACA,QACgC;CAChC,OAAO,IAAI,uBAAuB,OAAO;EAAE;EAAQ,GAAG;CAAO,CAAC;AAChE"}
+{"version":3,"file":"audio.js","names":[],"sources":["../../../src/adapters/audio.ts"],"sourcesContent":["import { BaseAudioAdapter } from '@tanstack/ai/adapters'\nimport {\n  arrayBufferToBase64,\n  createElevenLabsClient,\n  generateId,\n  parseOutputFormat,\n  readStreamToArrayBuffer,\n} from '../utils/client'\nimport {\n  ELEVENLABS_AUDIO_MODELS,\n  isElevenLabsMusicModel,\n  isElevenLabsSoundEffectsModel,\n} from '../model-meta'\nimport type { ElevenLabsClient } from '@elevenlabs/elevenlabs-js'\nimport type {\n  AudioGenerationOptions,\n  AudioGenerationResult,\n} from '@tanstack/ai'\nimport type { ElevenLabsClientConfig } from '../utils/client'\nimport type {\n  ElevenLabsAudioModel,\n  ElevenLabsMusicModel,\n  ElevenLabsOutputFormat,\n  ElevenLabsSoundEffectsModel,\n} from '../model-meta'\n\n/**\n * Structured composition plan for ElevenLabs music generation. Mutually\n * exclusive with a free-form `prompt` on the `generateAudio()` call — when\n * supplied, `prompt` is ignored by ElevenLabs.\n *\n * We mirror the SDK's camelCase naming. Lengths are in milliseconds.\n * @see https://elevenlabs.io/docs/api-reference/music/compose\n */\nexport interface ElevenLabsMusicCompositionPlan {\n  /** Positive global style descriptors (mood, instruments, tempo, …). */\n  positiveGlobalStyles?: Array<string>\n  /** Negative global style descriptors — styles to avoid. */\n  negativeGlobalStyles?: Array<string>\n  /** Section definitions (verse/chorus/bridge/…) with local style hints. */\n  sections?: Array<{\n    sectionName: string\n    positiveLocalStyles?: Array<string>\n    negativeLocalStyles?: Array<string>\n    durationMs?: number\n    lines?: Array<string>\n  }>\n}\n\n/**\n * Provider options common to all ElevenLabs audio endpoints.\n */\ninterface CommonAudioOptions {\n  /** Output audio format. Defaults to `mp3_44100_128`. */\n  outputFormat?: ElevenLabsOutputFormat\n}\n\n/**\n * Provider options for music generation (`music_v*`).\n */\nexport interface ElevenLabsMusicProviderOptions extends CommonAudioOptions {\n  /** Structured composition plan. Mutually exclusive with `prompt`/`duration`. */\n  compositionPlan?: ElevenLabsMusicCompositionPlan\n  /** Deterministic sampling seed (incompatible with `prompt`). */\n  seed?: number\n  /** Force the output to be purely instrumental (prompt-mode only). */\n  forceInstrumental?: boolean\n  /** Strictly respect section durations in `compositionPlan`. */\n  respectSectionsDurations?: boolean\n}\n\n/**\n * Provider options for sound-effect generation (`eleven_text_to_sound_v*`).\n */\nexport interface ElevenLabsSoundEffectsProviderOptions extends CommonAudioOptions {\n  /** Prompt influence, 0..1. Default 0.3. Higher = more prompt adherence. */\n  promptInfluence?: number\n  /** Generate a loopable SFX (v2 only). */\n  loop?: boolean\n}\n\n/**\n * Union of per-model provider options. We keep both branches on one type so\n * the adapter stays tree-shakeable; callers narrow by model at the factory.\n */\nexport type ElevenLabsAudioProviderOptions =\n  | (ElevenLabsMusicProviderOptions & ElevenLabsSoundEffectsProviderOptions)\n  | ElevenLabsMusicProviderOptions\n  | ElevenLabsSoundEffectsProviderOptions\n\n/**\n * ElevenLabs audio generation adapter. Dispatches to music or SFX endpoints\n * based on the model id. Music → `client.music.compose`, SFX →\n * `client.textToSoundEffects.convert`.\n *\n * @example\n * ```ts\n * const music = elevenlabsAudio('music_v2_5')\n * await generateAudio({ adapter: music, prompt: 'lo-fi beat', duration: 15 })\n *\n * const sfx = elevenlabsAudio('eleven_text_to_sound_v2')\n * await generateAudio({ adapter: sfx, prompt: 'glass shattering', duration: 3 })\n * ```\n */\nexport class ElevenLabsAudioAdapter<\n  TModel extends ElevenLabsAudioModel,\n> extends BaseAudioAdapter<TModel, ElevenLabsAudioProviderOptions> {\n  readonly name = 'elevenlabs' as const\n\n  private readonly client: ElevenLabsClient\n\n  constructor(model: TModel, config?: ElevenLabsClientConfig) {\n    super(model, config ?? {})\n    this.client = createElevenLabsClient(config)\n  }\n\n  async generateAudio(\n    options: AudioGenerationOptions<ElevenLabsAudioProviderOptions>,\n  ): Promise<AudioGenerationResult> {\n    const { logger } = options\n    logger.request(\n      `activity=generateAudio provider=elevenlabs model=${this.model}`,\n      { provider: 'elevenlabs', model: this.model },\n    )\n    try {\n      if (isElevenLabsMusicModel(this.model)) {\n        return await this.runMusic(options)\n      }\n      if (isElevenLabsSoundEffectsModel(this.model)) {\n        return await this.runSoundEffects(options)\n      }\n      throw new Error(\n        `Unsupported ElevenLabs audio model \"${this.model}\". Expected one of: ${ELEVENLABS_AUDIO_MODELS.join(', ')}.`,\n      )\n    } catch (error) {\n      logger.errors('elevenlabs.generateAudio fatal', {\n        error,\n        source: 'elevenlabs.generateAudio',\n      })\n      throw error\n    }\n  }\n\n  private async runMusic(\n    options: AudioGenerationOptions<ElevenLabsAudioProviderOptions>,\n  ): Promise<AudioGenerationResult> {\n    // Gated by isElevenLabsMusicModel() in generateAudio().\n    const modelId = this.model as ElevenLabsMusicModel\n    const music = (options.modelOptions ?? {}) as ElevenLabsMusicProviderOptions\n    const outputFormat = music.outputFormat\n\n    const stream = await this.client.music.compose({\n      modelId,\n      ...(options.prompt && !music.compositionPlan\n        ? { prompt: options.prompt }\n        : {}),\n      ...(music.compositionPlan\n        ? { compositionPlan: toMusicPrompt(music.compositionPlan) }\n        : {}),\n      ...(options.duration != null && !music.compositionPlan\n        ? { musicLengthMs: Math.round(options.duration * 1000) }\n        : {}),\n      ...(outputFormat ? { outputFormat } : {}),\n      ...(music.seed != null ? { seed: music.seed } : {}),\n      ...(music.forceInstrumental != null\n        ? { forceInstrumental: music.forceInstrumental }\n        : {}),\n      ...(music.respectSectionsDurations != null\n        ? { respectSectionsDurations: music.respectSectionsDurations }\n        : {}),\n    })\n\n    return this.finalize(stream, outputFormat, options.duration)\n  }\n\n  private async runSoundEffects(\n    options: AudioGenerationOptions<ElevenLabsAudioProviderOptions>,\n  ): Promise<AudioGenerationResult> {\n    // Gated by isElevenLabsSoundEffectsModel() in generateAudio().\n    const modelId = this.model as ElevenLabsSoundEffectsModel\n    const sfx = (options.modelOptions ??\n      {}) as ElevenLabsSoundEffectsProviderOptions\n    const outputFormat = sfx.outputFormat\n\n    const stream = await this.client.textToSoundEffects.convert({\n      text: options.prompt,\n      modelId,\n      ...(options.duration != null\n        ? { durationSeconds: options.duration }\n        : {}),\n      ...