@tanstack/ai-groq 0.5.3 → 0.5.4

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

@@ -1 +1 @@
-{"version":3,"file":"transcription.js","sources":["../../../src/adapters/transcription.ts"],"sourcesContent":["import { BaseTranscriptionAdapter } from '@tanstack/ai/adapters'\nimport { base64ToArrayBuffer, generateId } from '@tanstack/ai-utils'\nimport { getGroqApiKeyFromEnv, withGroqDefaults } from '../utils/client'\nimport type {\n  TranscriptionOptions,\n  TranscriptionResult,\n  TranscriptionSegment,\n} from '@tanstack/ai'\nimport type { GroqTranscriptionModel } from '../model-meta'\nimport type { GroqTranscriptionProviderOptions } from '../audio/transcription-provider-options'\nimport type { GroqClientConfig } from '../utils/client'\n\n/**\n * Configuration for the Groq Transcription adapter.\n */\nexport interface GroqTranscriptionConfig extends GroqClientConfig {}\n\n/**\n * Flattens the `openai` SDK's `HeadersLike` config value into a plain record so\n * it can be merged into the raw `fetch` request this adapter issues. Handles\n * the shapes callers actually pass (`Headers`, an entries array, or a plain\n * object); null/undefined values are dropped.\n *\n * ponytail: doesn't unwrap the SDK's internal `NullableHeaders` class; forward\n * that shape here if the SDK ever hands it to adapter config.\n */\nfunction normalizeHeaders(\n  headers: GroqTranscriptionConfig['defaultHeaders'],\n): Record<string, string> {\n  const out: Record<string, string> = {}\n  if (!headers) return out\n  const assign = (key: string, value: unknown) => {\n    if (value != null) out[key] = String(value)\n  }\n  if (headers instanceof Headers) {\n    headers.forEach((value, key) => assign(key, value))\n  } else if (Array.isArray(headers)) {\n    for (const [key, value] of headers) assign(key, value)\n  } else {\n    for (const [key, value] of Object.entries(headers)) assign(key, value)\n  }\n  return out\n}\n\n// Shape of Groq's verbose_json transcription response\ninterface GroqVerboseTranscriptionResponse {\n  task?: string\n  language?: string\n  duration?: number\n  text: string\n  segments?: Array<{\n    id: number\n    seek?: number\n    start: number\n    end: number\n    text: string\n    tokens?: Array<number>\n    temperature?: number\n    avg_logprob: number\n    compression_ratio?: number\n    no_speech_prob?: number\n  }>\n  words?: Array<{ word: string; start: number; end: number }>\n  x_groq?: { id?: string }\n}\n\n// Shape of Groq's json transcription response\ninterface GroqJsonTranscriptionResponse {\n  text: string\n  x_groq?: { id?: string }\n}\n\n/**\n * Groq Transcription (Speech-to-Text) Adapter\n *\n * Tree-shakeable adapter for Groq audio transcription. Supports\n * whisper-large-v3 and whisper-large-v3-turbo.\n *\n * Features:\n * - Audio file uploads (File, Blob, ArrayBuffer, base64/data URL)\n * - Remote audio URLs passed directly via Groq's `url` field — no upload needed\n * - Verbose JSON response with segment and word timestamps\n * - Language detection or specification (ISO-639-1)\n * - Confidence scores derived from segment avg_logprob\n */\nexport class GroqTranscriptionAdapter<\n  TModel extends GroqTranscriptionModel,\n> extends BaseTranscriptionAdapter<TModel, GroqTranscriptionProviderOptions> {\n  readonly name = 'groq' as const\n\n  private readonly apiKey: string\n  private readonly baseURL: string\n  private readonly defaultHeaders: Record<string, string>\n\n  constructor(config: GroqTranscriptionConfig, model: TModel) {\n    super(model, {})\n    const resolved = withGroqDefaults(config)\n    this.apiKey = resolved.apiKey\n    this.baseURL = resolved.baseURL ?? 'https://api.groq.com/openai/v1'\n    this.defaultHeaders = normalizeHeaders(resolved.defaultHeaders)\n  }\n\n  async transcribe(\n    options: TranscriptionOptions<GroqTranscriptionProviderOptions>,\n  ): Promise<TranscriptionResult> {\n    const { model, audio, language, prompt, responseFormat, modelOptions } =\n      options\n\n    // Groq's transcription endpoint only accepts 'json', 'text', and\n    // 'verbose_json'. Reject 'srt'/'vtt' up front so callers get a clear\n    // message instead of an opaque Groq HTTP error.\n    if (responseFormat === 'srt' || responseFormat === 'vtt') {\n      throw new Error(\n        `Groq transcription does not support responseFormat='${responseFormat}'. ` +\n          `Supported values: 'json', 'text', 'verbose_json'.`,\n      )\n    }\n\n    // Default to verbose_json so callers get language, duration, and timestamps\n    // without having to opt in explicitly. Both Groq whisper models support it.\n    const effectiveFormat = responseFormat ?? 'verbose_json'\n    const useVerbose = effectiveFormat === 'verbose_json'\n\n    const form = new FormData()\n    form.append('model', model)\n    form.append('response_format', effectiveFormat)\n    if (language !== undefined) form.append('language', language)\n    if (prompt !== undefined) form.append('prompt', prompt)\n    if (modelOptions?.temperature !== undefined) {\n      form.append('temperature', String(modelOptions.temperature))\n    }\n    if (modelOptions?.timestamp_granularities !== undefined) {\n      for (const g of modelOptions.timestamp_granularities) {\n        form.append('timestamp_granularities[]', g)\n      }\n    }\n\n    // HTTP/HTTPS URLs are forwarded directly via Groq's `url` field, which\n    // avoids a round-trip upload. All other inputs (File, Blob, ArrayBuffer,\n    // base64, data URL) are converted to a File and sent as `file`.\n    if (typeof audio === 'string' && /^https?:\\/\\//.test(audio)) {\n      form.append('url', audio)\n    } else {\n      form.append('file', this.prepareAudioFile(audio))\n    }\n\n    try {\n      options.logger.request(\n        `activity=transcription provider=${this.name} model=${model} verbose=${useVerbose}`,\n        { provider: this.name, model },\n      )\n\n      const response = await fetch(`${this.baseURL}/audio/transcriptions`, {\n        method: 'POST',\n        headers: {\n          ...this.defaultHeaders,\n          Authorization: `Bearer ${this.apiKey}`,\n        },\n        body: form,\n      })\n\n      if (!response.ok) {\n        const body = await response\n          .json()\n          .catch(() => null as Record<string, unknown> | null)\n        const message =\n          (body?.error as { message?: string } | undefined)?.message ??\n          `Groq API error ${response.status}`\n        throw new Error(message)\n      }\n\n      if (useVerbose) {\n        const data = (await response.json()) as GroqVerboseTranscriptionResponse\n        const requestId = data.x_groq?.id ?? generateId(this.name)\n\n        // `TranscriptionResult` declares optional fields without `| undefined`,\n        // so under exactOptionalPropertyTypes we must omit absent fields rather\n        // than assigning `undefined`.