@xopcai/xopc 0.0.143 → 0.0.145

package/dist/src/voice/stt/providers/alibaba-transcription.js

@@ -1,7 +1,7 @@
 import { createLogger } from "../../../utils/logger/index.js";
 import { init_logger } from "../../../utils/logger.js";
 import { ProviderHttpError } from "../../../media-shared/http/provider-http-errors.js";
-import { fetchWithTimeoutGuarded, postJsonRequest } from "../../../media-shared/http/provider-http.js";
+import { fetchWithTimeoutGuarded, postJsonRequest, waitProviderOperationPollInterval } from "../../../media-shared/http/provider-http.js";
 import "../../../media-shared/http/index.js";
 import { registerMediaUnderstandingProvider } from "../../../media-understanding/registry.js";
 //#region src/voice/stt/providers/alibaba-transcription.ts
@@ -43,7 +43,8 @@ async function submitTask(params) {
 			model: params.model,
 			input: { file_urls: [params.audioDataUrl] },
 			...params.language ? { parameters: { language_hint: params.language } } : {}
-		}
+		},
+		signal: params.signal
 	})).json();
 	if (data.code) throw new Error(`Alibaba STT API error: ${data.code} - ${data.message}`);
 	if (!data.output?.task_id) {
@@ -60,6 +61,7 @@ async function pollTask(params) {
 		const response = await fetchWithTimeoutGuarded(url, {
 			timeoutMs: params.timeoutMs,
 			label: "Alibaba STT poll",
+			signal: params.signal,
 			init: {
 				method: "GET",
 				headers: { Authorization: `Bearer ${params.apiKey}` }
@@ -78,17 +80,18 @@ async function pollTask(params) {
 		if (status === "SUCCEEDED") {
 			const result = data.output?.results?.[0];
 			if (!result) throw new Error("Alibaba STT task succeeded but no results found");
-			return { text: (await fetchTranscription(result.transcription_url, params.timeoutMs)).transcripts.map((t) => t.text).join("\n") };
+			return { text: (await fetchTranscription(result.transcription_url, params.timeoutMs, params.signal)).transcripts.map((t) => t.text).join("\n") };
 		}
 		if (status === "FAILED") throw new Error("Alibaba STT task failed");
-		await new Promise((resolve) => setTimeout(resolve, POLL_INTERVAL_MS));
+		await waitProviderOperationPollInterval(POLL_INTERVAL_MS, params.signal);
 	}
 	throw new Error(`Alibaba STT task did not complete within ${MAX_POLL_MS}ms`);
 }
-async function fetchTranscription(url, timeoutMs) {
+async function fetchTranscription(url, timeoutMs, signal) {
 	const response = await fetchWithTimeoutGuarded(url, {
 		timeoutMs,
-		label: "Alibaba STT transcription fetch"
+		label: "Alibaba STT transcription fetch",
+		signal
 	});
 	if (!response.ok) throw new ProviderHttpError({
 		label: "Alibaba STT transcription fetch",
@@ -116,12 +119,14 @@ async function transcribeAudio(req) {
 			model,
 			audioDataUrl,
 			...req.language ? { language: req.language } : {},
-			timeoutMs: req.timeoutMs
+			timeoutMs: req.timeoutMs,
+			signal: req.signal
 		});
 		const result = await pollTask({
 			apiKey: req.apiKey,
 			taskId,
-			timeoutMs: req.timeoutMs
+			timeoutMs: req.timeoutMs,
+			signal: req.signal
 		});
 		const durationSeconds = (Date.now() - startTime) / 1e3;
 		log.info({
@@ -136,6 +141,7 @@ async function transcribeAudio(req) {
 			durationSeconds
 		};
 	} catch (error) {
+		if (req.signal?.aborted) throw error;
 		const errorMsg = error instanceof Error ? error.message : String(error);