(outputFormat ? { outputFormat } : {}),\n      ...(sfx.promptInfluence != null\n        ? { promptInfluence: sfx.promptInfluence }\n        : {}),\n      ...(sfx.loop != null ? { loop: sfx.loop } : {}),\n    })\n\n    return this.finalize(stream, outputFormat, options.duration)\n  }\n\n  private async finalize(\n    stream: ReadableStream<Uint8Array>,\n    outputFormat: ElevenLabsOutputFormat | undefined,\n    duration: number | undefined,\n  ): Promise<AudioGenerationResult> {\n    const buffer = await readStreamToArrayBuffer(stream)\n    const base64 = arrayBufferToBase64(buffer)\n    const { contentType } = parseOutputFormat(outputFormat)\n    return {\n      id: generateId(this.name),\n      model: this.model,\n      audio: {\n        b64Json: base64,\n        contentType,\n        ...(duration != null ? { duration } : {}),\n      },\n    }\n  }\n\n  protected override generateId(): string {\n    return generateId(this.name)\n  }\n}\n\nfunction toMusicPrompt(plan: ElevenLabsMusicCompositionPlan) {\n  return {\n    positiveGlobalStyles: plan.positiveGlobalStyles ?? [],\n    negativeGlobalStyles: plan.negativeGlobalStyles ?? [],\n    sections: (plan.sections ?? []).map((section) => ({\n      sectionName: section.sectionName,\n      positiveLocalStyles: section.positiveLocalStyles ?? [],\n      negativeLocalStyles: section.negativeLocalStyles ?? [],\n      durationMs: section.durationMs ?? 10000,\n      lines: section.lines ?? [],\n    })),\n  }\n}\n\n/**\n * Create an ElevenLabs audio adapter using `ELEVENLABS_API_KEY` from env.\n */\nexport function elevenlabsAudio<TModel extends ElevenLabsAudioModel>(\n  model: TModel,\n  config?: ElevenLabsClientConfig,\n): ElevenLabsAudioAdapter<TModel> {\n  return new ElevenLabsAudioAdapter(model, config)\n}\n\n/**\n * Create an ElevenLabs audio adapter with an explicit API key.\n */\nexport function createElevenLabsAudio<TModel extends ElevenLabsAudioModel>(\n  model: TModel,\n  apiKey: string,\n  config?: Omit<ElevenLabsClientConfig, 'apiKey'>,\n): ElevenLabsAudioAdapter<TModel> {\n  return new ElevenLabsAudioAdapter(model, { apiKey, ...config })\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;AAwGA,IAAa,yBAAb,cAEU,iBAAyD;CACjE,OAAgB;CAEhB;CAEA,YAAY,OAAe,QAAiC;EAC1D,MAAM,OAAO,UAAU,CAAC,CAAC;EACzB,KAAK,SAAS,uBAAuB,MAAM;CAC7C;CAEA,MAAM,cACJ,SACgC;EAChC,MAAM,EAAE,WAAW;EACnB,OAAO,QACL,oDAAoD,KAAK,SACzD;GAAE,UAAU;GAAc,OAAO,KAAK;EAAM,CAC9C;EACA,IAAI;GACF,IAAI,uBAAuB,KAAK,KAAK,GACnC,OAAO,MAAM,KAAK,SAAS,OAAO;GAEpC,IAAI,8BAA8B,KAAK,KAAK,GAC1C,OAAO,MAAM,KAAK,gBAAgB,OAAO;GAE3C,MAAM,IAAI,MACR,uCAAuC,KAAK,MAAM,sBAAsB,wBAAwB,KAAK,IAAI,EAAE,EAC7G;EACF,SAAS,OAAO;GACd,OAAO,OAAO,kCAAkC;IAC9C;IACA,QAAQ;GACV,CAAC;GACD,MAAM;EACR;CACF;CAEA,MAAc,SACZ,SACgC;EAEhC,MAAM,UAAU,KAAK;EACrB,MAAM,QAAS,QAAQ,gBAAgB,CAAC;EACxC,MAAM,eAAe,MAAM;EAE3B,MAAM,SAAS,MAAM,KAAK,OAAO,MAAM,QAAQ;GAC7C;GACA,GAAI,QAAQ,UAAU,CAAC,MAAM,kBACzB,EAAE,QAAQ,QAAQ,OAAO,IACzB,CAAC;GACL,GAAI,MAAM,kBACN,EAAE,iBAAiB,cAAc,MAAM,eAAe,EAAE,IACxD,CAAC;GACL,GAAI,QAAQ,YAAY,QAAQ,CAAC,MAAM,kBACnC,EAAE,eAAe,KAAK,MAAM,QAAQ,WAAW,GAAI,EAAE,IACrD,CAAC;GACL,GAAI,eAAe,EAAE,aAAa,IAAI,CAAC;GACvC,GAAI,MAAM,QAAQ,OAAO,EAAE,MAAM,MAAM,KAAK,IAAI,CAAC;GACjD,GAAI,MAAM,qBAAqB,OAC3B,EAAE,mBAAmB,MAAM,kBAAkB,IAC7C,CAAC;GACL,GAAI,MAAM,4BAA4B,OAClC,EAAE,0BAA0B,MAAM,yBAAyB,IAC3D,CAAC;EACP,CAAC;EAED,OAAO,KAAK,SAAS,QAAQ,cAAc,QAAQ,QAAQ;CAC7D;CAEA,MAAc,gBACZ,SACgC;EAEhC,MAAM,UAAU,KAAK;EACrB,MAAM,MAAO,QAAQ,gBACnB,CAAC;EACH,MAAM,eAAe,IAAI;EAEzB,MAAM,SAAS,MAAM,KAAK,OAAO,mBAAmB,QAAQ;GAC1D,MAAM,QAAQ;GACd;GACA,GAAI,QAAQ,YAAY,OACpB,EAAE,iBAAiB,QAAQ,SAAS,IACpC,CAAC;GACL,GAAI,eAAe,EAAE,aAAa,IAAI,CAAC;GACvC,GAAI,IAAI,mBAAmB,OACvB,EAAE,iBAAiB,IAAI,gBAAgB,IACvC,CAAC;GACL,GAAI,IAAI,QAAQ,OAAO,EAAE,MAAM,IAAI,KAAK,IAAI,CAAC;EAC/C,CAAC;EAED,OAAO,KAAK,SAAS,QAAQ,cAAc,QAAQ,QAAQ;CAC7D;CAEA,MAAc,SACZ,QACA,cACA,UACgC;EAChC,MAAM,SAAS,MAAM,wBAAwB,MAAM;EACnD,MAAM,SAAS,oBAAoB,MAAM;EACzC,MAAM,EAAE,gBAAgB,kBAAkB,YAAY;EACtD,OAAO;GACL,IAAI,WAAW,KAAK,IAAI;GACxB,OAAO,KAAK;GACZ,OAAO;IACL,SAAS;IACT;IACA,GAAI,YAAY,OAAO,EAAE,SAAS,IAAI,CAAC;GACzC;EACF;CACF;CAEA,aAAwC;EACtC,OAAO,WAAW,KAAK,IAAI;CAC7B;AACF;AAEA,SAAS,cAAc,MAAsC;CAC3D,OAAO;EACL,sBAAsB,KAAK,wBAAwB,CAAC;EACpD,sBAAsB,KAAK,wBAAwB,CAAC;EACpD,WAAW,KAAK,YAAY,CAAC,EAAA,CAAG,KAAK,aAAa;GAChD,aAAa,QAAQ;GACrB,qBAAqB,QAAQ,uBAAuB,CAAC;GACrD,qBAAqB,QAAQ,uBAAuB,CAAC;GACrD,YAAY,QAAQ,cAAc;GAClC,OAAO,QAAQ,SAAS,CAAC;EAC3B,EAAE;CACJ;AACF;;;;AAKA,SAAgB,gBACd,OACA,QACgC;CAChC,OAAO,IAAI,uBAAuB,OAAO,MAAM;AACjD;;;;AAKA,SAAgB,sBACd,OACA,QACA,QACgC;CAChC,OAAO,IAAI,uBAAuB,OAAO;EAAE;EAAQ,GAAG;CAAO,CAAC;AAChE"}

package/dist/esm/adapters/speech.d.ts

@@ -1,5 +1,5 @@
 import { BaseTTSAdapter } from '@tanstack/ai/adapters';
-import { TTSOptions, TTSResult } from '@tanstack/ai';
+import { ListVoicesOptions, ListVoicesResult, TTSCapabilities, TTSOptions, TTSResult } from '@tanstack/ai';
 import { ElevenLabsClientConfig } from '../utils/client.js';
 import { ElevenLabsOutputFormat, ElevenLabsTTSModel } from '../model-meta.js';
 /**
@@ -27,7 +27,7 @@ export interface ElevenLabsVoiceSettings {
 export interface ElevenLabsSpeechProviderOptions {
     /** ElevenLabs voice ID to synthesize. Required if `generateSpeech().voice` is not set. */
     voiceId?: string;
-    /** Output audio format encoded as `codec_samplerate[_bitrate]`. Defaults to `mp3_44100_128`. */
+    /** Output audio format encoded as `codec_samplerate[_bitrate]`. Overrides `format`, including WAV wrapping. Defaults to `mp3_44100_128`. */
     outputFormat?: ElevenLabsOutputFormat;
     /** Voice-settings overrides for this request only. */
     voiceSettings?: ElevenLabsVoiceSettings;
@@ -68,9 +68,21 @@ export interface ElevenLabsSpeechProviderOptions {
  */
 export declare class ElevenLabsSpeechAdapter<TModel extends ElevenLabsTTSModel> extends BaseTTSAdapter<TModel, ElevenLabsSpeechProviderOptions> {
     readonly name: "elevenlabs";
+    /**
+     * `textToDialogue` accepts up to 10 distinct voice ids, and both endpoints
+     * have a `…WithTimestamps` twin that returns character alignment.
+     */
+    readonly capabilities: TTSCapabilities;
     private readonly client;
     constructor(model: TModel, config?: ElevenLabsClientConfig);
     generateSpeech(options: TTSOptions<ElevenLabsSpeechProviderOptions>): Promise<TTSResult>;
+    /**
+     * List the voices this API key can use, via `GET /v1/voices`.
+     *
+     * The catalog is per-account and grows every time `generateVoice()` saves a
+     * voice, so it has to be read at runtime rather than shipped as a const.
+     */
+    listVoices(options?: ListVoicesOptions): Promise<ListVoicesResult>;
     protected generateId(): string;
 }
 /**

package/dist/esm/adapters/speech.js

@@ -17,6 +17,14 @@ import { BaseTTSAdapter } from "@tanstack/ai/adapters";
 */
 var ElevenLabsSpeechAdapter = class extends BaseTTSAdapter {
 	name = "elevenlabs";
+	/**
+	* `textToDialogue` accepts up to 10 distinct voice ids, and both endpoints
+	* have a `…WithTimestamps` twin that returns character alignment.