\n        const segments = data.segments?.map(\n          (seg): TranscriptionSegment => ({\n            id: seg.id,\n            start: seg.start,\n            end: seg.end,\n            text: seg.text,\n            confidence: Math.exp(seg.avg_logprob),\n          }),\n        )\n        const words = data.words?.map((w) => ({\n          word: w.word,\n          start: w.start,\n          end: w.end,\n        }))\n\n        return {\n          id: requestId,\n          model,\n          text: data.text,\n          ...(data.language !== undefined && { language: data.language }),\n          ...(data.duration !== undefined && { duration: data.duration }),\n          ...(segments !== undefined && { segments }),\n          ...(words !== undefined && { words }),\n        }\n      } else if (effectiveFormat === 'text') {\n        const text = await response.text()\n        return {\n          id: generateId(this.name),\n          model,\n          text,\n          ...(language !== undefined && { language }),\n        }\n      } else {\n        const data = (await response.json()) as GroqJsonTranscriptionResponse\n        return {\n          id: data.x_groq?.id ?? generateId(this.name),\n          model,\n          text: data.text,\n          ...(language !== undefined && { language }),\n        }\n      }\n    } catch (error: unknown) {\n      options.logger.errors(`${this.name}.transcribe fatal`, {\n        error,\n        source: `${this.name}.transcribe`,\n      })\n      throw error\n    }\n  }\n\n  private prepareAudioFile(audio: string | File | Blob | ArrayBuffer): File {\n    if (typeof File !== 'undefined' && audio instanceof File) {\n      return audio\n    }\n    if (typeof Blob !== 'undefined' && audio instanceof Blob) {\n      this.ensureFileSupport()\n      return new File([audio], 'audio.mp3', {\n        type: audio.type || 'audio/mpeg',\n      })\n    }\n    if (typeof ArrayBuffer !== 'undefined' && audio instanceof ArrayBuffer) {\n      this.ensureFileSupport()\n      return new File([audio], 'audio.mp3', { type: 'audio/mpeg' })\n    }\n    if (typeof audio === 'string') {\n      this.ensureFileSupport()\n\n      if (audio.startsWith('data:')) {\n        const parts = audio.split(',')\n        const header = parts[0]\n        const base64Data = parts[1] || ''\n        const mimeMatch = header?.match(/data:([^;]+)/)\n        const mimeType = mimeMatch?.[1] || 'audio/mpeg'\n        const bytes = base64ToArrayBuffer(base64Data)\n        const extension = mimeType.split('/')[1] || 'mp3'\n        return new File([bytes], `audio.${extension}`, { type: mimeType })\n      }\n\n      const bytes = base64ToArrayBuffer(audio)\n      return new File([bytes], 'audio.mp3', { type: 'audio/mpeg' })\n    }\n\n    throw new Error('Invalid audio input type')\n  }\n\n  // Throws on Node < 20 where the global `File` constructor is unavailable.\n  private ensureFileSupport(): void {\n    if (typeof File === 'undefined') {\n      throw new Error(\n        '`File` is not available in this environment. ' +\n          'Use Node.js 20 or newer, or pass a File object directly.',\n      )\n    }\n  }\n}\n\n/**\n * Creates a Groq transcription adapter with an explicit API key.\n * Type resolution happens here at the call site.\n *\n * @param model - The model name (e.g., 'whisper-large-v3-turbo')\n * @param apiKey - Your Groq API key\n * @param config - Optional additional configuration\n * @returns Configured Groq transcription adapter instance\n *\n * @example\n * ```typescript\n * const adapter = createGroqTranscription('whisper-large-v3-turbo', 'gsk_...');\n *\n * const result = await generateTranscription({\n *   adapter,\n *   audio: audioFile,\n *   language: 'en',\n * });\n * ```\n */\nexport function createGroqTranscription<TModel extends GroqTranscriptionModel>(\n  model: TModel,\n  apiKey: string,\n  config?: Omit<GroqTranscriptionConfig, 'apiKey'>,\n): GroqTranscriptionAdapter<TModel> {\n  return new GroqTranscriptionAdapter({ apiKey, ...config }, model)\n}\n\n/**\n * Creates a Groq transcription adapter using the `GROQ_API_KEY` environment\n * variable. Type resolution happens here at the call site.\n *\n * Looks for `GROQ_API_KEY` in:\n * - `process.env` (Node.js)\n * - `window.env` (browser with injected env)\n *\n * @param model - The model name (e.g., 'whisper-large-v3-turbo')\n * @param config - Optional configuration (excluding apiKey which is auto-detected)\n * @returns Configured Groq transcription adapter instance\n * @throws Error if GROQ_API_KEY is not found in environment\n *\n * @example\n * ```typescript\n * const adapter = groqTranscription('whisper-large-v3-turbo');\n *\n * const result = await generateTranscription({\n *   adapter,\n *   audio: 'https://example.com/audio.mp3',\n * });\n *\n * console.log(result.text)\n * ```\n */\nexport function groqTranscription<TModel extends GroqTranscriptionModel>(\n  model: TModel,\n  config?: Omit<GroqTranscriptionConfig, 'apiKey'>,\n): GroqTranscriptionAdapter<TModel> {\n  const apiKey = getGroqApiKeyFromEnv()\n  return createGroqTranscription(model, apiKey, config)\n}\n"],"names":["bytes"],"mappings":";;;AA0BA,SAAS,iBACP,SACwB;AACxB,QAAM,MAA8B,CAAA;AACpC,MAAI,CAAC,QAAS,QAAO;AACrB,QAAM,SAAS,CAAC,KAAa,UAAmB;AAC9C,QAAI,SAAS,KAAM,KAAI,GAAG,IAAI,OAAO,KAAK;AAAA,EAC5C;AACA,MAAI,mBAAmB,SAAS;AAC9B,YAAQ,QAAQ,CAAC,OAAO,QAAQ,OAAO,KAAK,KAAK,CAAC;AAAA,EACpD,WAAW,MAAM,QAAQ,OAAO,GAAG;AACjC,eAAW,CAAC,KAAK,KAAK,KAAK,QAAS,QAAO,KAAK,KAAK;AAAA,EACvD,OAAO;AACL,eAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,OAAO,EAAG,QAAO,KAAK,KAAK;AAAA,EACvE;AACA,SAAO;AACT;AA2CO,MAAM,iCAEH,yBAAmE;AAAA,EAClE,OAAO;AAAA,EAEC;AAAA,EACA;AAAA,EACA;AAAA,EAEjB,YAAY,QAAiC,OAAe;AAC1D,UAAM,OAAO,EAAE;AACf,UAAM,WAAW,iBAAiB,MAAM;AACxC,SAAK,SAAS,SAAS;AACvB,SAAK,UAAU,SAAS,WAAW;AACnC,SAAK,iBAAiB,iBAAiB,SAAS,cAAc;AAAA,EAChE;AAAA,EAEA,MAAM,WACJ,SAC8B;AAC9B,UAAM,EAAE,OAAO,OAAO,UAAU,QAAQ,gBAAgB,iBACtD;AAKF,QAAI,mBAAmB,SAAS,mBAAmB,OAAO;AACxD,YAAM,IAAI;AAAA,QACR,uDAAuD,cAAc;AAAA,MAAA;AAAA,IAGzE;AAIA,UAAM,kBAAkB,kBAAkB;AAC1C,UAAM,aAAa,oBAAoB;AAEvC,UAAM,OAAO,IAAI,SAAA;AACjB,SAAK,OAAO,SAAS,KAAK;AAC1B,SAAK,OAAO,mBAAmB,eAAe;AAC9C,QAAI,aAAa,OAAW,MAAK,OAAO,YAAY,QAAQ;AAC5D,QAAI,WAAW,OAAW,MAAK,OAAO,UAAU,MAAM;AACtD,QAAI,cAAc,gBAAgB,QAAW;AAC3C,WAAK,OAAO,eAAe,OAAO,aAAa,WAAW,CAAC;AAAA,IAC7D;AACA,QAAI,cAAc,4BAA4B,QAAW;AACvD,iBAAW,KAAK,aAAa,yBAAyB;AACpD,aAAK,OAAO,6BAA6B,CAAC;AAAA,MAC5C;AAAA,IACF;AAKA,QAAI,OAAO,UAAU,YAAY,eAAe,KAAK,KAAK,GAAG;AAC3D,WAAK,OAAO,OAAO,KAAK;AAAA,IAC1B,OAAO;AACL,WAAK,OAAO,QAAQ,KAAK,iBAAiB,KAAK,CAAC;AAAA,IAClD;AAEA,QAAI;AACF,cAAQ,OAAO;AAAA,QACb,mCAAmC,KAAK,IAAI,UAAU,KAAK,YAAY,UAAU;AAAA,QACjF,EAAE,UAAU,KAAK,MAAM,MAAA;AAAA,MAAM;AAG/B,YAAM,WAAW,MAAM,MAAM,GAAG,KAAK,OAAO,yBAAyB;AAAA,QACnE,QAAQ;AAAA,QACR,SAAS;AAAA,UACP,GAAG,KAAK;AAAA,UACR,eAAe,UAAU,KAAK,MAAM;AAAA,QAAA;AAAA,QAEtC,MAAM;AAAA,MAAA,CACP;AAED,UAAI,CAAC,SAAS,IAAI;AAChB,cAAM,OAAO,MAAM,SAChB,OACA,MAAM,MAAM,IAAsC;AACrD,cAAM,UACH,MAAM,OAA4C,WACnD,kBAAkB,SAAS,MAAM;AACnC,cAAM,IAAI,MAAM,OAAO;AAAA,MACzB;AAEA,UAAI,YAAY;AACd,cAAM,OAAQ,MAAM,SAAS,KAAA;AAC7B,cAAM,YAAY,KAAK,QAAQ,MAAM,WAAW,KAAK,IAAI;AAKzD,cAAM,WAAW,KAAK,UAAU;AAAA,UAC9B,CAAC,SAA+B;AAAA,YAC9B,IAAI,IAAI;AAAA,YACR,OAAO,IAAI;AAAA,YACX,KAAK,IAAI;AAAA,YACT,MAAM,IAAI;AAAA,YACV,YAAY,KAAK,IAAI,IAAI,WAAW;AAAA,UAAA;AAAA,QACtC;AAEF,cAAM,QAAQ,KAAK,OAAO,IAAI,CAAC,OAAO;AAAA,UACpC,MAAM,EAAE;AAAA,UACR,OAAO,EAAE;AAAA,UACT,KAAK,EAAE;AAAA,QAAA,EACP;AAEF,eAAO;AAAA,UACL,IAAI;AAAA,UACJ;AAAA,UACA,MAAM,KAAK;AAAA,UACX,GAAI,KAAK,aAAa,UAAa,EAAE,UAAU,KAAK,SAAA;AAAA,UACpD,GAAI,KAAK,aAAa,UAAa,EAAE,UAAU,KAAK,SAAA;AAAA,UACpD,GAAI,aAAa,UAAa,EAAE,SAAA;AAAA,UAChC,GAAI,UAAU,UAAa,EAAE,MAAA;AAAA,QAAM;AAAA,MAEvC,WAAW,oBAAoB,QAAQ;AACrC,cAAM,OAAO,MAAM,SAAS,KAAA;AAC5B,eAAO;AAAA,UACL,IAAI,WAAW,KAAK,IAAI;AAAA,UACxB;AAAA,UACA;AAAA,UACA,GAAI,aAAa,UAAa,EAAE,SAAA;AAAA,QAAS;AAAA,MAE7C,OAAO;AACL,cAAM,OAAQ,MAAM,SAAS,KAAA;AAC7B,eAAO;AAAA,UACL,IAAI,KAAK,QAAQ,MAAM,WAAW,KAAK,IAAI;AAAA,UAC3C;AAAA,UACA,MAAM,KAAK;AAAA,UACX,GAAI,aAAa,UAAa,EAAE,SAAA;AAAA,QAAS;AAAA,MAE7C;AAAA,IACF,SAAS,OAAgB;AACvB,cAAQ,OAAO,OAAO,GAAG,KAAK,IAAI,qBAAqB;AAAA,QACrD;AAAA,QACA,QAAQ,GAAG,KAAK,IAAI;AAAA,MAAA,CACrB;AACD,YAAM;AAAA,IACR;AAAA,EACF;AAAA,EAEQ,iBAAiB,OAAiD;AACxE,QAAI,OAAO,SAAS,eAAe,iBAAiB,MAAM;AACxD,aAAO;AAAA,IACT;AACA,QAAI,OAAO,SAAS,eAAe,iBAAiB,MAAM;AACxD,WAAK,kBAAA;AACL,aAAO,IAAI,KAAK,CAAC,KAAK,GAAG,aAAa;AAAA,QACpC,MAAM,MAAM,QAAQ;AAAA,MAAA,CACrB;AAAA,IACH;AACA,QAAI,OAAO,gBAAgB,eAAe,iBAAiB,aAAa;AACtE,WAAK,kBAAA;AACL,aAAO,IAAI,KAAK,CAAC,KAAK,GAAG,aAAa,EAAE,MAAM,cAAc;AAAA,IAC9D;AACA,QAAI,OAAO,UAAU,UAAU;AAC7B,WAAK,kBAAA;AAEL,UAAI,MAAM,WAAW,OAAO,GAAG;AAC7B,cAAM,QAAQ,MAAM,MAAM,GAAG;AAC7B,cAAM,SAAS,MAAM,CAAC;AACtB,cAAM,aAAa,MAAM,CAAC,KAAK;AAC/B,cAAM,YAAY,QAAQ,MAAM,cAAc;AAC9C,cAAM,WAAW,YAAY,CAAC,KAAK;AACnC,cAAMA,SAAQ,oBAAoB,UAAU;AAC5C,cAAM,YAAY,SAAS,MAAM,GAAG,EAAE,CAAC,KAAK;AAC5C,eAAO,IAAI,KAAK,CAACA,MAAK,GAAG,SAAS,SAAS,IAAI,EAAE,MAAM,UAAU;AAAA,MACnE;AAEA,YAAM,QAAQ,oBAAoB,KAAK;AACvC,aAAO,IAAI,KAAK,CAAC,KAAK,GAAG,aAAa,EAAE,MAAM,cAAc;AAAA,IAC9D;AAEA,UAAM,IAAI,MAAM,0BAA0B;AAAA,EAC5C;AAAA;AAAA,EAGQ,oBAA0B;AAChC,QAAI,OAAO,SAAS,aAAa;AAC/B,YAAM,IAAI;AAAA,QACR;AAAA,MAAA;AAAA,IAGJ;AAAA,EACF;AACF;AAsBO,SAAS,wBACd,OACA,QACA,QACkC;AAClC,SAAO,IAAI,yBAAyB,EAAE,QAAQ,GAAG,OAAA,GAAU,KAAK;AAClE;AA2BO,SAAS,kBACd,OACA,QACkC;AAClC,QAAM,SAAS,qBAAA;AACf,SAAO,wBAAwB,OAAO,QAAQ,MAAM;AACtD;"}
+{"version":3,"file":"transcription.js","names":[],"sources":["../../../src/adapters/transcription.ts"],"sourcesContent":["import { BaseTranscriptionAdapter } from '@tanstack/ai/adapters'\nimport { base64ToArrayBuffer, generateId } from '@tanstack/ai-utils'\nimport { getGroqApiKeyFromEnv, withGroqDefaults } from '../utils/client'\nimport type {\n  TranscriptionOptions,\n  TranscriptionResult,\n  TranscriptionSegment,\n} from '@tanstack/ai'\nimport type { GroqTranscriptionModel } from '../model-meta'\nimport type { GroqTranscriptionProviderOptions } from '../audio/transcription-provider-options'\nimport type { GroqClientConfig } from '../utils/client'\n\n/**\n * Configuration for the Groq Transcription adapter.\n */\nexport interface GroqTranscriptionConfig extends GroqClientConfig {}\n\n/**\n * Flattens the `openai` SDK's `HeadersLike` config value into a plain record so\n * it can be merged into the raw `fetch` request this adapter issues. Handles\n * the shapes callers actually pass (`Headers`, an entries array, or a plain\n * object); null/undefined values are dropped.\n *\n * ponytail: doesn't unwrap the SDK's internal `NullableHeaders` class; forward\n * that shape here if the SDK ever hands it to adapter config.\n */\nfunction normalizeHeaders(\n  headers: GroqTranscriptionConfig['defaultHeaders'],\n): Record<string, string> {\n  const out: Record<string, string> = {}\n  if (!headers) return out\n  const assign = (key: string, value: unknown) => {\n    if (value != null) out[key] = String(value)\n  }\n  if (headers instanceof Headers) {\n    headers.forEach((value, key) => assign(key, value))\n  } else if (Array.isArray(headers)) {\n    for (const [key, value] of headers) assign(key, value)\n  } else {\n    for (const [key, value] of Object.entries(headers)) assign(key, value)\n  }\n  return out\n}\n\n// Shape of Groq's verbose_json transcription response\ninterface GroqVerboseTranscriptionResponse {\n  task?: string\n  language?: string\n  duration?: number\n  text: string\n  segments?: Array<{\n    id: number\n    seek?: number\n    start: number\n    end: number\n    text: string\n    tokens?: Array<number>\n    temperature?: number\n    avg_logprob: number\n    compression_ratio?: number\n    no_speech_prob?: number\n  }>\n  words?: Array<{ word: string; start: number; end: number }>\n  x_groq?: { id?: string }\n}\n\n// Shape of Groq's json transcription response\ninterface GroqJsonTranscriptionResponse {\n  text: string\n  x_groq?: { id?: string }\n}\n\n/**\n * Groq Transcription (Speech-to-Text) Adapter\n *\n * Tree-shakeable adapter for Groq audio transcription. Supports\n * whisper-large-v3 and whisper-large-v3-turbo.\n *\n * Features:\n * - Audio file uploads (File, Blob, ArrayBuffer, base64/data URL)\n * - Remote audio URLs passed directly via Groq's `url` field — no upload needed\n * - Verbose JSON response with segment and word timestamps\n * - Language detection or specification (ISO-639-1)\n * - Confidence scores derived from segment avg_logprob\n */\nexport class GroqTranscriptionAdapter<\n  TModel extends GroqTranscriptionModel,\n> extends BaseTranscriptionAdapter<TModel, GroqTranscriptionProviderOptions> {\n  readonly name = 'groq' as const\n\n  private readonly apiKey: string\n  private readonly baseURL: string\n  private readonly defaultHeaders: Record<string, string>\n\n  constructor(config: GroqTranscriptionConfig, model: TModel) {\n    super(model, {})\n    const resolved = withGroqDefaults(config)\n    this.apiKey = resolved.apiKey\n    this.baseURL = resolved.baseURL ?? 'https://api.groq.com/openai/v1'\n    this.defaultHeaders = normalizeHeaders(resolved.defaultHeaders)\n  }\n\n  async transcribe(\n    options: TranscriptionOptions<GroqTranscriptionProviderOptions>,\n  ): Promise<TranscriptionResult> {\n    const { model, audio, language, prompt, responseFormat, modelOptions } =\n      options\n\n    // Groq's transcription endpoint only accepts 'json', 'text', and\n    // 'verbose_json'. Reject 'srt'/'vtt' up front so callers get a clear\n    // message instead of an opaque Groq HTTP error.\n    if (responseFormat === 'srt' || responseFormat === 'vtt') {\n      throw new Error(\n        `Groq transcription does not support responseFormat='${responseFormat}'. ` +\n          `Supported values: 'json', 'text', 'verbose_json'.`,\n      )\n    }\n\n    // Default to verbose_json so callers get language, duration, and timestamps\n    // without having to opt in explicitly. Both Groq whisper models support it.\n    const effectiveFormat = responseFormat ?? 