 		log.error({
 			err: error,

package/dist/src/voice/stt/providers/alibaba-transcription.js.map

@@ -1 +1 @@
-{"version":3,"file":"alibaba-transcription.js","names":[],"sources":["../../../../../src/voice/stt/providers/alibaba-transcription.ts"],"sourcesContent":["/**\n * Alibaba DashScope Paraformer STT — implements MediaUnderstandingProvider.transcribeAudio.\n *\n * DashScope's async API is a 4-step dance:\n *   1. POST /api/v1/services/audio/asr/transcription with `X-DashScope-Async: enable`\n *      and `input.file_urls = [data:audio/ogg;base64,...]` → returns task_id\n *   2. Poll GET /api/v1/tasks/{task_id} until task_status === 'SUCCEEDED'\n *   3. The success result contains `transcription_url` pointing at a JSON document\n *   4. Fetch that JSON to extract the actual `transcripts[].text`\n *\n * All HTTP calls go through `fetchWithTimeoutGuarded` so the SSRF guard catches\n * malicious config injection. The transcription_url is a public OSS URL\n * (CDN-hosted); we keep `allowPrivateNetwork: false` so a malicious tenant\n * cannot redirect to an internal address.\n *\n * Polling caps at 60s by default (60 * 1s). The runner's `timeoutMs` controls\n * HTTP timeouts on each individual submit/poll/fetch call, not the overall\n * poll loop ceiling.\n */\n\nimport {\n  ProviderHttpError,\n  fetchWithTimeoutGuarded,\n  postJsonRequest,\n} from '../../../media-shared/http/index.js';\nimport { registerMediaUnderstandingProvider } from '../../../media-understanding/registry.js';\nimport type {\n  AudioTranscriptionRequest,\n  AudioTranscriptionResult,\n  MediaUnderstandingProvider,\n} from '../../../media-understanding/types.js';\nimport { createLogger } from '../../../utils/logger.js';\n\nconst log = createLogger('STT:Alibaba');\n\nconst DEFAULT_MODEL = 'paraformer-v2';\nconst DEFAULT_BASE_URL =\n  'https://dashscope.aliyuncs.com/api/v1/services/audio/asr/transcription';\nconst TASKS_BASE_URL = 'https://dashscope.aliyuncs.com/api/v1/tasks';\nconst POLL_INTERVAL_MS = 1000;\nconst MAX_POLL_MS = 60_000;\n\ninterface TaskResponse {\n  status_code?: number;\n  request_id?: string;\n  code?: string;\n  message?: string;\n  output?: {\n    task_id: string;\n    task_status: 'PENDING' | 'RUNNING' | 'SUCCEEDED' | 'FAILED';\n  };\n}\n\ninterface TranscriptionResultEntry {\n  file_url: string;\n  transcription_url: string;\n  subtask_status: string;\n}\n\ninterface FetchResponse {\n  status_code?: number;\n  request_id?: string;\n  output?: {\n    task_id: string;\n    task_status: 'PENDING' | 'RUNNING' | 'SUCCEEDED' | 'FAILED';\n    results?: TranscriptionResultEntry[];\n  };\n  usage?: { duration?: number };\n}\n\ninterface TranscriptionDetail {\n  file_url: string;\n  properties: {\n    audio_format: string;\n    channels: number[];\n    original_sampling_rate: number;\n    original_duration_in_milliseconds: number;\n  };\n  transcripts: Array<{\n    channel_id: number;\n    content_duration_in_milliseconds: number;\n    text: string;\n  }>;\n}\n\nasync function submitTask(params: {\n  apiKey: string;\n  baseUrl: string;\n  model: string;\n  audioDataUrl: string;\n  language?: string;\n  timeoutMs: number;\n}): Promise<string> {\n  const response = await postJsonRequest(params.baseUrl, {\n    timeoutMs: params.timeoutMs,\n    label: 'Alibaba STT submit',\n    headers: {\n      Authorization: `Bearer ${params.apiKey}`,\n      'X-DashScope-Async': 'enable',\n    },\n    body: {\n      model: params.model,\n      input: { file_urls: [params.audioDataUrl] },\n      ...