+	*/
+	capabilities = {
+		maxSpeakers: 10,
+		timestamps: true
+	};
 	client;
 	constructor(model, config) {
 		super(model, config ?? {});
@@ -29,33 +37,84 @@ var ElevenLabsSpeechAdapter = class extends BaseTTSAdapter {
 			model: this.model
 		});
 		try {
-			const voiceId = options.voice ?? options.modelOptions?.voiceId;
-			if (!voiceId) throw new Error("ElevenLabs TTS requires a voice. Pass `voice` on generateSpeech() or `voiceId` in modelOptions.");
 			const { outputFormat, voiceSettings, languageCode, seed, previousText, nextText, previousRequestIds, nextRequestIds, applyTextNormalization, applyLanguageTextNormalization, optimizeStreamingLatency, enableLogging } = options.modelOptions ?? {};
 			const effectiveOutputFormat = outputFormat ?? inferOutputFormatFromResponseFormat(options.format);
-			const stream = await this.client.textToSpeech.convert(voiceId, {
-				text: options.text,
+			const wrapAsWav = outputFormat == null && options.format === "wav";
+			const { format, contentType } = wrapAsWav ? {
+				format: "wav",
+				contentType: "audio/wav"
+			} : parseOutputFormat(effectiveOutputFormat);
+			const encodeAudio = (buffer) => arrayBufferToBase64(wrapAsWav ? wrapPcmAsWav(buffer) : buffer);
+			const encodeAudioBase64 = (audioBase64) => wrapAsWav ? encodeAudio(base64ToArrayBuffer(audioBase64)) : audioBase64;
+			const base = {
 				modelId: this.model,
 				...effectiveOutputFormat ? { outputFormat: effectiveOutputFormat } : {},
-				...voiceSettings ? { voiceSettings: mapVoiceSettings(voiceSettings, options.speed) } : options.speed != null ? { voiceSettings: { speed: options.speed } } : {},
 				...languageCode ? { languageCode } : {},
 				...seed != null ? { seed } : {},
+				...applyTextNormalization ? { applyTextNormalization } : {}
+			};
+			if (options.turns) {
+				const inputs = options.turns.map((turn) => ({
+					text: turn.text,
+					voiceId: turn.voice
+				}));
+				const dialogue = {
+					...base,
+					inputs,
+					...voiceSettings?.stability != null ? { settings: { stability: voiceSettings.stability } } : {}
+				};
+				if (options.timestamps) {
+					const response = await this.client.textToDialogue.convertWithTimestamps(dialogue);
+					return {
+						id: generateId(this.name),
+						model: this.model,
+						audio: encodeAudioBase64(response.audioBase64),
+						format,
+						contentType,
+						...toAlignmentFields(response.alignment, response.voiceSegments)
+					};
+				}
+				const stream = await this.client.textToDialogue.convert(dialogue);
+				return {
+					id: generateId(this.name),
+					model: this.model,
+					audio: encodeAudio(await readStreamToArrayBuffer(stream)),
+					format,
+					contentType
+				};
+			}
+			const voiceId = options.voice ?? options.modelOptions?.voiceId;
+			if (!voiceId) throw new Error("ElevenLabs TTS requires a voice. Pass `voice` on generateSpeech() or `voiceId` in modelOptions.");
+			const single = {
+				...base,
+				text: options.text,
+				...voiceSettings ? { voiceSettings: mapVoiceSettings(voiceSettings, options.speed) } : options.speed != null ? { voiceSettings: { speed: options.speed } } : {},
 				...previousText ? { previousText } : {},
 				...nextText ? { nextText } : {},
 				...previousRequestIds ? { previousRequestIds } : {},
 				...nextRequestIds ? { nextRequestIds } : {},
-				...applyTextNormalization ? { applyTextNormalization } : {},
 				...applyLanguageTextNormalization != null ? { applyLanguageTextNormalization } : {},
-				...optimizeStreamingLatency != null ? { optimizeStreamingLatency } : {},
 				...enableLogging != null ? { enableLogging } : {}
+			};
+			if (options.timestamps) {
+				const response = await this.client.textToSpeech.convertWithTimestamps(voiceId, single);
+				return {
+					id: generateId(this.name),
+					model: this.model,
+					audio: encodeAudioBase64(response.audioBase64),
+					format,
+					contentType,
+					...toAlignmentFields(response.alignment, void 0)
+				};
+			}
+			const stream = await this.client.textToSpeech.convert(voiceId, {
+				...single,
+				...optimizeStreamingLatency != null ? { optimizeStreamingLatency } : {}
 			});
-			const buffer = await readStreamToArrayBuffer(stream);
-			const base64 = arrayBufferToBase64(buffer);
-			const { format, contentType } = parseOutputFormat(effectiveOutputFormat);
 			return {
 				id: generateId(this.name),
 				model: this.model,
-				audio: base64,
+				audio: encodeAudio(await readStreamToArrayBuffer(stream)),
 				format,
 				contentType
 			};
@@ -67,10 +126,43 @@ var ElevenLabsSpeechAdapter = class extends BaseTTSAdapter {
 			throw error;
 		}
 	}
+	/**
+	* List the voices this API key can use, via `GET /v1/voices`.
+	*
+	* The catalog is per-account and grows every time `generateVoice()` saves a
+	* voice, so it has to be read at runtime rather than shipped as a const.
+	*/
+	async listVoices(options) {
+		const voices = (await this.client.voices.getAll({}, options?.abortSignal ? { abortSignal: options.abortSignal } : {})).voices.map(toCatalogVoice);
+		const origins = options?.origins;
+		return { voices: origins ? voices.filter((voice) => voice.origin && origins.includes(voice.origin)) : voices };
+	}
 	generateId() {
 		return generateId(this.name);
 	}
 };
+/**
+* Map the ElevenLabs timestamp payload onto the core `alignment` / `segments`
+* fields. `voiceSegments` only comes back from the dialogue endpoint; its
+* character indices slice the alignment into per-turn text.
+*/
+function toAlignmentFields(alignment, voiceSegments) {
+	const fields = {};
+	if (alignment) fields.alignment = {
+		unit: "character",
+		texts: alignment.characters,
+		startSeconds: alignment.characterStartTimesSeconds,
+		endSeconds: alignment.characterEndTimesSeconds
+	};
+	if (voiceSegments && voiceSegments.length > 0) fields.segments = voiceSegments.map((segment) => ({
+		startSeconds: segment.startTimeSeconds,
+		endSeconds: segment.endTimeSeconds,
+		turnIndex: segment.dialogueInputIndex,
+		voice: segment.voiceId,
+		...alignment ? { text: alignment.characters.slice(segment.characterStartIndex, segment.characterEndIndex).join("") } : {}
+	}));
+	return fields;
+}
 function mapVoiceSettings(settings, speedOverride) {
 	return {
 		...settings.stability != null ? { stability: settings.stability } : {},
@@ -81,6 +173,35 @@ function mapVoiceSettings(settings, speedOverride) {
 	};
 }
 /**
+* ElevenLabs voice categories map onto {@link VoiceOrigin} with two joins:
+* `famous` and `high_quality` are both curated tiers, so both read as
+* `'professional'`. An unrecognized category is dropped rather than guessed,
+* which keeps an `origins` filter from silently matching the wrong thing.
+*/
+function toVoiceOrigin(category) {
+	switch (category) {
+		case "premade": return "premade";
+		case "generated": return "generated";
+		case "cloned": return "cloned";
+		case "professional":
+		case "famous":
+		case "high_quality": return "professional";
+		case void 0:
+		default: return;
+	}
+}
+function toCatalogVoice(voice) {
+	const origin = toVoiceOrigin(voice.category);
+	return {
+		voiceId: voice.voiceId,
+		...voice.name ? { name: voice.name } : {},
+		...origin ? { origin } : {},
+		...voice.description ? { description: voice.description } : {},
+		...voice.previewUrl ? { previewUrl: voice.previewUrl } : {},
+		...voice.labels ? { labels: voice.labels } : {}
+	};
+}
+/**
 * Map the standard TTSOptions `format` (mp3/opus/aac/flac/wav/pcm) to a
 * reasonable ElevenLabs `outputFormat` so callers don't need to know the
 * full codec/samplerate string for the common case.
@@ -88,12 +209,40 @@ function mapVoiceSettings(settings, speedOverride) {
 function inferOutputFormatFromResponseFormat(format) {
 	switch (format) {
 		case "mp3": return "mp3_44100_128";
-		case "pcm": return "pcm_44100";
+		case "pcm":
+		case "wav": return "pcm_44100";
 		case "opus": return "opus_48000_128";
 		case void 0: return;
-		default: return "mp3_44100_128";
+		default: throw new Error(`ElevenLabs TTS does not support format '${format}'. Use mp3, pcm, opus, or wav.`);
 	}
 }
+function base64ToArrayBuffer(base64) {
+	const binary = atob(base64);
+	const bytes = new Uint8Array(binary.length);
+	for (let i = 0; i < binary.length; i += 1) bytes[i] = binary.charCodeAt(i);
+	return bytes.buffer;
+}
+/** Wrap ElevenLabs pcm_44100 (16-bit little-endian mono) in a RIFF/WAV container. */
+function wrapPcmAsWav(pcm) {
+	const wav = new ArrayBuffer(44 + pcm.byteLength);
+	const bytes = new Uint8Array(wav);
+	const header = new DataView(wav);
+	const encoder = new TextEncoder();
+	bytes.set(encoder.encode("RIFF"), 0);
+	header.setUint32(4, 36 + pcm.byteLength, true);
+	bytes.set(encoder.encode("WAVEfmt "), 8);
+	header.setUint32(16, 16, true);
+	header.setUint16(20, 1, true);
+	header.setUint16(22, 1, true);
+	header.setUint32(24, 44100, true);
+	header.setUint32(28, 88200, true);
+	header.setUint16(32, 2, true);
+	header.setUint16(34, 16, true);
+	bytes.set(encoder.encode("data"), 36);
+	header.setUint32(40, pcm.byteLength, true);
+	bytes.set(new Uint8Array(pcm), 44);
+	return wav;
+}
 /**
 * Create an ElevenLabs speech adapter using `ELEVENLABS_API_KEY` from env.