'verbose_json'\n    const useVerbose = effectiveFormat === 'verbose_json'\n\n    const form = new FormData()\n    form.append('model', model)\n    form.append('response_format', effectiveFormat)\n    if (language !== undefined) form.append('language', language)\n    if (prompt !== undefined) form.append('prompt', prompt)\n    if (modelOptions?.temperature !== undefined) {\n      form.append('temperature', String(modelOptions.temperature))\n    }\n    if (modelOptions?.timestamp_granularities !== undefined) {\n      for (const g of modelOptions.timestamp_granularities) {\n        form.append('timestamp_granularities[]', g)\n      }\n    }\n\n    // HTTP/HTTPS URLs are forwarded directly via Groq's `url` field, which\n    // avoids a round-trip upload. All other inputs (File, Blob, ArrayBuffer,\n    // base64, data URL) are converted to a File and sent as `file`.\n    if (typeof audio === 'string' && /^https?:\\/\\//.test(audio)) {\n      form.append('url', audio)\n    } else {\n      form.append('file', this.prepareAudioFile(audio))\n    }\n\n    try {\n      options.logger.request(\n        `activity=transcription provider=${this.name} model=${model} verbose=${useVerbose}`,\n        { provider: this.name, model },\n      )\n\n      const response = await fetch(`${this.baseURL}/audio/transcriptions`, {\n        method: 'POST',\n        headers: {\n          ...this.defaultHeaders,\n          Authorization: `Bearer ${this.apiKey}`,\n        },\n        body: form,\n      })\n\n      if (!response.ok) {\n        const body = await response\n          .json()\n          .catch(() => null as Record<string, unknown> | null)\n        const message =\n          (body?.error as { message?: string } | undefined)?.message ??\n          `Groq API error ${response.status}`\n        throw new Error(message)\n      }\n\n      if (useVerbose) {\n        const data = (await response.json()) as GroqVerboseTranscriptionResponse\n        const requestId = data.x_groq?.id ?? generateId(this.name)\n\n        // `TranscriptionResult` declares optional fields without `| undefined`,\n        // so under exactOptionalPropertyTypes we must omit absent fields rather\n        // than assigning `undefined`.\n        const segments = data.segments?.map(\n          (seg): TranscriptionSegment => ({\n            id: seg.id,\n            start: seg.start,\n            end: seg.end,\n            text: seg.text,\n            confidence: Math.exp(seg.avg_logprob),\n          }),\n        )\n        const words = data.words?.map((w) => ({\n          word: w.word,\n          start: w.start,\n          end: w.end,\n        }))\n\n        return {\n          id: requestId,\n          model,\n          text: data.text,\n          ...(data.language !== undefined && { language: data.language }),\n          ...(data.duration !== undefined && { duration: data.duration }),\n          ...(segments !== undefined && { segments }),\n          ...(words !== undefined && { words }),\n        }\n      } else if (effectiveFormat === 'text') {\n        const text = await response.text()\n        return {\n          id: generateId(this.name),\n          model,\n          text,\n          ...(language !== undefined && { language }),\n        }\n      } else {\n        const data = (await response.json()) as GroqJsonTranscriptionResponse\n        return {\n          id: data.x_groq?.id ?? generateId(this.name),\n          model,\n          text: data.text,\n          ...(language !== undefined && { language }),\n        }\n      }\n    } catch (error: unknown) {\n      options.logger.errors(`${this.name}.transcribe fatal`, {\n        error,\n        source: `${this.name}.transcribe`,\n      })\n      throw error\n    }\n  }\n\n  private prepareAudioFile(audio: string | File | Blob | ArrayBuffer): File {\n    if (typeof File !== 'undefined' && audio instanceof File) {\n      return audio\n    }\n    if (typeof Blob !== 'undefined' && audio instanceof Blob) {\n      this.ensureFileSupport()\n      return new File([audio], 'audio.mp3', {\n        type: audio.type || 'audio/mpeg',\n      })\n    }\n    if (typeof ArrayBuffer !== 'undefined' && audio instanceof ArrayBuffer) {\n      this.ensureFileSupport()\n      return new File([audio], 'audio.mp3', { type: 'audio/mpeg' })\n    }\n    if (typeof audio === 'string') {\n      this.ensureFileSupport()\n\n      if (audio.startsWith('data:')) {\n        const parts = audio.split(',')\n        const header = parts[0]\n        const base64Data = parts[1] || ''\n        const mimeMatch = header?.match(/data:([^;]+)/)\n        const mimeType = mimeMatch?.[1] || 'audio/mpeg'\n        const bytes = base64ToArrayBuffer(base64Data)\n        const extension = mimeType.split('/')[1] || 'mp3'\n        return new File([bytes], `audio.${extension}`, { type: mimeType })\n      }\n\n      const bytes = base64ToArrayBuffer(audio)\n      return new File([bytes], 'audio.mp3', { type: 'audio/mpeg' })\n    }\n\n    throw new Error('Invalid audio input type')\n  }\n\n  // Throws on Node < 20 where the global `File` constructor is unavailable.\n  private ensureFileSupport(): void {\n    if (typeof File === 'undefined') {\n      throw new Error(\n        '`File` is not available in this environment. ' +\n          'Use Node.js 20 or newer, or pass a File object directly.',\n      )\n    }\n  }\n}\n\n/**\n * Creates a Groq transcription adapter with an explicit API key.\n * Type resolution happens here at the call site.\n *\n * @param model - The model name (e.g., 'whisper-large-v3-turbo')\n * @param apiKey - Your Groq API key\n * @param config - Optional additional configuration\n * @returns Configured Groq transcription adapter instance\n *\n * @example\n * ```typescript\n * const adapter = createGroqTranscription('whisper-large-v3-turbo', 'gsk_...');\n *\n * const result = await generateTranscription({\n *   adapter,\n *   audio: audioFile,\n *   language: 'en',\n * });\n * ```\n */\nexport function createGroqTranscription<TModel extends GroqTranscriptionModel>(\n  model: TModel,\n  apiKey: string,\n  config?: Omit<GroqTranscriptionConfig, 'apiKey'>,\n): GroqTranscriptionAdapter<TModel> {\n  return new GroqTranscriptionAdapter({ apiKey, ...config }, model)\n}\n\n/**\n * Creates a Groq transcription adapter using the `GROQ_API_KEY` environment\n * variable. Type resolution happens here at the call site.