(params.language ? { parameters: { language_hint: params.language } } : {}),\n    },\n  });\n  const data = (await response.json()) as TaskResponse;\n  if (data.code) {\n    throw new Error(`Alibaba STT API error: ${data.code} - ${data.message}`);\n  }\n  if (!data.output?.task_id) {\n    log.error({ response: data }, 'Alibaba STT API response missing task_id');\n    throw new Error(`No task_id returned from Alibaba STT API: ${JSON.stringify(data)}`);\n  }\n  log.debug({ taskId: data.output.task_id }, 'Task submitted');\n  return data.output.task_id;\n}\n\nasync function pollTask(params: {\n  apiKey: string;\n  taskId: string;\n  timeoutMs: number;\n}): Promise<{ text: string }> {\n  const startTime = Date.now();\n  const url = `${TASKS_BASE_URL}/${params.taskId}`;\n  while (Date.now() - startTime < MAX_POLL_MS) {\n    const response = await fetchWithTimeoutGuarded(url, {\n      timeoutMs: params.timeoutMs,\n      label: 'Alibaba STT poll',\n      init: {\n        method: 'GET',\n        headers: { Authorization: `Bearer ${params.apiKey}` },\n      },\n    });\n    if (!response.ok) {\n      const errorText = await response.text().catch(() => '');\n      throw new ProviderHttpError({\n        label: 'Alibaba STT poll',\n        status: response.status,\n        detail: errorText.slice(0, 220) || response.statusText,\n      });\n    }\n    const data = (await response.json()) as FetchResponse;\n    const status = data.output?.task_status;\n    if (status === 'SUCCEEDED') {\n      const result = data.output?.results?.[0];\n      if (!result) {\n        throw new Error('Alibaba STT task succeeded but no results found');\n      }\n      const transcription = await fetchTranscription(result.transcription_url, params.timeoutMs);\n      const fullText = transcription.transcripts.map((t) => t.text).join('\\n');\n      return { text: fullText };\n    }\n    if (status === 'FAILED') {\n      throw new Error('Alibaba STT task failed');\n    }\n    await new Promise((resolve) => setTimeout(resolve, POLL_INTERVAL_MS));\n  }\n  throw new Error(`Alibaba STT task did not complete within ${MAX_POLL_MS}ms`);\n}\n\nasync function fetchTranscription(url: string, timeoutMs: number): Promise<TranscriptionDetail> {\n  const response = await fetchWithTimeoutGuarded(url, {\n    timeoutMs,\n    label: 'Alibaba STT transcription fetch',\n  });\n  if (!response.ok) {\n    throw new ProviderHttpError({\n      label: 'Alibaba STT transcription fetch',\n      status: response.status,\n      detail: response.statusText,\n    });\n  }\n  return (await response.json()) as TranscriptionDetail;\n}\n\nasync function transcribeAudio(req: AudioTranscriptionRequest): Promise<AudioTranscriptionResult> {\n  const startTime = Date.now();\n  const model = req.model ?? DEFAULT_MODEL;\n  const baseUrl = req.baseUrl ?? DEFAULT_BASE_URL;\n\n  // Paraformer accepts `data:` URLs directly so we don't need to upload\n  // the audio first. This avoids the OSS pre-signing dance.\n  const base64Audio = req.buffer.toString('base64');\n  const audioDataUrl = `data:${req.mime ?? 