 */

package/dist/esm/adapters/speech.js.map

@@ -1 +1 @@
-{"version":3,"file":"speech.js","names":[],"sources":["../../../src/adapters/speech.ts"],"sourcesContent":["import { BaseTTSAdapter } from '@tanstack/ai/adapters'\nimport {\n  arrayBufferToBase64,\n  createElevenLabsClient,\n  generateId,\n  parseOutputFormat,\n  readStreamToArrayBuffer,\n} from '../utils/client'\nimport type { ElevenLabsClient } from '@elevenlabs/elevenlabs-js'\nimport type { TTSOptions, TTSResult } from '@tanstack/ai'\nimport type { ElevenLabsClientConfig } from '../utils/client'\nimport type { ElevenLabsOutputFormat, ElevenLabsTTSModel } from '../model-meta'\n\n/**\n * ElevenLabs voice settings overrides. All fields are optional — omitted\n * values fall back to the voice's stored defaults.\n * @see https://elevenlabs.io/docs/api-reference/text-to-speech/convert\n */\nexport interface ElevenLabsVoiceSettings {\n  /** Voice stability, 0..1. Default 0.5. */\n  stability?: number\n  /** Similarity boost, 0..1. Default 0.75. */\n  similarityBoost?: number\n  /** Style exaggeration, 0..1. Default 0. */\n  style?: number\n  /** Playback speed. Default 1.0. */\n  speed?: number\n  /** Clarity/presence boost. Default true. */\n  useSpeakerBoost?: boolean\n}\n\n/**\n * Provider-specific TTS options. `voice` on `generateSpeech()` takes priority\n * over `voiceId` here, but we expose the same field for callers that prefer\n * to keep voice configuration inside the adapter config.\n */\nexport interface ElevenLabsSpeechProviderOptions {\n  /** ElevenLabs voice ID to synthesize. Required if `generateSpeech().voice` is not set. */\n  voiceId?: string\n  /** Output audio format encoded as `codec_samplerate[_bitrate]`. Defaults to `mp3_44100_128`. */\n  outputFormat?: ElevenLabsOutputFormat\n  /** Voice-settings overrides for this request only. */\n  voiceSettings?: ElevenLabsVoiceSettings\n  /** ISO-639-1 language code to enforce (e.g. `'en'`, `'ja'`). */\n  languageCode?: string\n  /** Deterministic sampling seed, 0..4294967295. */\n  seed?: number\n  /** Previous text for stitching adjacent clips. */\n  previousText?: string\n  /** Next text for stitching adjacent clips. */\n  nextText?: string\n  /** Previous request IDs for stitching (max 3). */\n  previousRequestIds?: Array<string>\n  /** Next request IDs for stitching (max 3). */\n  nextRequestIds?: Array<string>\n  /** Text normalization toggle. Default `'auto'`. */\n  applyTextNormalization?: 'auto' | 'on' | 'off'\n  /** Language-specific text normalization (currently Japanese only, adds latency). */\n  applyLanguageTextNormalization?: boolean\n  /** Latency optimization level, 0..4. */\n  optimizeStreamingLatency?: number\n  /** Enable logging. Set false for zero-retention mode (enterprise only). */\n  enableLogging?: boolean\n}\n\n/**\n * ElevenLabs text-to-speech adapter built on the official\n * `@elevenlabs/elevenlabs-js` SDK.\n *\n * @example\n * ```ts\n * const adapter = elevenlabsSpeech('eleven_multilingual_v2')\n * const result = await generateSpeech({\n *   adapter,\n *   text: 'Hello, world!',\n *   voice: '21m00Tcm4TlvDq8ikWAM',\n * })\n * ```\n */\nexport class ElevenLabsSpeechAdapter<\n  TModel extends ElevenLabsTTSModel,\n> extends BaseTTSAdapter<TModel, ElevenLabsSpeechProviderOptions> {\n  readonly name = 'elevenlabs' as const\n\n  private readonly client: ElevenLabsClient\n\n  constructor(model: TModel, config?: ElevenLabsClientConfig) {\n    super(model, config ?? {})\n    this.client = createElevenLabsClient(config)\n  }\n\n  async generateSpeech(\n    options: TTSOptions<ElevenLabsSpeechProviderOptions>,\n  ): Promise<TTSResult> {\n    const { logger } = options\n    logger.request(\n      `activity=generateSpeech provider=elevenlabs model=${this.model}`,\n      { provider: 'elevenlabs', model: this.model },\n    )\n    try {\n      const voiceId = options.voice ?? options.modelOptions?.voiceId\n      if (!voiceId) {\n        throw new Error(\n          'ElevenLabs TTS requires a voice. Pass `voice` on generateSpeech() or `voiceId` in modelOptions.',\n        )\n      }\n      const {\n        outputFormat,\n        voiceSettings,\n        languageCode,\n        seed,\n        previousText,\n        nextText,\n        previousRequestIds,\n        nextRequestIds,\n        applyTextNormalization,\n        applyLanguageTextNormalization,\n        optimizeStreamingLatency,\n        enableLogging,\n      } = options.modelOptions ?? {}\n      const effectiveOutputFormat =\n        outputFormat ?? inferOutputFormatFromResponseFormat(options.format)\n\n      const stream = await this.client.textToSpeech.convert(voiceId, {\n        text: options.text,\n        modelId: this.model,\n        ...(effectiveOutputFormat\n          ? { outputFormat: effectiveOutputFormat }\n          : {}),\n        ...(voiceSettings\n          ? { voiceSettings: mapVoiceSettings(voiceSettings, options.speed) }\n          : options.speed != null\n            ? { voiceSettings: { speed: options.speed } }\n            : {}),\n        ...(languageCode ? { languageCode } : {}),\n        ...(seed != null ? { seed } : {}),\n        ...(previousText ? { previousText } : {}),\n        ...(nextText ? { nextText } : {}),\n        ...(previousRequestIds ? { previousRequestIds } : {}),\n        ...(nextRequestIds ? { nextRequestIds } : {}),\n        ...(applyTextNormalization ? { applyTextNormalization } : {}),\n        ...(applyLanguageTextNormalization != null\n          ? { applyLanguageTextNormalization }\n          : {}),\n        ...(optimizeStreamingLatency != null\n          ? { optimizeStreamingLatency }\n          : {}),\n        ...(enableLogging != null ? { enableLogging } : {}),\n      })\n\n      const buffer = await readStreamToArrayBuffer(stream)\n      const base64 = arrayBufferToBase64(buffer)\n      const { format, contentType } = parseOutputFormat(effectiveOutputFormat)\n\n      return {\n        id: generateId(this.name),\n        model: this.model,\n        audio: base64,\n        format,\n        contentType,\n      }\n    } catch (error) {\n      logger.errors('elevenlabs.generateSpeech fatal', {\n        error,\n        source: 'elevenlabs.generateSpeech',\n      })\n      throw error\n    }\n  }\n\n  protected override generateId(): string {\n    return generateId(this.name)\n  }\n}\n\nfunction mapVoiceSettings(\n  settings: ElevenLabsVoiceSettings,\n  speedOverride: number | undefined,\n): Record<string, unknown> {\n  return {\n    ...(settings.stability != null ? { stability: settings.stability } : {}),\n    ...(settings.similarityBoost != null\n      ? { similarityBoost: settings.similarityBoost }\n      : {}),\n    ...(settings.style != null ? { style: settings.style } : {}),\n    ...(speedOverride != null\n      ? { speed: speedOverride }\n      : settings.speed != null\n        ? { speed: settings.speed }\n        : {}),\n    ...(settings.useSpeakerBoost != null\n      ? { useSpeakerBoost: settings.useSpeakerBoost }\n      : {}),\n  }\n}\n\n/**\n * Map the standard TTSOptions `format` (mp3/opus/aac/flac/wav/pcm) to a\n * reasonable ElevenLabs `outputFormat` so callers don't need to know the\n * full codec/samplerate string for the common case.