\n *\n * Looks for `GROQ_API_KEY` in:\n * - `process.env` (Node.js)\n * - `window.env` (browser with injected env)\n *\n * @param model - The model name (e.g., 'whisper-large-v3-turbo')\n * @param config - Optional configuration (excluding apiKey which is auto-detected)\n * @returns Configured Groq transcription adapter instance\n * @throws Error if GROQ_API_KEY is not found in environment\n *\n * @example\n * ```typescript\n * const adapter = groqTranscription('whisper-large-v3-turbo');\n *\n * const result = await generateTranscription({\n *   adapter,\n *   audio: 'https://example.com/audio.mp3',\n * });\n *\n * console.log(result.text)\n * ```\n */\nexport function groqTranscription<TModel extends GroqTranscriptionModel>(\n  model: TModel,\n  config?: Omit<GroqTranscriptionConfig, 'apiKey'>,\n): GroqTranscriptionAdapter<TModel> {\n  const apiKey = getGroqApiKeyFromEnv()\n  return createGroqTranscription(model, apiKey, config)\n}\n"],"mappings":";;;;;;;;;;;;;AA0BA,SAAS,iBACP,SACwB;CACxB,MAAM,MAA8B,CAAC;CACrC,IAAI,CAAC,SAAS,OAAO;CACrB,MAAM,UAAU,KAAa,UAAmB;EAC9C,IAAI,SAAS,MAAM,IAAI,OAAO,OAAO,KAAK;CAC5C;CACA,IAAI,mBAAmB,SACrB,QAAQ,SAAS,OAAO,QAAQ,OAAO,KAAK,KAAK,CAAC;MAC7C,IAAI,MAAM,QAAQ,OAAO,GAC9B,KAAK,MAAM,CAAC,KAAK,UAAU,SAAS,OAAO,KAAK,KAAK;MAErD,KAAK,MAAM,CAAC,KAAK,UAAU,OAAO,QAAQ,OAAO,GAAG,OAAO,KAAK,KAAK;CAEvE,OAAO;AACT;;;;;;;;;;;;;;AA2CA,IAAa,2BAAb,cAEU,yBAAmE;CAC3E,OAAgB;CAEhB;CACA;CACA;CAEA,YAAY,QAAiC,OAAe;EAC1D,MAAM,OAAO,CAAC,CAAC;EACf,MAAM,WAAW,iBAAiB,MAAM;EACxC,KAAK,SAAS,SAAS;EACvB,KAAK,UAAU,SAAS,WAAW;EACnC,KAAK,iBAAiB,iBAAiB,SAAS,cAAc;CAChE;CAEA,MAAM,WACJ,SAC8B;EAC9B,MAAM,EAAE,OAAO,OAAO,UAAU,QAAQ,gBAAgB,iBACtD;EAKF,IAAI,mBAAmB,SAAS,mBAAmB,OACjD,MAAM,IAAI,MACR,uDAAuD,eAAe,qDAExE;EAKF,MAAM,kBAAkB,kBAAkB;EAC1C,MAAM,aAAa,oBAAoB;EAEvC,MAAM,OAAO,IAAI,SAAS;EAC1B,KAAK,OAAO,SAAS,KAAK;EAC1B,KAAK,OAAO,mBAAmB,eAAe;EAC9C,IAAI,aAAa,KAAA,GAAW,KAAK,OAAO,YAAY,QAAQ;EAC5D,IAAI,WAAW,KAAA,GAAW,KAAK,OAAO,UAAU,MAAM;EACtD,IAAI,cAAc,gBAAgB,KAAA,GAChC,KAAK,OAAO,eAAe,OAAO,aAAa,WAAW,CAAC;EAE7D,IAAI,cAAc,4BAA4B,KAAA,GAC5C,KAAK,MAAM,KAAK,aAAa,yBAC3B,KAAK,OAAO,6BAA6B,CAAC;EAO9C,IAAI,OAAO,UAAU,YAAY,eAAe,KAAK,KAAK,GACxD,KAAK,OAAO,OAAO,KAAK;OAExB,KAAK,OAAO,QAAQ,KAAK,iBAAiB,KAAK,CAAC;EAGlD,IAAI;GACF,QAAQ,OAAO,QACb,mCAAmC,KAAK,KAAK,SAAS,MAAM,WAAW,cACvE;IAAE,UAAU,KAAK;IAAM;GAAM,CAC/B;GAEA,MAAM,WAAW,MAAM,MAAM,GAAG,KAAK,QAAQ,wBAAwB;IACnE,QAAQ;IACR,SAAS;KACP,GAAG,KAAK;KACR,eAAe,UAAU,KAAK;IAChC;IACA,MAAM;GACR,CAAC;GAED,IAAI,CAAC,SAAS,IAAI;IAIhB,MAAM,YACH,MAJgB,SAChB,KAAK,CAAC,CACN,YAAY,IAAsC,EAAA,EAE5C,MAAA,EAA4C,WACnD,kBAAkB,SAAS;IAC7B,MAAM,IAAI,MAAM,OAAO;GACzB;GAEA,IAAI,YAAY;IACd,MAAM,OAAQ,MAAM,SAAS,KAAK;IAClC,MAAM,YAAY,KAAK,QAAQ,MAAM,WAAW,KAAK,IAAI;IAKzD,MAAM,WAAW,KAAK,UAAU,KAC7B,SAA+B;KAC9B,IAAI,IAAI;KACR,OAAO,IAAI;KACX,KAAK,IAAI;KACT,MAAM,IAAI;KACV,YAAY,KAAK,IAAI,IAAI,WAAW;IACtC,EACF;IACA,MAAM,QAAQ,KAAK,OAAO,KAAK,OAAO;KACpC,MAAM,EAAE;KACR,OAAO,EAAE;KACT,KAAK,EAAE;IACT,EAAE;IAEF,OAAO;KACL,IAAI;KACJ;KACA,MAAM,KAAK;KACX,GAAI,KAAK,aAAa,KAAA,KAAa,EAAE,UAAU,KAAK,SAAS;KAC7D,GAAI,KAAK,aAAa,KAAA,KAAa,EAAE,UAAU,KAAK,SAAS;KAC7D,GAAI,aAAa,KAAA,KAAa,EAAE,SAAS;KACzC,GAAI,UAAU,KAAA,KAAa,EAAE,MAAM;IACrC;GACF,OAAO,IAAI,oBAAoB,QAAQ;IACrC,MAAM,OAAO,MAAM,SAAS,KAAK;IACjC,OAAO;KACL,IAAI,WAAW,KAAK,IAAI;KACxB;KACA;KACA,GAAI,aAAa,KAAA,KAAa,EAAE,SAAS;IAC3C;GACF,OAAO;IACL,MAAM,OAAQ,MAAM,SAAS,KAAK;IAClC,OAAO;KACL,IAAI,KAAK,QAAQ,MAAM,WAAW,KAAK,IAAI;KAC3C;KACA,MAAM,KAAK;KACX,GAAI,aAAa,KAAA,KAAa,EAAE,SAAS;IAC3C;GACF;EACF,SAAS,OAAgB;GACvB,QAAQ,OAAO,OAAO,GAAG,KAAK,KAAK,oBAAoB;IACrD;IACA,QAAQ,GAAG,KAAK,KAAK;GACvB,CAAC;GACD,MAAM;EACR;CACF;CAEA,iBAAyB,OAAiD;EACxE,IAAI,OAAO,SAAS,eAAe,iBAAiB,MAClD,OAAO;EAET,IAAI,OAAO,SAAS,eAAe,iBAAiB,MAAM;GACxD,KAAK,kBAAkB;GACvB,OAAO,IAAI,KAAK,CAAC,KAAK,GAAG,aAAa,EACpC,MAAM,MAAM,QAAQ,aACtB,CAAC;EACH;EACA,IAAI,OAAO,gBAAgB,eAAe,iBAAiB,aAAa;GACtE,KAAK,kBAAkB;GACvB,OAAO,IAAI,KAAK,CAAC,KAAK,GAAG,aAAa,EAAE,MAAM,aAAa,CAAC;EAC9D;EACA,IAAI,OAAO,UAAU,UAAU;GAC7B,KAAK,kBAAkB;GAEvB,IAAI,MAAM,WAAW,OAAO,GAAG;IAC7B,MAAM,QAAQ,MAAM,MAAM,GAAG;IAC7B,MAAM,SAAS,MAAM;IACrB,MAAM,aAAa,MAAM,MAAM;IAE/B,MAAM,YADY,QAAQ,MAAM,cAAc,EAAA,GACjB,MAAM;IACnC,MAAM,QAAQ,oBAAoB,UAAU;IAC5C,MAAM,YAAY,SAAS,MAAM,GAAG,CAAC,CAAC,MAAM;IAC5C,OAAO,IAAI,KAAK,CAAC,KAAK,GAAG,SAAS,aAAa,EAAE,MAAM,SAAS,CAAC;GACnE;GAEA,MAAM,QAAQ,oBAAoB,KAAK;GACvC,OAAO,IAAI,KAAK,CAAC,KAAK,GAAG,aAAa,EAAE,MAAM,aAAa,CAAC;EAC9D;EAEA,MAAM,IAAI,MAAM,0BAA0B;CAC5C;CAGA,oBAAkC;EAChC,IAAI,OAAO,SAAS,aAClB,MAAM,IAAI,MACR,uGAEF;CAEJ;AACF;;;;;;;;;;;;;;;;;;;;;AAsBA,SAAgB,wBACd,OACA,QACA,QACkC;CAClC,OAAO,IAAI,yBAAyB;EAAE;EAAQ,GAAG;CAAO,GAAG,KAAK;AAClE;;;;;;;;;;;;;;;;;;;;;;;;;;AA2BA,SAAgB,kBACd,OACA,QACkC;CAElC,OAAO,wBAAwB,OADhB,qBACuB,GAAQ,MAAM;AACtD"}

package/dist/esm/adapters/tts.js

@@ -1,73 +1,135 @@
+import { getGroqApiKeyFromEnv, withGroqDefaults } from "../utils/client.js";
+import { validateAudioInput } from "../audio/audio-provider-options.js";
 import OpenAI from "openai";
+import { arrayBufferToBase64, generateId } from "@tanstack/ai-utils";
 import { BaseTTSAdapter } from "@tanstack/ai/adapters";
 import { toRunErrorPayload } from "@tanstack/ai/adapter-internals";
-import { arrayBufferToBase64, generateId } from "@tanstack/ai-utils";
-import { withGroqDefaults, getGroqApiKeyFromEnv } from "../utils/client.js";
-import { validateAudioInput } from "../audio/audio-provider-options.js";
-class GroqTTSAdapter extends BaseTTSAdapter {
-  name = "groq";
-  client;
-  constructor(config, model) {
-    super(model, {});
-    this.client = new OpenAI(withGroqDefaults(config));
-  }
-  async generateSpeech(options) {
-    const { model, text, voice, format, speed, modelOptions } = options;
-    validateAudioInput({ input: text, model: this.model });
-    const request = {
-      model,
-      input: text,
-      voice: voice ?? "autumn",
-      response_format: format ?? "wav",
-      ...speed !== void 0 && { speed },
-      ...modelOptions ?? {}
-    };
-    try {
-      options.logger.request(
-        `activity=tts provider=${this.name} model=${model} format=${request.response_format ?? "default"} voice=${request.voice}`,
-        { provider: this.name, model }
-      );
-      const response = await this.client.audio.speech.create(request);
-      const arrayBuffer = await response.arrayBuffer();
-      const base64 = arrayBufferToBase64(arrayBuffer);
-      const outputFormat = request.response_format ?? "wav";
-      const contentType = this.getContentType(outputFormat);
-      return {
-        id: generateId(this.name),
-        model,
-        audio: base64,
-        format: outputFormat,
-        contentType
-      };
-    } catch (error) {
-      options.logger.errors(`${this.name}.generateSpeech fatal`, {
-        error: toRunErrorPayload(error, `${this.name}.generateSpeech failed`),
-        source: `${this.name}.generateSpeech`
-      });
-      throw error;
-    }
-  }
-  getContentType(format) {
-    const contentTypes = {
-      flac: "audio/flac",
-      mp3: "audio/mpeg",
-      mulaw: "audio/basic",
-      ogg: "audio/ogg",
-      wav: "audio/wav"
-    };
-    return contentTypes[format] || "audio/wav";
-  }
-}
+//#region src/adapters/tts.ts
+/**
+* Groq Text-to-Speech Adapter
+*
+* Tree-shakeable adapter for Groq TTS functionality. Groq exposes an
+* OpenAI-compatible `/audio/speech` endpoint, so the adapter drives it with
+* the OpenAI SDK via a `baseURL` override (the same pattern as the Groq text
+* adapter).
+*
+* Supports `canopylabs/orpheus-v1-english` and
+* `canopylabs/orpheus-arabic-saudi`.