'audio/ogg'};base64,${base64Audio}`;\n\n  log.debug(\n    { model, bufferSize: req.buffer.length, language: req.language, fileName: req.fileName },\n    'Sending to Alibaba Paraformer',\n  );\n\n  try {\n    const taskId = await submitTask({\n      apiKey: req.apiKey,\n      baseUrl,\n      model,\n      audioDataUrl,\n      ...(req.language ? { language: req.language } : {}),\n      timeoutMs: req.timeoutMs,\n    });\n    const result = await pollTask({\n      apiKey: req.apiKey,\n      taskId,\n      timeoutMs: req.timeoutMs,\n    });\n    const durationSeconds = (Date.now() - startTime) / 1000;\n    log.info(\n      { provider: 'alibaba', durationSeconds, textLength: result.text.length },\n      'Transcription completed',\n    );\n    return {\n      text: result.text,\n      model,\n      ...(req.language ? { language: req.language } : {}),\n      durationSeconds,\n    };\n  } catch (error) {\n    const errorMsg = error instanceof Error ? error.message : String(error);\n    log.error(\n      { err: error, bufferSize: req.buffer.length, model },\n      `Alibaba transcription failed: ${errorMsg}`,\n    );\n    throw new Error(`Alibaba STT failed: ${errorMsg}`);\n  }\n}\n\nexport const alibabaTranscriptionProvider: MediaUnderstandingProvider = {\n  id: 'alibaba',\n  aliases: ['dashscope', 'paraformer'],\n  capabilities: ['audio'],\n  envKey: 'DASHSCOPE_API_KEY',\n  defaultModels: { audio: DEFAULT_MODEL },\n  // Higher priority than openai for audio because Paraformer-v2 has better\n  // Chinese accuracy and is the project's typical primary STT.\n  autoPriority: { audio: 10 },\n  transcribeAudio,\n};\n\nregisterMediaUnderstandingProvider(alibabaTranscriptionProvider);\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;aA+BwD;AAExD,MAAM,MAAM,aAAa,cAAc;AAEvC,MAAM,gBAAgB;AACtB,MAAM,mBACJ;AACF,MAAM,iBAAiB;AACvB,MAAM,mBAAmB;AACzB,MAAM,cAAc;AA6CpB,eAAe,WAAW,QAON;CAclB,MAAM,OAAQ,OAAM,MAbG,gBAAgB,OAAO,SAAS;EACrD,WAAW,OAAO;EAClB,OAAO;EACP,SAAS;GACP,eAAe,UAAU,OAAO;GAChC,qBAAqB;GACtB;EACD,MAAM;GACJ,OAAO,OAAO;GACd,OAAO,EAAE,WAAW,CAAC,OAAO,aAAa,EAAE;GAC3C,GAAI,OAAO,WAAW,EAAE,YAAY,EAAE,eAAe,OAAO,UAAU,EAAE,GAAG,EAAE;GAC9E;EACF,CAAC,EAC2B,MAAM;AACnC,KAAI,KAAK,KACP,OAAM,IAAI,MAAM,0BAA0B,KAAK,KAAK,KAAK,KAAK,UAAU;AAE1E,KAAI,CAAC,KAAK,QAAQ,SAAS;AACzB,MAAI,MAAM,EAAE,UAAU,MAAM,EAAE,2CAA2C;AACzE,QAAM,IAAI,MAAM,6CAA6C,KAAK,UAAU,KAAK,GAAG;;AAEtF,KAAI,MAAM,EAAE,QAAQ,KAAK,OAAO,SAAS,EAAE,iBAAiB;AAC5D,QAAO,KAAK,OAAO;;AAGrB,eAAe,SAAS,QAIM;CAC5B,MAAM,YAAY,KAAK,KAAK;CAC5B,MAAM,MAAM,GAAG,eAAe,GAAG,OAAO;AACxC,QAAO,KAAK,KAAK,GAAG,YAAY,aAAa;EAC3C,MAAM,WAAW,MAAM,wBAAwB,KAAK;GAClD,WAAW,OAAO;GAClB,OAAO;GACP,MAAM;IACJ,QAAQ;IACR,SAAS,EAAE,eAAe,UAAU,OAAO,UAAU;IACtD;GACF,CAAC;AACF,MAAI,CAAC,SAAS,IAAI;GAChB,MAAM,YAAY,MAAM,SAAS,MAAM,CAAC,YAAY,GAAG;AACvD,SAAM,IAAI,kBAAkB;IAC1B,OAAO;IACP,QAAQ,SAAS;IACjB,QAAQ,UAAU,MAAM,GAAG,IAAI,IAAI,SAAS;IAC7C,CAAC;;EAEJ,MAAM,OAAQ,MAAM,SAAS,MAAM;EACnC,MAAM,SAAS,KAAK,QAAQ;AAC5B,MAAI,WAAW,aAAa;GAC1B,MAAM,SAAS,KAAK,QAAQ,UAAU;AACtC,OAAI,CAAC,OACH,OAAM,IAAI,MAAM,kDAAkD;AAIpE,UAAO,EAAE,OADQ,MADW,mBAAmB,OAAO,mBAAmB,OAAO,UAAU,EAC3D,YAAY,KAAK,MAAM,EAAE,KAAK,CAAC,KAAK,KAC5C,EAAE;;AAE3B,MAAI,WAAW,SACb,OAAM,IAAI,MAAM,0BAA0B;AAE5C,QAAM,IAAI,SAAS,YAAY,WAAW,SAAS,iBAAiB,CAAC;;AAEvE,OAAM,IAAI,MAAM,4CAA4C,YAAY,IAAI;;AAG9E,eAAe,mBAAmB,KAAa,WAAiD;CAC9F,MAAM,WAAW,MAAM,wBAAwB,KAAK;EAClD;EACA,OAAO;EACR,CAAC;AACF,KAAI,CAAC,SAAS,GACZ,OAAM,IAAI,kBAAkB;EAC1B,OAAO;EACP,QAAQ,SAAS;EACjB,QAAQ,SAAS;EAClB,CAAC;AAEJ,QAAQ,MAAM,SAAS,MAAM;;AAG/B,eAAe,gBAAgB,KAAmE;CAChG,MAAM,YAAY,KAAK,KAAK;CAC5B,MAAM,QAAQ,IAAI,SAAS;CAC3B,MAAM,UAAU,IAAI,WAAW;CAI/B,MAAM,cAAc,IAAI,OAAO,SAAS,SAAS;CACjD,MAAM,eAAe,QAAQ,IAAI,QAAQ,YAAY,UAAU;AAE/D,KAAI,MACF;EAAE;EAAO,YAAY,IAAI,OAAO;EAAQ,UAAU,IAAI;EAAU,UAAU,IAAI;EAAU,EACxF,gCACD;AAED,KAAI;EACF,MAAM,SAAS,MAAM,WAAW;GAC9B,QAAQ,IAAI;GACZ;GACA;GACA;GACA,GAAI,IAAI,WAAW,EAAE,UAAU,IAAI,UAAU,GAAG,EAAE;GAClD,WAAW,IAAI;GAChB,CAAC;EACF,MAAM,SAAS,MAAM,SAAS;GAC5B,QAAQ,IAAI;GACZ;GACA,WAAW,IAAI;GAChB,CAAC;EACF,MAAM,mBAAmB,KAAK,KAAK,GAAG,aAAa;AACnD,MAAI,KACF;GAAE,UAAU;GAAW;GAAiB,YAAY,OAAO,KAAK;GAAQ,EACxE,0BACD;AACD,SAAO;GACL,MAAM,OAAO;GACb;GACA,GAAI,IAAI,WAAW,EAAE,UAAU,IAAI,UAAU,GAAG,EAAE;GAClD;GACD;UACM,OAAO;EACd,MAAM,WAAW,iBAAiB,QAAQ,MAAM,UAAU,OAAO,MAAM;AACvE,MAAI,MACF;GAAE,KAAK;GAAO,YAAY,IAAI,OAAO;GAAQ;GAAO,EACpD,iCAAiC,WAClC;AACD,QAAM,IAAI,MAAM,uBAAuB,WAAW;;;AAItD,MAAa,+BAA2D;CACtE,IAAI;CACJ,SAAS,CAAC,aAAa,aAAa;CACpC,cAAc,CAAC,QAAQ;CACvB,QAAQ;CACR,eAAe,EAAE,OAAO,eAAe;CAGvC,cAAc,EAAE,OAAO,IAAI;CAC3B;CACD;AAED,mCAAmC,6BAA6B"}
+{"version":3,"file":"alibaba-transcription.js","names":[],"sources":["../../../../../src/voice/stt/providers/alibaba-transcription.ts"],"sourcesContent":["/**\n * Alibaba DashScope Paraformer STT — implements MediaUnderstandingProvider.transcribeAudio.\n *\n * DashScope's async API is a 4-step dance:\n *   1. POST /api/v1/services/audio/asr/transcription with `X-DashScope-Async: enable`\n *      and `input.file_urls = [data:audio/ogg;base64,...]` → returns task_id\n *   2. Poll GET /api/v1/tasks/{task_id} until task_status === 'SUCCEEDED'\n *   3. The success result contains `transcription_url` pointing at a JSON document\n *   4. Fetch that JSON to extract the actual `transcripts[].text`\n *\n * All HTTP calls go through `fetchWithTimeoutGuarded` so the SSRF guard catches\n * malicious config injection. The transcription_url is a public OSS URL\n * (CDN-hosted); we keep `allowPrivateNetwork: false` so a malicious tenant\n * cannot redirect to an internal address.\n *\n * Polling caps at 60s by default (60 * 1s). The runner's `timeoutMs` controls\n * HTTP timeouts on each individual submit/poll/fetch call, not the overall\n * poll loop ceiling.\n */\n\nimport {\n  ProviderHttpError,\n  fetchWithTimeoutGuarded,\n  postJsonRequest,\n  waitProviderOperationPollInterval,\n} from '../../../media-shared/http/index.js';\nimport { registerMediaUnderstandingProvider } from '../../../media-understanding/registry.js';\nimport type {\n  AudioTranscriptionRequest,\n  AudioTranscriptionResult,\n  MediaUnderstandingProvider,\n} from '../../../media-understanding/types.js';\nimport { createLogger } from '../../../utils/logger.js';\n\nconst log = createLogger('STT:Alibaba');\n\nconst DEFAULT_MODEL = 'paraformer-v2';\nconst DEFAULT_BASE_URL =\n  'https://dashscope.aliyuncs.com/api/v1/services/audio/asr/transcription';\nconst TASKS_BASE_URL = 'https://dashscope.aliyuncs.com/api/v1/tasks';\nconst POLL_INTERVAL_MS = 1000;\nconst MAX_POLL_MS = 60_000;\n\ninterface TaskResponse {\n  status_code?: number;\n  request_id?: string;\n  code?: string;\n  message?: string;\n  output?: {\n    task_id: string;\n    task_status: 'PENDING' | 'RUNNING' | 'SUCCEEDED' | 'FAILED';\n  };\n}\n\ninterface