\n */\nfunction inferOutputFormatFromResponseFormat(\n  format: TTSOptions['format'] | undefined,\n): ElevenLabsOutputFormat | undefined {\n  switch (format) {\n    case 'mp3':\n      return 'mp3_44100_128'\n    case 'pcm':\n      return 'pcm_44100'\n    case 'opus':\n      return 'opus_48000_128'\n    case undefined:\n      return undefined\n    case 'aac':\n    case 'flac':\n    case 'wav':\n    default:\n      // `aac` / `flac` / `wav` are not native ElevenLabs formats —\n      // fall back to mp3 rather than blowing up mid-request.\n      return 'mp3_44100_128'\n  }\n}\n\n/**\n * Create an ElevenLabs speech adapter using `ELEVENLABS_API_KEY` from env.\n */\nexport function elevenlabsSpeech<TModel extends ElevenLabsTTSModel>(\n  model: TModel,\n  config?: ElevenLabsClientConfig,\n): ElevenLabsSpeechAdapter<TModel> {\n  return new ElevenLabsSpeechAdapter(model, config)\n}\n\n/**\n * Create an ElevenLabs speech adapter with an explicit API key.\n */\nexport function createElevenLabsSpeech<TModel extends ElevenLabsTTSModel>(\n  model: TModel,\n  apiKey: string,\n  config?: Omit<ElevenLabsClientConfig, 'apiKey'>,\n): ElevenLabsSpeechAdapter<TModel> {\n  return new ElevenLabsSpeechAdapter(model, { apiKey, ...config })\n}\n"],"mappings":";;;;;;;;;;;;;;;;;AA+EA,IAAa,0BAAb,cAEU,eAAwD;CAChE,OAAgB;CAEhB;CAEA,YAAY,OAAe,QAAiC;EAC1D,MAAM,OAAO,UAAU,CAAC,CAAC;EACzB,KAAK,SAAS,uBAAuB,MAAM;CAC7C;CAEA,MAAM,eACJ,SACoB;EACpB,MAAM,EAAE,WAAW;EACnB,OAAO,QACL,qDAAqD,KAAK,SAC1D;GAAE,UAAU;GAAc,OAAO,KAAK;EAAM,CAC9C;EACA,IAAI;GACF,MAAM,UAAU,QAAQ,SAAS,QAAQ,cAAc;GACvD,IAAI,CAAC,SACH,MAAM,IAAI,MACR,iGACF;GAEF,MAAM,EACJ,cACA,eACA,cACA,MACA,cACA,UACA,oBACA,gBACA,wBACA,gCACA,0BACA,kBACE,QAAQ,gBAAgB,CAAC;GAC7B,MAAM,wBACJ,gBAAgB,oCAAoC,QAAQ,MAAM;GAEpE,MAAM,SAAS,MAAM,KAAK,OAAO,aAAa,QAAQ,SAAS;IAC7D,MAAM,QAAQ;IACd,SAAS,KAAK;IACd,GAAI,wBACA,EAAE,cAAc,sBAAsB,IACtC,CAAC;IACL,GAAI,gBACA,EAAE,eAAe,iBAAiB,eAAe,QAAQ,KAAK,EAAE,IAChE,QAAQ,SAAS,OACf,EAAE,eAAe,EAAE,OAAO,QAAQ,MAAM,EAAE,IAC1C,CAAC;IACP,GAAI,eAAe,EAAE,aAAa,IAAI,CAAC;IACvC,GAAI,QAAQ,OAAO,EAAE,KAAK,IAAI,CAAC;IAC/B,GAAI,eAAe,EAAE,aAAa,IAAI,CAAC;IACvC,GAAI,WAAW,EAAE,SAAS,IAAI,CAAC;IAC/B,GAAI,qBAAqB,EAAE,mBAAmB,IAAI,CAAC;IACnD,GAAI,iBAAiB,EAAE,eAAe,IAAI,CAAC;IAC3C,GAAI,yBAAyB,EAAE,uBAAuB,IAAI,CAAC;IAC3D,GAAI,kCAAkC,OAClC,EAAE,+BAA+B,IACjC,CAAC;IACL,GAAI,4BAA4B,OAC5B,EAAE,yBAAyB,IAC3B,CAAC;IACL,GAAI,iBAAiB,OAAO,EAAE,cAAc,IAAI,CAAC;GACnD,CAAC;GAED,MAAM,SAAS,MAAM,wBAAwB,MAAM;GACnD,MAAM,SAAS,oBAAoB,MAAM;GACzC,MAAM,EAAE,QAAQ,gBAAgB,kBAAkB,qBAAqB;GAEvE,OAAO;IACL,IAAI,WAAW,KAAK,IAAI;IACxB,OAAO,KAAK;IACZ,OAAO;IACP;IACA;GACF;EACF,SAAS,OAAO;GACd,OAAO,OAAO,mCAAmC;IAC/C;IACA,QAAQ;GACV,CAAC;GACD,MAAM;EACR;CACF;CAEA,aAAwC;EACtC,OAAO,WAAW,KAAK,IAAI;CAC7B;AACF;AAEA,SAAS,iBACP,UACA,eACyB;CACzB,OAAO;EACL,GAAI,SAAS,aAAa,OAAO,EAAE,WAAW,SAAS,UAAU,IAAI,CAAC;EACtE,GAAI,SAAS,mBAAmB,OAC5B,EAAE,iBAAiB,SAAS,gBAAgB,IAC5C,CAAC;EACL,GAAI,SAAS,SAAS,OAAO,EAAE,OAAO,SAAS,MAAM,IAAI,CAAC;EAC1D,GAAI,iBAAiB,OACjB,EAAE,OAAO,cAAc,IACvB,SAAS,SAAS,OAChB,EAAE,OAAO,SAAS,MAAM,IACxB,CAAC;EACP,GAAI,SAAS,mBAAmB,OAC5B,EAAE,iBAAiB,SAAS,gBAAgB,IAC5C,CAAC;CACP;AACF;;;;;;AAOA,SAAS,oCACP,QACoC;CACpC,QAAQ,QAAR;EACE,KAAK,OACH,OAAO;EACT,KAAK,OACH,OAAO;EACT,KAAK,QACH,OAAO;EACT,KAAK,KAAA,GACH;EAIF,SAGE,OAAO;CACX;AACF;;;;AAKA,SAAgB,iBACd,OACA,QACiC;CACjC,OAAO,IAAI,wBAAwB,OAAO,MAAM;AAClD;;;;AAKA,SAAgB,uBACd,OACA,QACA,QACiC;CACjC,OAAO,IAAI,wBAAwB,OAAO;EAAE;EAAQ,GAAG;CAAO,CAAC;AACjE"}
+{"version":3,"file":"speech.js","names":[],"sources":["../../../src/adapters/speech.ts"],"sourcesContent":["import { BaseTTSAdapter } from '@tanstack/ai/adapters'\nimport {\n  arrayBufferToBase64,\n  createElevenLabsClient,\n  generateId,\n  parseOutputFormat,\n  readStreamToArrayBuffer,\n} from '../utils/client'\nimport type { ElevenLabs, ElevenLabsClient } from '@elevenlabs/elevenlabs-js'\nimport type {\n  CatalogVoice,\n  ListVoicesOptions,\n  ListVoicesResult,\n  TTSAlignment,\n  TTSCapabilities,\n  TTSOptions,\n  TTSResult,\n  TTSSegment,\n  VoiceOrigin,\n} from '@tanstack/ai'\nimport type { ElevenLabsClientConfig } from '../utils/client'\nimport type { ElevenLabsOutputFormat, ElevenLabsTTSModel } from '../model-meta'\n\n/**\n * ElevenLabs voice settings overrides. All fields are optional — omitted\n * values fall back to the voice's stored defaults.\n * @see https://elevenlabs.io/docs/api-reference/text-to-speech/convert\n */\nexport interface ElevenLabsVoiceSettings {\n  /** Voice stability, 0..1. Default 0.5. */\n  stability?: number\n  /** Similarity boost, 0..1. Default 0.75. */\n  similarityBoost?: number\n  /** Style exaggeration, 0..1. Default 0. */\n  style?: number\n  /** Playback speed. Default 1.0. */\n  speed?: number\n  /** Clarity/presence boost. Default true. */\n  useSpeakerBoost?: boolean\n}\n\n/**\n * Provider-specific TTS options. `voice` on `generateSpeech()` takes priority\n * over `voiceId` here, but we expose the same field for callers that prefer\n * to keep voice configuration inside the adapter config.\n */\nexport interface ElevenLabsSpeechProviderOptions {\n  /** ElevenLabs voice ID to synthesize. Required if `generateSpeech().voice` is not set. */\n  voiceId?: string\n  /** Output audio format encoded as `codec_samplerate[_bitrate]`. Overrides `format`, including WAV wrapping. Defaults to `mp3_44100_128`. */\n  outputFormat?: ElevenLabsOutputFormat\n  /** Voice-settings overrides for this request only. */\n  voiceSettings?: ElevenLabsVoiceSettings\n  /** ISO-639-1 language code to enforce (e.g. `'en'`, `'ja'`). */\n  languageCode?: string\n  /** Deterministic sampling seed, 0..4294967295. */\n  seed?: number\n  /** Previous text for stitching adjacent clips. */\n  previousText?: string\n  /** Next text for stitching adjacent clips. */\n  nextText?: string\n  /** Previous request IDs for stitching (max 3). */\n  previousRequestIds?: Array<string>\n  /** Next request IDs for stitching (max 3). */\n  nextRequestIds?: Array<string>\n  /** Text normalization toggle. Default `'auto'`. */\n  applyTextNormalization?: 'auto' | 'on' | 'off'\n  /** Language-specific text normalization (currently Japanese only, adds latency). */\n  applyLanguageTextNormalization?: boolean\n  /** Latency optimization level, 0..4. */\n  optimizeStreamingLatency?: number\n  /** Enable logging. Set false for zero-retention mode (enterprise only). */\n  enableLogging?: boolean\n}\n\n/**\n * ElevenLabs text-to-speech adapter built on the official\n * `@elevenlabs/elevenlabs-js` SDK.\n *\n * @example\n * ```ts\n * const adapter = elevenlabsSpeech('eleven_multilingual_v2')\n * const result = await generateSpeech({\n *   adapter,\n *   text: 'Hello, world!',\n *   voice: '21m00Tcm4TlvDq8ikWAM',\n * })\n * ```\n */\nexport class ElevenLabsSpeechAdapter<\n  TModel extends ElevenLabsTTSModel,\n> extends BaseTTSAdapter<TModel, ElevenLabsSpeechProviderOptions> {\n  readonly name = 'elevenlabs' as const\n\n  /**\n   * `textToDialogue` accepts up to 10 distinct voice ids, and both endpoints\n   * have a `…WithTimestamps` twin that returns character alignment.