+*
+* Features:
+* - English voices: autumn(f), diana(f), hannah(f), austin(m), daniel(m), troy(m)
+* - Arabic voices: fahad(m), sultan(m), lulwa(f), noura(f)
+* - Output formats: flac, mp3, mulaw, ogg, wav (default wav)
+* - Speed control
+* - Configurable sample rate via `modelOptions`
+*/
+var GroqTTSAdapter = class extends BaseTTSAdapter {
+	name = "groq";
+	client;
+	constructor(config, model) {
+		super(model, {});
+		this.client = new OpenAI(withGroqDefaults(config));
+	}
+	async generateSpeech(options) {
+		const { model, text, voice, format, speed, modelOptions } = options;
+		validateAudioInput({
+			input: text,
+			model: this.model
+		});
+		const request = {
+			model,
+			input: text,
+			voice: voice ?? "autumn",
+			response_format: format ?? "wav",
+			...speed !== void 0 && { speed },
+			...modelOptions ?? {}
+		};
+		try {
+			options.logger.request(`activity=tts provider=${this.name} model=${model} format=${request.response_format ?? "default"} voice=${request.voice}`, {
+				provider: this.name,
+				model
+			});
+			const base64 = arrayBufferToBase64(await (await this.client.audio.speech.create(request)).arrayBuffer());
+			const outputFormat = request.response_format ?? "wav";
+			const contentType = this.getContentType(outputFormat);
+			return {
+				id: generateId(this.name),
+				model,
+				audio: base64,
+				format: outputFormat,
+				contentType
+			};
+		} catch (error) {
+			options.logger.errors(`${this.name}.generateSpeech fatal`, {
+				error: toRunErrorPayload(error, `${this.name}.generateSpeech failed`),
+				source: `${this.name}.generateSpeech`
+			});
+			throw error;
+		}
+	}
+	getContentType(format) {
+		return {
+			flac: "audio/flac",
+			mp3: "audio/mpeg",
+			mulaw: "audio/basic",
+			ogg: "audio/ogg",
+			wav: "audio/wav"
+		}[format] || "audio/wav";
+	}
+};
+/**
+* Creates a Groq speech adapter with explicit API key.
+* Type resolution happens here at the call site.
+*
+* @param model - The model name (e.g., 'canopylabs/orpheus-v1-english')
+* @param apiKey - Your Groq API key
+* @param config - Optional additional configuration
+* @returns Configured Groq speech adapter instance with resolved types
+*
+* @example
+* ```typescript
+* const adapter = createGroqSpeech('canopylabs/orpheus-v1-english', 'gsk_...')
+*
+* const result = await generateSpeech({
+*   adapter,
+*   text: 'Hello, world!',
+*   voice: 'autumn',
+* })
+* ```
+*/
 function createGroqSpeech(model, apiKey, config) {
-  return new GroqTTSAdapter({ apiKey, ...config }, model);
+	return new GroqTTSAdapter({
+		apiKey,
+		...config
+	}, model);
 }
+/**
+* Creates a Groq speech adapter with automatic API key detection from
+* environment variables.
+*
+* Looks for `GROQ_API_KEY` in the environment.
+*
+* @param model - The model name (e.g., 'canopylabs/orpheus-v1-english')
+* @param config - Optional configuration (excluding apiKey which is auto-detected)
+* @returns Configured Groq speech adapter instance with resolved types
+* @throws Error if GROQ_API_KEY is not found in environment
+*
+* @example
+* ```typescript
+* const adapter = groqSpeech('canopylabs/orpheus-v1-english')
+*
+* const result = await generateSpeech({
+*   adapter,
+*   text: 'Welcome to TanStack AI!',
+*   voice: 'autumn',
+*   format: 'wav',
+* })
+* ```
+*/
 function groqSpeech(model, config) {
-  const apiKey = getGroqApiKeyFromEnv();
-  return createGroqSpeech(model, apiKey, config);
+	return createGroqSpeech(model, getGroqApiKeyFromEnv(), config);
 }
-export {
-  GroqTTSAdapter,
-  createGroqSpeech,
-  groqSpeech
-};
-//# sourceMappingURL=tts.js.map
+//#endregion
+export { GroqTTSAdapter, createGroqSpeech, groqSpeech };
+
+//# sourceMappingURL=tts.js.map

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

@@ -1 +1 @@
-{"version":3,"file":"tts.js","sources":["../../../src/adapters/tts.ts"],"sourcesContent":["import OpenAI from 'openai'\nimport { BaseTTSAdapter } from '@tanstack/ai/adapters'\nimport { toRunErrorPayload } from '@tanstack/ai/adapter-internals'\nimport { arrayBufferToBase64, generateId } from '@tanstack/ai-utils'\nimport { getGroqApiKeyFromEnv, withGroqDefaults } from '../utils/client'\nimport { validateAudioInput } from '../audio/audio-provider-options'\nimport type { TTSOptions, TTSResult } from '@tanstack/ai'\nimport type OpenAI_SDK from 'openai'\nimport type { GroqTTSModel } from '../model-meta'\nimport type { GroqTTSProviderOptions } from '../audio/tts-provider-options'\nimport type { GroqClientConfig } from '../utils/client'\n\n/**\n * Configuration for Groq TTS adapter\n */\nexport interface GroqTTSConfig extends GroqClientConfig {}\n\n/**\n * Groq Text-to-Speech Adapter\n *\n * Tree-shakeable adapter for Groq TTS functionality. Groq exposes an\n * OpenAI-compatible `/audio/speech` endpoint, so the adapter drives it with\n * the OpenAI SDK via a `baseURL` override (the same pattern as the Groq text\n * adapter).\n *\n * Supports `canopylabs/orpheus-v1-english` and\n * `canopylabs/orpheus-arabic-saudi`.\n *\n * Features:\n * - English voices: autumn(f), diana(f), hannah(f), austin(m), daniel(m), troy(m)\n * - Arabic voices: fahad(m), sultan(m), lulwa(f), noura(f)\n * - Output formats: flac, mp3, mulaw, ogg, wav (default wav)\n * - Speed control\n * - Configurable sample rate via `modelOptions`\n */\nexport class GroqTTSAdapter<TModel extends GroqTTSModel> extends BaseTTSAdapter<\n  TModel,\n  GroqTTSProviderOptions\n> {\n  readonly name = 'groq' as const\n\n  protected client: OpenAI\n\n  constructor(config: GroqTTSConfig, model: TModel) {\n    super(model, {})\n    this.client = new OpenAI(withGroqDefaults(config))\n  }\n\n  async generateSpeech(\n    options: TTSOptions<GroqTTSProviderOptions>,\n  ): Promise<TTSResult> {\n    const { model, text, voice, format, speed, modelOptions } = options\n\n    validateAudioInput({ input: text, model: this.model })\n\n    // Spreading optional inputs conditionally keeps the request compatible\n    // with the vendor SDK shape under exactOptionalPropertyTypes. `sample_rate`\n    // is a Groq-only body field carried via modelOptions.\n    const request: OpenAI_SDK.Audio.SpeechCreateParams = {\n      model,\n      input: text,\n      voice: voice ?? 'autumn',\n      response_format: format ?? 'wav',\n      ...(speed !== undefined && { speed }),\n      ...(modelOptions ?? {}),\n    }\n\n    try {\n      options.logger.request(\n        `activity=tts provider=${this.name} model=${model} format=${request.response_format ?? 'default'} voice=${request.voice}`,\n        { provider: this.name, model },\n      )\n      const response = await this.client.audio.speech.create(request)\n\n      const arrayBuffer = await response.arrayBuffer()\n      const base64 = arrayBufferToBase64(arrayBuffer)\n\n      const outputFormat = request.response_format ?? 'wav'\n      const contentType = this.getContentType(outputFormat)\n\n      return {\n        id: generateId(this.name),\n        model,\n        audio: base64,\n        format: outputFormat,\n        contentType,\n      }\n    } catch (error: unknown) {\n      // Narrow before logging: raw SDK errors can carry request metadata\n      // (including auth headers) which we must never surface to user loggers.