TranscriptionResultEntry {\n  file_url: string;\n  transcription_url: string;\n  subtask_status: string;\n}\n\ninterface FetchResponse {\n  status_code?: number;\n  request_id?: string;\n  output?: {\n    task_id: string;\n    task_status: 'PENDING' | 'RUNNING' | 'SUCCEEDED' | 'FAILED';\n    results?: TranscriptionResultEntry[];\n  };\n  usage?: { duration?: number };\n}\n\ninterface TranscriptionDetail {\n  file_url: string;\n  properties: {\n    audio_format: string;\n    channels: number[];\n    original_sampling_rate: number;\n    original_duration_in_milliseconds: number;\n  };\n  transcripts: Array<{\n    channel_id: number;\n    content_duration_in_milliseconds: number;\n    text: string;\n  }>;\n}\n\nasync function submitTask(params: {\n  apiKey: string;\n  baseUrl: string;\n  model: string;\n  audioDataUrl: string;\n  language?: string;\n  timeoutMs: number;\n  signal?: AbortSignal;\n}): Promise<string> {\n  const response = await postJsonRequest(params.baseUrl, {\n    timeoutMs: params.timeoutMs,\n    label: 'Alibaba STT submit',\n    headers: {\n      Authorization: `Bearer ${params.apiKey}`,\n      'X-DashScope-Async': 'enable',\n    },\n    body: {\n      model: params.model,\n      input: { file_urls: [params.audioDataUrl] },\n      ...(params.language ? { parameters: { language_hint: params.language } } : {}),\n    },\n    signal: params.signal,\n  });\n  const data = (await response.json()) as TaskResponse;\n  if (data.code) {\n    throw new Error(`Alibaba STT API error: ${data.code} - ${data.message}`);\n  }\n  if (!data.output?.task_id) {\n    log.error({ response: data }, 'Alibaba STT API response missing task_id');\n    throw new Error(`No task_id returned from Alibaba STT API: ${JSON.stringify(data)}`);\n  }\n  log.debug({ taskId: data.output.task_id }, 'Task submitted');\n  return data.output.task_id;\n}\n\nasync function pollTask(params: {\n  apiKey: string;\n  taskId: string;\n  timeoutMs: number;\n  signal?: AbortSignal;\n}): Promise<{ text: string }> {\n  const startTime = Date.now();\n  const url = `${TASKS_BASE_URL}/${params.taskId}`;\n  while (Date.now() - startTime < MAX_POLL_MS) {\n    const response = await fetchWithTimeoutGuarded(url, {\n      timeoutMs: params.timeoutMs,\n      label: 'Alibaba STT poll',\n      signal: params.signal,\n      init: {\n        method: 'GET',\n        headers: { Authorization: `Bearer ${params.apiKey}` },\n      },\n    });\n    if (!response.ok) {\n      const errorText = await response.text().catch(() => '');\n      throw new ProviderHttpError({\n        label: 'Alibaba STT poll',\n        status: response.status,\n        detail: errorText.slice(0, 220) || response.statusText,\n      });\n    }\n    const data = (await response.json()) as FetchResponse;\n    const status = data.output?.task_status;\n    if (status === 'SUCCEEDED') {\n      const result = data.output?.results?.