\n   */\n  override readonly capabilities: TTSCapabilities = {\n    maxSpeakers: 10,\n    timestamps: true,\n  }\n\n  private readonly client: ElevenLabsClient\n\n  constructor(model: TModel, config?: ElevenLabsClientConfig) {\n    super(model, config ?? {})\n    this.client = createElevenLabsClient(config)\n  }\n\n  async generateSpeech(\n    options: TTSOptions<ElevenLabsSpeechProviderOptions>,\n  ): Promise<TTSResult> {\n    const { logger } = options\n    logger.request(\n      `activity=generateSpeech provider=elevenlabs model=${this.model}`,\n      { provider: 'elevenlabs', model: this.model },\n    )\n    try {\n      const {\n        outputFormat,\n        voiceSettings,\n        languageCode,\n        seed,\n        previousText,\n        nextText,\n        previousRequestIds,\n        nextRequestIds,\n        applyTextNormalization,\n        applyLanguageTextNormalization,\n        optimizeStreamingLatency,\n        enableLogging,\n      } = options.modelOptions ?? {}\n      const effectiveOutputFormat =\n        outputFormat ?? inferOutputFormatFromResponseFormat(options.format)\n      const wrapAsWav = outputFormat == null && options.format === 'wav'\n      const { format, contentType } = wrapAsWav\n        ? { format: 'wav', contentType: 'audio/wav' }\n        : parseOutputFormat(effectiveOutputFormat)\n      // `format: 'wav'` asks ElevenLabs for pcm_44100 and wraps it here, on\n      // every endpoint: the byte-stream pair hands back raw bytes, the\n      // timestamped pair hands back the same bytes as base64.\n      const encodeAudio = (buffer: ArrayBuffer) =>\n        arrayBufferToBase64(wrapAsWav ? wrapPcmAsWav(buffer) : buffer)\n      const encodeAudioBase64 = (audioBase64: string) =>\n        wrapAsWav ? encodeAudio(base64ToArrayBuffer(audioBase64)) : audioBase64\n      const base = {\n        modelId: this.model,\n        ...(effectiveOutputFormat\n          ? { outputFormat: effectiveOutputFormat }\n          : {}),\n        ...(languageCode ? { languageCode } : {}),\n        ...(seed != null ? { seed } : {}),\n        ...(applyTextNormalization ? { applyTextNormalization } : {}),\n      }\n\n      // Four endpoints, picked by (turns?, timestamps?). The dialogue pair\n      // takes `inputs` instead of a voice id in the path, and the timestamped\n      // pair returns JSON (base64 audio + alignment) instead of a byte stream.\n      if (options.turns) {\n        const inputs = options.turns.map((turn) => ({\n          text: turn.text,\n          voiceId: turn.voice,\n        }))\n        const dialogue = {\n          ...base,\n          inputs,\n          ...(voiceSettings?.stability != null\n            ? { settings: { stability: voiceSettings.stability } }\n            : {}),\n        }\n\n        if (options.timestamps) {\n          const response =\n            await this.client.textToDialogue.convertWithTimestamps(dialogue)\n          return {\n            id: generateId(this.name),\n            model: this.model,\n            audio: encodeAudioBase64(response.audioBase64),\n            format,\n            contentType,\n            ...toAlignmentFields(response.alignment, response.voiceSegments),\n          }\n        }\n\n        const stream = await this.client.textToDialogue.convert(dialogue)\n        return {\n          id: generateId(this.name),\n          model: this.model,\n          audio: encodeAudio(await readStreamToArrayBuffer(stream)),\n          format,\n          contentType,\n        }\n      }\n\n      const voiceId = options.voice ?? options.modelOptions?.voiceId\n      if (!voiceId) {\n        throw new Error(\n          'ElevenLabs TTS requires a voice. Pass `voice` on generateSpeech() or `voiceId` in modelOptions.',\n        )\n      }\n      const single = {\n        ...base,\n        text: options.text,\n        ...(voiceSettings\n          ? { voiceSettings: mapVoiceSettings(voiceSettings, options.speed) }\n          : options.speed != null\n            ? { voiceSettings: { speed: options.speed } }\n            : {}),\n        ...(previousText ? { previousText } : {}),\n        ...(nextText ? { nextText } : {}),\n        ...(previousRequestIds ? { previousRequestIds } : {}),\n        ...(nextRequestIds ? { nextRequestIds } : {}),\n        ...(applyLanguageTextNormalization != null\n          ? { applyLanguageTextNormalization }\n          : {}),\n        ...(enableLogging != null ? { enableLogging } : {}),\n      }\n\n      if (options.timestamps) {\n        const response = await this.client.textToSpeech.convertWithTimestamps(\n          voiceId,\n          single,\n        )\n        return {\n          id: generateId(this.name),\n          model: this.model,\n          audio: encodeAudioBase64(response.audioBase64),\n          format,\n          contentType,\n          ...toAlignmentFields(response.alignment, undefined),\n        }\n      }\n\n      const stream = await this.client.textToSpeech.convert(voiceId, {\n        ...single,\n        ...(optimizeStreamingLatency != null\n          ? { optimizeStreamingLatency }\n          : {}),\n      })\n\n      return {\n        id: generateId(this.name),\n        model: this.model,\n        audio: encodeAudio(await readStreamToArrayBuffer(stream)),\n        format,\n        contentType,\n      }\n    } catch (error) {\n      logger.errors('elevenlabs.generateSpeech fatal', {\n        error,\n        source: 'elevenlabs.generateSpeech',\n      })\n      throw error\n    }\n  }\n\n  /**\n   * List the voices this API key can use, via `GET /v1/voices`.\n   *\n   * The catalog is per-account and grows every time `generateVoice()` saves a\n   * voice, so it has to be read at runtime rather than shipped as a const.\n   */\n  override async listVoices(\n    options?: ListVoicesOptions,\n  ): Promise<ListVoicesResult> {\n    const response = await this.client.voices.getAll(\n      {},\n      options?.abortSignal ? { abortSignal: options.abortSignal } : {},\n    )\n\n    const voices = response.voices.map(toCatalogVoice)\n    const origins = options?.origins\n    // ElevenLabs has no origin filter on /v1/voices, so narrow in memory.\n    return {\n      voices: origins\n        ? voices.filter(\n            (voice) => voice.origin && origins.includes(voice.origin),\n          )\n        : voices,\n    }\n  }\n\n  protected override generateId(): string {\n    return generateId(this.name)\n  }\n}\n\n/**\n * Map the ElevenLabs timestamp payload onto the core `alignment` / `segments`\n * fields. `voiceSegments` only comes back from the dialogue endpoint; its\n * character indices slice the alignment into per-turn text.\n */\nfunction toAlignmentFields(\n  alignment: ElevenLabs.CharacterAlignmentResponseModel | undefined,\n  voiceSegments: Array<ElevenLabs.VoiceSegment> | undefined,\n): { alignment?: TTSAlignment; segments?: Array<TTSSegment> } {\n  const fields: { alignment?: TTSAlignment; segments?: Array<TTSSegment> } = {}\n  if (alignment) {\n    fields.alignment = {\n      unit: 'character',\n      texts: alignment.characters,\n      startSeconds: alignment.characterStartTimesSeconds,\n      endSeconds: alignment.characterEndTimesSeconds,\n    }\n  }\n  if (voiceSegments && voiceSegments.length > 0) {\n    fields.segments = voiceSegments.map((segment) => ({\n      startSeconds: segment.startTimeSeconds,\n      endSeconds: segment.endTimeSeconds,\n      turnIndex: segment.dialogueInputIndex,\n      voice: segment.voiceId,\n      ...