\n      options.logger.errors(`${this.name}.generateSpeech fatal`, {\n        error: toRunErrorPayload(error, `${this.name}.generateSpeech failed`),\n        source: `${this.name}.generateSpeech`,\n      })\n      throw error\n    }\n  }\n\n  private getContentType(format: string): string {\n    const contentTypes: Record<string, string> = {\n      flac: 'audio/flac',\n      mp3: 'audio/mpeg',\n      mulaw: 'audio/basic',\n      ogg: 'audio/ogg',\n      wav: 'audio/wav',\n    }\n    return contentTypes[format] || 'audio/wav'\n  }\n}\n\n/**\n * Creates a Groq speech adapter with explicit API key.\n * Type resolution happens here at the call site.\n *\n * @param model - The model name (e.g., 'canopylabs/orpheus-v1-english')\n * @param apiKey - Your Groq API key\n * @param config - Optional additional configuration\n * @returns Configured Groq speech adapter instance with resolved types\n *\n * @example\n * ```typescript\n * const adapter = createGroqSpeech('canopylabs/orpheus-v1-english', 'gsk_...')\n *\n * const result = await generateSpeech({\n *   adapter,\n *   text: 'Hello, world!',\n *   voice: 'autumn',\n * })\n * ```\n */\nexport function createGroqSpeech<TModel extends GroqTTSModel>(\n  model: TModel,\n  apiKey: string,\n  config?: Omit<GroqTTSConfig, 'apiKey'>,\n): GroqTTSAdapter<TModel> {\n  return new GroqTTSAdapter({ apiKey, ...config }, model)\n}\n\n/**\n * Creates a Groq speech adapter with automatic API key detection from\n * environment variables.\n *\n * Looks for `GROQ_API_KEY` in the environment.\n *\n * @param model - The model name (e.g., 'canopylabs/orpheus-v1-english')\n * @param config - Optional configuration (excluding apiKey which is auto-detected)\n * @returns Configured Groq speech adapter instance with resolved types\n * @throws Error if GROQ_API_KEY is not found in environment\n *\n * @example\n * ```typescript\n * const adapter = groqSpeech('canopylabs/orpheus-v1-english')\n *\n * const result = await generateSpeech({\n *   adapter,\n *   text: 'Welcome to TanStack AI!',\n *   voice: 'autumn',\n *   format: 'wav',\n * })\n * ```\n */\nexport function groqSpeech<TModel extends GroqTTSModel>(\n  model: TModel,\n  config?: Omit<GroqTTSConfig, 'apiKey'>,\n): GroqTTSAdapter<TModel> {\n  const apiKey = getGroqApiKeyFromEnv()\n  return createGroqSpeech(model, apiKey, config)\n}\n"],"names":[],"mappings":";;;;;;AAmCO,MAAM,uBAAoD,eAG/D;AAAA,EACS,OAAO;AAAA,EAEN;AAAA,EAEV,YAAY,QAAuB,OAAe;AAChD,UAAM,OAAO,EAAE;AACf,SAAK,SAAS,IAAI,OAAO,iBAAiB,MAAM,CAAC;AAAA,EACnD;AAAA,EAEA,MAAM,eACJ,SACoB;AACpB,UAAM,EAAE,OAAO,MAAM,OAAO,QAAQ,OAAO,iBAAiB;AAE5D,uBAAmB,EAAE,OAAO,MAAM,OAAO,KAAK,OAAO;AAKrD,UAAM,UAA+C;AAAA,MACnD;AAAA,MACA,OAAO;AAAA,MACP,OAAO,SAAS;AAAA,MAChB,iBAAiB,UAAU;AAAA,MAC3B,GAAI,UAAU,UAAa,EAAE,MAAA;AAAA,MAC7B,GAAI,gBAAgB,CAAA;AAAA,IAAC;AAGvB,QAAI;AACF,cAAQ,OAAO;AAAA,QACb,yBAAyB,KAAK,IAAI,UAAU,KAAK,WAAW,QAAQ,mBAAmB,SAAS,UAAU,QAAQ,KAAK;AAAA,QACvH,EAAE,UAAU,KAAK,MAAM,MAAA;AAAA,MAAM;AAE/B,YAAM,WAAW,MAAM,KAAK,OAAO,MAAM,OAAO,OAAO,OAAO;AAE9D,YAAM,cAAc,MAAM,SAAS,YAAA;AACnC,YAAM,SAAS,oBAAoB,WAAW;AAE9C,YAAM,eAAe,QAAQ,mBAAmB;AAChD,YAAM,cAAc,KAAK,eAAe,YAAY;AAEpD,aAAO;AAAA,QACL,IAAI,WAAW,KAAK,IAAI;AAAA,QACxB;AAAA,QACA,OAAO;AAAA,QACP,QAAQ;AAAA,QACR;AAAA,MAAA;AAAA,IAEJ,SAAS,OAAgB;AAGvB,cAAQ,OAAO,OAAO,GAAG,KAAK,IAAI,yBAAyB;AAAA,QACzD,OAAO,kBAAkB,OAAO,GAAG,KAAK,IAAI,wBAAwB;AAAA,QACpE,QAAQ,GAAG,KAAK,IAAI;AAAA,MAAA,CACrB;AACD,YAAM;AAAA,IACR;AAAA,EACF;AAAA,EAEQ,eAAe,QAAwB;AAC7C,UAAM,eAAuC;AAAA,MAC3C,MAAM;AAAA,MACN,KAAK;AAAA,MACL,OAAO;AAAA,MACP,KAAK;AAAA,MACL,KAAK;AAAA,IAAA;AAEP,WAAO,aAAa,MAAM,KAAK;AAAA,EACjC;AACF;AAsBO,SAAS,iBACd,OACA,QACA,QACwB;AACxB,SAAO,IAAI,eAAe,EAAE,QAAQ,GAAG,OAAA,GAAU,KAAK;AACxD;AAyBO,SAAS,WACd,OACA,QACwB;AACxB,QAAM,SAAS,qBAAA;AACf,SAAO,iBAAiB,OAAO,QAAQ,MAAM;AAC/C;"}
+{"version":3,"file":"tts.js","names":[],"sources":["../../../src/adapters/tts.ts"],"sourcesContent":["import OpenAI from 'openai'\nimport { BaseTTSAdapter } from '@tanstack/ai/adapters'\nimport { toRunErrorPayload } from '@tanstack/ai/adapter-internals'\nimport { arrayBufferToBase64, generateId } from '@tanstack/ai-utils'\nimport { getGroqApiKeyFromEnv, withGroqDefaults } from '../utils/client'\nimport { validateAudioInput } from '../audio/audio-provider-options'\nimport type { TTSOptions, TTSResult } from '@tanstack/ai'\nimport type OpenAI_SDK from 'openai'\nimport type { GroqTTSModel } from '../model-meta'\nimport type { GroqTTSProviderOptions } from '../audio/tts-provider-options'\nimport type { GroqClientConfig } from '../utils/client'\n\n/**\n * Configuration for Groq TTS adapter\n */\nexport interface GroqTTSConfig extends GroqClientConfig {}\n\n/**\n * Groq Text-to-Speech Adapter\n *\n * Tree-shakeable adapter for Groq TTS functionality. Groq exposes an\n * OpenAI-compatible `/audio/speech` endpoint, so the adapter drives it with\n * the OpenAI SDK via a `baseURL` override (the same pattern as the Groq text\n * adapter).\n *\n * Supports `canopylabs/orpheus-v1-english` and\n * `canopylabs/orpheus-arabic-saudi`.\n *\n * Features:\n * - English voices: autumn(f), diana(f), hannah(f), austin(m), daniel(m), troy(m)\n * - Arabic voices: fahad(m), sultan(m), lulwa(f), noura(f)\n * - Output formats: flac, mp3, mulaw, ogg, wav (default wav)\n * - Speed control\n * - Configurable sample rate via `modelOptions`\n */\nexport class GroqTTSAdapter<TModel extends GroqTTSModel> extends BaseTTSAdapter<\n  TModel,\n  GroqTTSProviderOptions\n> {\n  readonly name = 'groq' as const\n\n  protected client: OpenAI\n\n  constructor(config: GroqTTSConfig, model: TModel) {\n    super(model, {})\n    this.client = new OpenAI(withGroqDefaults(config))\n  }\n\n  async generateSpeech(\n    options: TTSOptions<GroqTTSProviderOptions>,\n  ): Promise<TTSResult> {\n    const { model, text, voice, format, speed, modelOptions } = options\n\n    validateAudioInput({ input: text, model: this.model })\n\n    // Spreading optional inputs conditionally keeps the request compatible\n    // with the vendor SDK shape under exactOptionalPropertyTypes. `sample_rate`\n    // is a Groq-only body field carried via modelOptions.\n    const request: OpenAI_SDK.Audio.SpeechCreateParams = {\n      model,\n      input: text,\n      voice: voice ?? 'autumn',\n      response_format: format ?? 'wav',\n      ...(speed !== undefined && { speed }),\n      ...(modelOptions ?? {}),\n    }\n\n    try {\n      options.logger.request(\n        `activity=tts provider=${this.name} model=${model} format=${request.response_format ?? 'default'} voice=${request.voice}`,\n        { provider: this.name, model },\n      )\n      const response = await this.client.audio.speech.create(request)\n\n      const arrayBuffer = await response.arrayBuffer()\n      const base64 = arrayBufferToBase64(arrayBuffer)\n\n      const outputFormat = request.response_format ?? 'wav'\n      const contentType = this.getContentType(outputFormat)\n\n      return {\n        id: generateId(this.name),\n        model,\n        audio: base64,\n        format: outputFormat,\n        contentType,\n      }\n    } catch (error: unknown) {\n      // Narrow before logging: raw SDK errors can carry request metadata\n      // (including auth headers) which we must never surface to user loggers.