(alignment\n        ? {\n            text: alignment.characters\n              .slice(segment.characterStartIndex, segment.characterEndIndex)\n              .join(''),\n          }\n        : {}),\n    }))\n  }\n  return fields\n}\n\nfunction mapVoiceSettings(\n  settings: ElevenLabsVoiceSettings,\n  speedOverride: number | undefined,\n): Record<string, unknown> {\n  return {\n    ...(settings.stability != null ? { stability: settings.stability } : {}),\n    ...(settings.similarityBoost != null\n      ? { similarityBoost: settings.similarityBoost }\n      : {}),\n    ...(settings.style != null ? { style: settings.style } : {}),\n    ...(speedOverride != null\n      ? { speed: speedOverride }\n      : settings.speed != null\n        ? { speed: settings.speed }\n        : {}),\n    ...(settings.useSpeakerBoost != null\n      ? { useSpeakerBoost: settings.useSpeakerBoost }\n      : {}),\n  }\n}\n\n/**\n * ElevenLabs voice categories map onto {@link VoiceOrigin} with two joins:\n * `famous` and `high_quality` are both curated tiers, so both read as\n * `'professional'`. An unrecognized category is dropped rather than guessed,\n * which keeps an `origins` filter from silently matching the wrong thing.\n */\nfunction toVoiceOrigin(\n  category: ElevenLabs.VoiceCategory | undefined,\n): VoiceOrigin | undefined {\n  switch (category) {\n    case 'premade':\n      return 'premade'\n    case 'generated':\n      return 'generated'\n    case 'cloned':\n      return 'cloned'\n    case 'professional':\n    case 'famous':\n    case 'high_quality':\n      return 'professional'\n    case undefined:\n    default:\n      return undefined\n  }\n}\n\nfunction toCatalogVoice(voice: ElevenLabs.Voice): CatalogVoice {\n  const origin = toVoiceOrigin(voice.category)\n  return {\n    voiceId: voice.voiceId,\n    ...(voice.name ? { name: voice.name } : {}),\n    ...(origin ? { origin } : {}),\n    ...(voice.description ? { description: voice.description } : {}),\n    ...(voice.previewUrl ? { previewUrl: voice.previewUrl } : {}),\n    ...(voice.labels ? { labels: voice.labels } : {}),\n  }\n}\n\n/**\n * Map the standard TTSOptions `format` (mp3/opus/aac/flac/wav/pcm) to a\n * reasonable ElevenLabs `outputFormat` so callers don't need to know the\n * full codec/samplerate string for the common case.\n */\nfunction inferOutputFormatFromResponseFormat(\n  format: TTSOptions['format'] | undefined,\n): ElevenLabsOutputFormat | undefined {\n  switch (format) {\n    case 'mp3':\n      return 'mp3_44100_128'\n    case 'pcm':\n    case 'wav':\n      return 'pcm_44100'\n    case 'opus':\n      return 'opus_48000_128'\n    case undefined:\n      return undefined\n    case 'aac':\n    case 'flac':\n    default:\n      throw new Error(\n        `ElevenLabs TTS does not support format '${format}'. Use mp3, pcm, opus, or wav.`,\n      )\n  }\n}\n\nfunction base64ToArrayBuffer(base64: string): ArrayBuffer {\n  const binary = atob(base64)\n  const bytes = new Uint8Array(binary.length)\n  for (let i = 0; i < binary.length; i += 1) bytes[i] = binary.charCodeAt(i)\n  return bytes.buffer\n}\n\n/** Wrap ElevenLabs pcm_44100 (16-bit little-endian mono) in a RIFF/WAV container. */\nfunction wrapPcmAsWav(pcm: ArrayBuffer): ArrayBuffer {\n  const wav = new ArrayBuffer(44 + pcm.byteLength)\n  const bytes = new Uint8Array(wav)\n  const header = new DataView(wav)\n  const encoder = new TextEncoder()\n  bytes.set(encoder.encode('RIFF'), 0)\n  header.setUint32(4, 36 + pcm.byteLength, true)\n  bytes.set(encoder.encode('WAVEfmt '), 8)\n  header.setUint32(16, 16, true) // PCM format chunk size\n  header.setUint16(20, 1, true) // PCM encoding\n  header.setUint16(22, 1, true) // mono\n  header.setUint32(24, 44100, true) // sample rate\n  header.setUint32(28, 44100 * 2, true) // bytes per second\n  header.setUint16(32, 2, true) // bytes per sample\n  header.setUint16(34, 16, true) // bits per sample\n  bytes.set(encoder.encode('data'), 36)\n  header.setUint32(40, pcm.byteLength, true)\n  bytes.set(new Uint8Array(pcm), 44)\n  return wav\n}\n\n/**\n * Create an ElevenLabs speech adapter using `ELEVENLABS_API_KEY` from env.\n */\nexport function elevenlabsSpeech<TModel extends ElevenLabsTTSModel>(\n  model: TModel,\n  config?: ElevenLabsClientConfig,\n): ElevenLabsSpeechAdapter<TModel> {\n  return new ElevenLabsSpeechAdapter(model, config)\n}\n\n/**\n * Create an ElevenLabs speech adapter with an explicit API key.\n */\nexport function createElevenLabsSpeech<TModel extends ElevenLabsTTSModel>(\n  model: TModel,\n  apiKey: string,\n  config?: Omit<ElevenLabsClientConfig, 'apiKey'>,\n): ElevenLabsSpeechAdapter<TModel> {\n  return new ElevenLabsSpeechAdapter(model, { apiKey, ...config })\n}\n"],"mappings":";;;;;;;;;;;;;;;;;AAyFA,IAAa,0BAAb,cAEU,eAAwD;CAChE,OAAgB;;;;;CAMhB,eAAkD;EAChD,aAAa;EACb,YAAY;CACd;CAEA;CAEA,YAAY,OAAe,QAAiC;EAC1D,MAAM,OAAO,UAAU,CAAC,CAAC;EACzB,KAAK,SAAS,uBAAuB,MAAM;CAC7C;CAEA,MAAM,eACJ,SACoB;EACpB,MAAM,EAAE,WAAW;EACnB,OAAO,QACL,qDAAqD,KAAK,SAC1D;GAAE,UAAU;GAAc,OAAO,KAAK;EAAM,CAC9C;EACA,IAAI;GACF,MAAM,EACJ,cACA,eACA,cACA,MACA,cACA,UACA,oBACA,gBACA,wBACA,gCACA,0BACA,kBACE,QAAQ,gBAAgB,CAAC;GAC7B,MAAM,wBACJ,gBAAgB,oCAAoC,QAAQ,MAAM;GACpE,MAAM,YAAY,gBAAgB,QAAQ,QAAQ,WAAW;GAC7D,MAAM,EAAE,QAAQ,gBAAgB,YAC5B;IAAE,QAAQ;IAAO,aAAa;GAAY,IAC1C,kBAAkB,qBAAqB;GAI3C,MAAM,eAAe,WACnB,oBAAoB,YAAY,aAAa,MAAM,IAAI,MAAM;GAC/D,MAAM,qBAAqB,gBACzB,YAAY,YAAY,oBAAoB,WAAW,CAAC,IAAI;GAC9D,MAAM,OAAO;IACX,SAAS,KAAK;IACd,GAAI,wBACA,EAAE,cAAc,sBAAsB,IACtC,CAAC;IACL,GAAI,eAAe,EAAE,aAAa,IAAI,CAAC;IACvC,GAAI,QAAQ,OAAO,EAAE,KAAK,IAAI,CAAC;IAC/B,GAAI,yBAAyB,EAAE,uBAAuB,IAAI,CAAC;GAC7D;GAKA,IAAI,QAAQ,OAAO;IACjB,MAAM,SAAS,QAAQ,MAAM,KAAK,UAAU;KAC1C,MAAM,KAAK;KACX,SAAS,KAAK;IAChB,EAAE;IACF,MAAM,WAAW;KACf,GAAG;KACH;KACA,GAAI,eAAe,aAAa,OAC5B,EAAE,UAAU,EAAE,WAAW,cAAc,UAAU,EAAE,IACnD,CAAC;IACP;IAEA,IAAI,QAAQ,YAAY;KACtB,MAAM,WACJ,MAAM,KAAK,OAAO,eAAe,sBAAsB,QAAQ;KACjE,OAAO;MACL,IAAI,WAAW,KAAK,IAAI;MACxB,OAAO,KAAK;MACZ,OAAO,kBAAkB,SAAS,WAAW;MAC7C;MACA;MACA,GAAG,kBAAkB,SAAS,WAAW,SAAS,aAAa;KACjE;IACF;IAEA,MAAM,SAAS,MAAM,KAAK,OAAO,eAAe,QAAQ,QAAQ;IAChE,OAAO;KACL,IAAI,WAAW,KAAK,IAAI;KACxB,OAAO,KAAK;KACZ,OAAO,YAAY,MAAM,wBAAwB,MAAM,CAAC;KACxD;KACA;IACF;GACF;GAEA,MAAM,UAAU,QAAQ,SAAS,QAAQ,cAAc;GACvD,IAAI,CAAC,SACH,MAAM,IAAI,MACR,iGACF;GAEF,MAAM,SAAS;IACb,GAAG;IACH,MAAM,QAAQ;IACd,GAAI,gBACA,EAAE,eAAe,iBAAiB,eAAe,QAAQ,KAAK,EAAE,IAChE,QAAQ,SAAS,OACf,EAAE,eAAe,EAAE,OAAO,QAAQ,MAAM,EAAE,IAC1C,CAAC;IACP,GAAI,eAAe,EAAE,aAAa,IAAI,CAAC;IACvC,GAAI,WAAW,EAAE,SAAS,IAAI,CAAC;IAC/B,GAAI,qBAAqB,EAAE,mBAAmB,IAAI,CAAC;IACnD,GAAI,iBAAiB,EAAE,eAAe,IAAI,CAAC;IAC3C,GAAI,kCAAkC,OAClC,EAAE,+BAA+B,IACjC,CAAC;IACL,GAAI,iBAAiB,OAAO,EAAE,cAAc,IAAI,CAAC;GACnD;GAEA,IAAI,QAAQ,YAAY;IACtB,MAAM,WAAW,MAAM,KAAK,OAAO,aAAa,sBAC9C,SACA,MACF;IACA,OAAO;KACL,IAAI,WAAW,KAAK,IAAI;KACxB,OAAO,KAAK;KACZ,OAAO,kBAAkB,SAAS,WAAW;KAC7C;KACA;KACA,GAAG,kBAAkB,SAAS,WAAW,KAAA,CAAS;IACpD;GACF;GAEA,MAAM,SAAS,MAAM,KAAK,OAAO,aAAa,QAAQ,SAAS;IAC7D,GAAG;IACH,GAAI,4BAA4B,OAC5B,EAAE,yBAAyB,IAC3B,CAAC;GACP,CAAC;GAED,OAAO;IACL,IAAI,WAAW,KAAK,IAAI;IACxB,OAAO,KAAK;IACZ,OAAO,YAAY,MAAM,wBAAwB,MAAM,CAAC;IACxD;IACA;GACF;EACF,SAAS,OAAO;GACd,OAAO,OAAO,mCAAmC;IAC/C;IACA,QAAQ;GACV,CAAC;GACD,MAAM;EACR;CACF;;;;;;;CAQA,MAAe,WACb,SAC2B;EAM3B,MAAM,UAAS,MALQ,KAAK,OAAO,OAAO,OACxC,CAAC,GACD,SAAS,cAAc,EAAE,aAAa,QAAQ,YAAY,IAAI,CAAC,CACjE,EAAA,CAEwB,OAAO,IAAI,cAAc;EACjD,MAAM,UAAU,SAAS;EAEzB,OAAO,EACL,QAAQ,UACJ,OAAO,QACJ,UAAU,MAAM,UAAU,QAAQ,SAAS,MAAM,MAAM,CAC1D,IACA,OACN;CACF;CAEA,aAAwC;EACtC,OAAO,WAAW,KAAK,IAAI;CAC7B;AACF;;;;;;AAOA,SAAS,kBACP,WACA,eAC4D;CAC5D,MAAM,SAAqE,CAAC;CAC5E,IAAI,WACF,OAAO,YAAY;EACjB,MAAM;EACN,OAAO,UAAU;EACjB,cAAc,UAAU;EACxB,YAAY,UAAU;CACxB;CAEF,IAAI,iBAAiB,cAAc,SAAS,GAC1C,OAAO,WAAW,cAAc,KAAK,aAAa;EAChD,cAAc,QAAQ;EACtB,YAAY,QAAQ;EACpB,WAAW,QAAQ;EACnB,OAAO,QAAQ;EACf,GAAI,YACA,EACE,MAAM,UAAU,WACb,MAAM,QAAQ,qBAAqB,QAAQ,iBAAiB,CAAC,CAC7D,KAAK,EAAE,EACZ,IACA,CAAC;CACP,EAAE;CAEJ,OAAO;AACT;AAEA,SAAS,iBACP,UACA,eACyB;CACzB,OAAO;EACL,GAAI,SAAS,aAAa,OAAO,EAAE,WAAW,SAAS,UAAU,IAAI,CAAC;EACtE,GAAI,SAAS,mBAAmB,OAC5B,EAAE,iBAAiB,SAAS,gBAAgB,IAC5C,CAAC;EACL,GAAI,SAAS,SAAS,OAAO,EAAE,OAAO,SAAS,MAAM,IAAI,CAAC;EAC1D,GAAI,iBAAiB,OACjB,EAAE,OAAO,cAAc,IACvB,SAAS,SAAS,OAChB,EAAE,OAAO,SAAS,MAAM,IACxB,CAAC;EACP,GAAI,SAAS,mBAAmB,OAC5B,EAAE,iBAAiB,SAAS,gBAAgB,IAC5C,CAAC;CACP;AACF;;;;;;;AAQA,SAAS,cACP,UACyB;CACzB,QAAQ,UAAR;EACE,KAAK,WACH,OAAO;EACT,KAAK,aACH,OAAO;EACT,KAAK,UACH,OAAO;EACT,KAAK;EACL,KAAK;EACL,KAAK,gBACH,OAAO;EACT,KAAK,KAAA;EACL,SACE;CACJ;AACF;AAEA,SAAS,eAAe,OAAuC;CAC7D,MAAM,SAAS,cAAc,MAAM,QAAQ;CAC3C,OAAO;EACL,SAAS,MAAM;EACf,GAAI,MAAM,OAAO,EAAE,MAAM,MAAM,KAAK,IAAI,CAAC;EACzC,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC;EAC3B,GAAI,MAAM,cAAc,EAAE,aAAa,MAAM,YAAY,IAAI,CAAC;EAC9D,GAAI,MAAM,aAAa,EAAE,YAAY,MAAM,WAAW,IAAI,CAAC;EAC3D,GAAI,MAAM,SAAS,EAAE,QAAQ,MAAM,OAAO,IAAI,CAAC;CACjD;AACF;;;;;;AAOA,SAAS,oCACP,QACoC;CACpC,QAAQ,QAAR;EACE,KAAK,OACH,OAAO;EACT,KAAK;EACL,KAAK,OACH,OAAO;EACT,KAAK,QACH,OAAO;EACT,KAAK,KAAA,GACH;EAGF,SACE,MAAM,IAAI,MACR,2CAA2C,OAAO,+BACpD;CACJ;AACF;AAEA,SAAS,oBAAoB,QAA6B;CACxD,MAAM,SAAS,KAAK,MAAM;CAC1B,MAAM,QAAQ,IAAI,WAAW,OAAO,MAAM;CAC1C,KAAK,IAAI,IAAI,GAAG,IAAI,OAAO,QAAQ,KAAK,GAAG,MAAM,KAAK,OAAO,WAAW,CAAC;CACzE,OAAO,MAAM;AACf;;AAGA,SAAS,aAAa,KAA+B;CACnD,MAAM,MAAM,IAAI,YAAY,KAAK,IAAI,UAAU;CAC/C,MAAM,QAAQ,IAAI,WAAW,GAAG;CAChC,MAAM,SAAS,IAAI,SAAS,GAAG;CAC/B,MAAM,UAAU,IAAI,YAAY;CAChC,MAAM,IAAI,QAAQ,OAAO,MAAM,GAAG,CAAC;CACnC,OAAO,UAAU,GAAG,KAAK,IAAI,YAAY,IAAI;CAC7C,MAAM,IAAI,QAAQ,OAAO,UAAU,GAAG,CAAC;CACvC,OAAO,UAAU,IAAI,IAAI,IAAI;CAC7B,OAAO,UAAU,IAAI,GAAG,IAAI;CAC5B,OAAO,UAAU,IAAI,GAAG,IAAI;CAC5B,OAAO,UAAU,IAAI,OAAO,IAAI;CAChC,OAAO,UAAU,IAAI,OAAW,IAAI;CACpC,OAAO,UAAU,IAAI,GAAG,IAAI;CAC5B,OAAO,UAAU,IAAI,IAAI,IAAI;CAC7B,MAAM,IAAI,QAAQ,OAAO,MAAM,GAAG,EAAE;CACpC,OAAO,UAAU,IAAI,IAAI,YAAY,IAAI;CACzC,MAAM,IAAI,IAAI,WAAW,GAAG,GAAG,EAAE;CACjC,OAAO;AACT;;;;AAKA,SAAgB,iBACd,OACA,QACiC;CACjC,OAAO,IAAI,wBAAwB,OAAO,MAAM;AAClD;;;;AAKA,SAAgB,uBACd,OACA,QACA,QACiC;CACjC,OAAO,IAAI,wBAAwB,OAAO;EAAE;EAAQ,GAAG;CAAO,CAAC;AACjE"}

package/dist/esm/adapters/voice.d.ts

@@ -0,0 +1,85 @@
+import { BaseVoiceAdapter } from '@tanstack/ai/adapters';
+import { VoiceGenerationOptions, VoiceResult } from '@tanstack/ai';
+import { ElevenLabsClientConfig } from '../utils/client.js';
+import { ElevenLabsOutputFormat, ElevenLabsVoiceModel } from '../model-meta.js';
+/**
+ * Provider-specific voice-design options. Fields map 1:1 onto the SDK's
+ * `VoiceDesignRequestModel` — mirroring the names so ElevenLabs'
+ * documentation stays useful here.
+ * @see https://elevenlabs.io/docs/api-reference/text-to-voice/design
+ */
+export interface ElevenLabsVoiceProviderOptions {
+    /** Output audio format for the previews, `codec_samplerate[_bitrate]`. */
+    outputFormat?: ElevenLabsOutputFormat;
+    /** Line the previews speak, 100..1000 characters. */
+    text?: string;
+    /** Let ElevenLabs write the preview line from the description. */
+    autoGenerateText?: boolean;
+    /** Preview loudness, -1 (quietest) to 1 (loudest). 0 is roughly -24 LUFS. */
+    loudness?: number;
+    /** Deterministic sampling seed — same seed and inputs produce the same voice. */
+    seed?: number;
+    /** How closely to follow the description. High values can sound robotic. */
+    guidanceScale?: number;
+    /** Higher quality trades variety for fidelity. */
+    quality?: number;
+    /** Let ElevenLabs expand a short description into a detailed one. */
+    shouldEnhance?: boolean;
+    /**
+     * Balance of description against reference audio, 0 (almost all reference)
+     * to 1 (almost all description). `eleven_ttv_v3` only.
+     */
+    promptStrength?: number;
+    /** Metadata stored on the voice. Only used when the voice is saved. */
+    labels?: Record<string, string>;
+    /** Remixing session to attach these generations to. */
+    remixingSessionId?: string;
+    /** Remixing session iteration to attach these generations to. */
+    remixingSessionIterationId?: string;
+}
+/**
+ * ElevenLabs voice-design adapter built on the official
+ * `@elevenlabs/elevenlabs-js` SDK.
+ *
+ * ElevenLabs designs voices in two steps — generate previews, then promote one
+ * into a real voice — but that is an implementation detail. Pass `name` to get
+ * a saved voice back; leave it off to audition previews first.
+ *
+ * @example Audition previews
+ * ```ts
+ * const result = await generateVoice({
+ *   adapter: elevenlabsVoiceDesign('eleven_ttv_v3'),
+ *   prompt: 'A warm, gravelly narrator in his sixties with a slight Irish lilt',
+ * })
+ * ```
+ *
+ * @example Save the voice
+ * ```ts
+ * const result = await generateVoice({
+ *   adapter: elevenlabsVoiceDesign('eleven_ttv_v3'),
+ *   prompt: 'A bright, upbeat product demo host',
+ *   name: 'Demo Host',
+ * })
+ * ```
+ */
+export declare class ElevenLabsVoiceAdapter<TModel extends ElevenLabsVoiceModel> extends BaseVoiceAdapter<TModel, ElevenLabsVoiceProviderOptions> {
+    readonly name: "elevenlabs";
+    private readonly client;
+    constructor(model: TModel, config?: ElevenLabsClientConfig);
+    generateVoice(options: VoiceGenerationOptions<ElevenLabsVoiceProviderOptions>): Promise<VoiceResult>;
+    /**
+     * Promote the best preview into a real library voice. The remaining
+     * generated ids go along as `playedNotSelectedVoiceIds` — ElevenLabs uses
+     * them as RLHF signal for future designs.
+     */
+    private promoteFirstPreview;
+    protected generateId(): string;
+}
+/**
+ * Create an ElevenLabs voice-design adapter using `ELEVENLABS_API_KEY` from env.
+ */
+export declare function elevenlabsVoiceDesign<TModel extends ElevenLabsVoiceModel>(model: TModel, config?: ElevenLabsClientConfig): ElevenLabsVoiceAdapter<TModel>;
+/**
+ * Create an ElevenLabs voice-design adapter with an explicit API key.
+ */
+export declare function createElevenLabsVoiceDesign<TModel extends ElevenLabsVoiceModel>(model: TModel, apiKey: string, config?: Omit<ElevenLabsClientConfig, 'apiKey'>): ElevenLabsVoiceAdapter<TModel>;