\n      options.logger.errors(`${this.name}.generateSpeech fatal`, {\n        error: toRunErrorPayload(error, `${this.name}.generateSpeech failed`),\n        source: `${this.name}.generateSpeech`,\n      })\n      throw error\n    }\n  }\n\n  private getContentType(format: string): string {\n    const contentTypes: Record<string, string> = {\n      flac: 'audio/flac',\n      mp3: 'audio/mpeg',\n      mulaw: 'audio/basic',\n      ogg: 'audio/ogg',\n      wav: 'audio/wav',\n    }\n    return contentTypes[format] || 'audio/wav'\n  }\n}\n\n/**\n * Creates a Groq speech adapter with explicit API key.\n * Type resolution happens here at the call site.\n *\n * @param model - The model name (e.g., 'canopylabs/orpheus-v1-english')\n * @param apiKey - Your Groq API key\n * @param config - Optional additional configuration\n * @returns Configured Groq speech adapter instance with resolved types\n *\n * @example\n * ```typescript\n * const adapter = createGroqSpeech('canopylabs/orpheus-v1-english', 'gsk_...')\n *\n * const result = await generateSpeech({\n *   adapter,\n *   text: 'Hello, world!',\n *   voice: 'autumn',\n * })\n * ```\n */\nexport function createGroqSpeech<TModel extends GroqTTSModel>(\n  model: TModel,\n  apiKey: string,\n  config?: Omit<GroqTTSConfig, 'apiKey'>,\n): GroqTTSAdapter<TModel> {\n  return new GroqTTSAdapter({ apiKey, ...config }, model)\n}\n\n/**\n * Creates a Groq speech adapter with automatic API key detection from\n * environment variables.\n *\n * Looks for `GROQ_API_KEY` in the environment.\n *\n * @param model - The model name (e.g., 'canopylabs/orpheus-v1-english')\n * @param config - Optional configuration (excluding apiKey which is auto-detected)\n * @returns Configured Groq speech adapter instance with resolved types\n * @throws Error if GROQ_API_KEY is not found in environment\n *\n * @example\n * ```typescript\n * const adapter = groqSpeech('canopylabs/orpheus-v1-english')\n *\n * const result = await generateSpeech({\n *   adapter,\n *   text: 'Welcome to TanStack AI!',\n *   voice: 'autumn',\n *   format: 'wav',\n * })\n * ```\n */\nexport function groqSpeech<TModel extends GroqTTSModel>(\n  model: TModel,\n  config?: Omit<GroqTTSConfig, 'apiKey'>,\n): GroqTTSAdapter<TModel> {\n  const apiKey = getGroqApiKeyFromEnv()\n  return createGroqSpeech(model, apiKey, config)\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;AAmCA,IAAa,iBAAb,cAAiE,eAG/D;CACA,OAAgB;CAEhB;CAEA,YAAY,QAAuB,OAAe;EAChD,MAAM,OAAO,CAAC,CAAC;EACf,KAAK,SAAS,IAAI,OAAO,iBAAiB,MAAM,CAAC;CACnD;CAEA,MAAM,eACJ,SACoB;EACpB,MAAM,EAAE,OAAO,MAAM,OAAO,QAAQ,OAAO,iBAAiB;EAE5D,mBAAmB;GAAE,OAAO;GAAM,OAAO,KAAK;EAAM,CAAC;EAKrD,MAAM,UAA+C;GACnD;GACA,OAAO;GACP,OAAO,SAAS;GAChB,iBAAiB,UAAU;GAC3B,GAAI,UAAU,KAAA,KAAa,EAAE,MAAM;GACnC,GAAI,gBAAgB,CAAC;EACvB;EAEA,IAAI;GACF,QAAQ,OAAO,QACb,yBAAyB,KAAK,KAAK,SAAS,MAAM,UAAU,QAAQ,mBAAmB,UAAU,SAAS,QAAQ,SAClH;IAAE,UAAU,KAAK;IAAM;GAAM,CAC/B;GAIA,MAAM,SAAS,oBAAoB,OADT,MAFH,KAAK,OAAO,MAAM,OAAO,OAAO,OAAO,EAAA,CAE3B,YAAY,CACD;GAE9C,MAAM,eAAe,QAAQ,mBAAmB;GAChD,MAAM,cAAc,KAAK,eAAe,YAAY;GAEpD,OAAO;IACL,IAAI,WAAW,KAAK,IAAI;IACxB;IACA,OAAO;IACP,QAAQ;IACR;GACF;EACF,SAAS,OAAgB;GAGvB,QAAQ,OAAO,OAAO,GAAG,KAAK,KAAK,wBAAwB;IACzD,OAAO,kBAAkB,OAAO,GAAG,KAAK,KAAK,uBAAuB;IACpE,QAAQ,GAAG,KAAK,KAAK;GACvB,CAAC;GACD,MAAM;EACR;CACF;CAEA,eAAuB,QAAwB;EAQ7C,OAAO;GANL,MAAM;GACN,KAAK;GACL,OAAO;GACP,KAAK;GACL,KAAK;EAEA,EAAa,WAAW;CACjC;AACF;;;;;;;;;;;;;;;;;;;;;AAsBA,SAAgB,iBACd,OACA,QACA,QACwB;CACxB,OAAO,IAAI,eAAe;EAAE;EAAQ,GAAG;CAAO,GAAG,KAAK;AACxD;;;;;;;;;;;;;;;;;;;;;;;;AAyBA,SAAgB,WACd,OACA,QACwB;CAExB,OAAO,iBAAiB,OADT,qBACgB,GAAQ,MAAM;AAC/C"}

package/dist/esm/audio/audio-provider-options.js

@@ -1,9 +1,12 @@
-const validateAudioInput = (options) => {
-  if (options.input.length > 200) {
-    throw new Error("Input text exceeds maximum length of 200 characters.");
-  }
+//#region src/audio/audio-provider-options.ts
+/**
+* Validates that the audio input text does not exceed the maximum length.
+* @throws Error if input text exceeds 200 characters
+*/
+var validateAudioInput = (options) => {
+	if (options.input.length > 200) throw new Error("Input text exceeds maximum length of 200 characters.");
 };
-export {
-  validateAudioInput
-};
-//# sourceMappingURL=audio-provider-options.js.map
+//#endregion
+export { validateAudioInput };
+
+//# sourceMappingURL=audio-provider-options.js.map

package/dist/esm/audio/audio-provider-options.js.map

@@ -1 +1 @@
-{"version":3,"file":"audio-provider-options.js","sources":["../../../src/audio/audio-provider-options.ts"],"sourcesContent":["/**\n * Common audio provider options for Groq audio endpoints.\n */\nexport interface AudioProviderOptions {\n  /**\n   * The text to generate audio for.\n   * Maximum length is 200 characters.\n   * Use [directions] for vocal control (English voices only).\n   */\n  input: string\n  /**\n   * The audio model to use for generation.\n   */\n  model: string\n}\n\n/**\n * Validates that the audio input text does not exceed the maximum length.\n * @throws Error if input text exceeds 200 characters\n */\nexport const validateAudioInput = (options: AudioProviderOptions) => {\n  if (options.input.length > 200) {\n    throw new Error('Input text exceeds maximum length of 200 characters.')\n  }\n}\n"],"names":[],"mappings":"AAoBO,MAAM,qBAAqB,CAAC,YAAkC;AACnE,MAAI,QAAQ,MAAM,SAAS,KAAK;AAC9B,UAAM,IAAI,MAAM,sDAAsD;AAAA,EACxE;AACF;"}
+{"version":3,"file":"audio-provider-options.js","names":[],"sources":["../../../src/audio/audio-provider-options.ts"],"sourcesContent":["/**\n * Common audio provider options for Groq audio endpoints.\n */\nexport interface AudioProviderOptions {\n  /**\n   * The text to generate audio for.\n   * Maximum length is 200 characters.\n   * Use [directions] for vocal control (English voices only).\n   */\n  input: string\n  /**\n   * The audio model to use for generation.\n   */\n  model: string\n}\n\n/**\n * Validates that the audio input text does not exceed the maximum length.\n * @throws Error if input text exceeds 200 characters\n */\nexport const validateAudioInput = (options: AudioProviderOptions) => {\n  if (options.input.length > 200) {\n    throw new Error('Input text exceeds maximum length of 200 characters.')\n  }\n}\n"],"mappings":";;;;;AAoBA,IAAa,sBAAsB,YAAkC;CACnE,IAAI,QAAQ,MAAM,SAAS,KACzB,MAAM,IAAI,MAAM,sDAAsD;AAE1E"}

package/dist/esm/index.js

@@ -2,18 +2,4 @@ import { GroqTextAdapter, createGroqText, groqText } from "./adapters/text.js";
 import { GroqTranscriptionAdapter, createGroqTranscription, groqTranscription } from "./adapters/transcription.js";
 import { GroqTTSAdapter, createGroqSpeech, groqSpeech } from "./adapters/tts.js";
 import { GROQ_CHAT_MODELS, GROQ_TRANSCRIPTION_MODELS, GROQ_TTS_MODELS } from "./model-meta.js";
-export {
-  GROQ_CHAT_MODELS,
-  GROQ_TRANSCRIPTION_MODELS,
-  GROQ_TTS_MODELS,
-  GroqTTSAdapter,
-  GroqTextAdapter,
-  GroqTranscriptionAdapter,
-  createGroqSpeech,
-  createGroqText,
-  createGroqTranscription,
-  groqSpeech,
-  groqText,
-  groqTranscription
-};
-//# sourceMappingURL=index.js.map
+export { GROQ_CHAT_MODELS, GROQ_TRANSCRIPTION_MODELS, GROQ_TTS_MODELS, GroqTTSAdapter, GroqTextAdapter, GroqTranscriptionAdapter, createGroqSpeech, createGroqText, createGroqTranscription, groqSpeech, groqText, groqTranscription };