[0];\n      if (!result) {\n        throw new Error('Alibaba STT task succeeded but no results found');\n      }\n      const transcription = await fetchTranscription(\n        result.transcription_url,\n        params.timeoutMs,\n        params.signal,\n      );\n      const fullText = transcription.transcripts.map((t) => t.text).join('\\n');\n      return { text: fullText };\n    }\n    if (status === 'FAILED') {\n      throw new Error('Alibaba STT task failed');\n    }\n    await waitProviderOperationPollInterval(POLL_INTERVAL_MS, params.signal);\n  }\n  throw new Error(`Alibaba STT task did not complete within ${MAX_POLL_MS}ms`);\n}\n\nasync function fetchTranscription(\n  url: string,\n  timeoutMs: number,\n  signal?: AbortSignal,\n): Promise<TranscriptionDetail> {\n  const response = await fetchWithTimeoutGuarded(url, {\n    timeoutMs,\n    label: 'Alibaba STT transcription fetch',\n    signal,\n  });\n  if (!response.ok) {\n    throw new ProviderHttpError({\n      label: 'Alibaba STT transcription fetch',\n      status: response.status,\n      detail: response.statusText,\n    });\n  }\n  return (await response.json()) as TranscriptionDetail;\n}\n\nasync function transcribeAudio(req: AudioTranscriptionRequest): Promise<AudioTranscriptionResult> {\n  const startTime = Date.now();\n  const model = req.model ?? DEFAULT_MODEL;\n  const baseUrl = req.baseUrl ?? DEFAULT_BASE_URL;\n\n  // Paraformer accepts `data:` URLs directly so we don't need to upload\n  // the audio first. This avoids the OSS pre-signing dance.\n  const base64Audio = req.buffer.toString('base64');\n  const audioDataUrl = `data:${req.mime ?? 'audio/ogg'};base64,${base64Audio}`;\n\n  log.debug(\n    { model, bufferSize: req.buffer.length, language: req.language, fileName: req.fileName },\n    'Sending to Alibaba Paraformer',\n  );\n\n  try {\n    const taskId = await submitTask({\n      apiKey: req.apiKey,\n      baseUrl,\n      model,\n      audioDataUrl,\n      ...(req.language ? { language: req.language } : {}),\n      timeoutMs: req.timeoutMs,\n      signal: req.signal,\n    });\n    const result = await pollTask({\n      apiKey: req.apiKey,\n      taskId,\n      timeoutMs: req.timeoutMs,\n      signal: req.signal,\n    });\n    const durationSeconds = (Date.now() - startTime) / 1000;\n    log.info(\n      { provider: 'alibaba', durationSeconds, textLength: result.text.length },\n      'Transcription completed',\n    );\n    return {\n      text: result.text,\n      model,\n      ...(req.language ? { language: req.language } : {}),\n      durationSeconds,\n    };\n  } catch (error) {\n    if (req.signal?.aborted) {\n      throw error;\n    }\n    const errorMsg = error instanceof Error ? error.message : String(error);\n    log.error(\n      { err: error, bufferSize: req.buffer.length, model },\n      `Alibaba transcription failed: ${errorMsg}`,\n    );\n    throw new Error(`Alibaba STT failed: ${errorMsg}`);\n  }\n}\n\nexport const alibabaTranscriptionProvider: MediaUnderstandingProvider = {\n  id: 'alibaba',\n  aliases: ['dashscope', 'paraformer'],\n  capabilities: ['audio'],\n  envKey: 'DASHSCOPE_API_KEY',\n  defaultModels: { audio: DEFAULT_MODEL },\n  // Higher priority than openai for audio because Paraformer-v2 has better\n  // Chinese accuracy and is the project's typical primary STT.\n  autoPriority: { audio: 10 },\n  transcribeAudio,\n};\n\nregisterMediaUnderstandingProvider(alibabaTranscriptionProvider);\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;aAgCwD;AAExD,MAAM,MAAM,aAAa,cAAc;AAEvC,MAAM,gBAAgB;AACtB,MAAM,mBACJ;AACF,MAAM,iBAAiB;AACvB,MAAM,mBAAmB;AACzB,MAAM,cAAc;AA6CpB,eAAe,WAAW,QAQN;CAelB,MAAM,OAAQ,OAAM,MAdG,gBAAgB,OAAO,SAAS;EACrD,WAAW,OAAO;EAClB,OAAO;EACP,SAAS;GACP,eAAe,UAAU,OAAO;GAChC,qBAAqB;GACtB;EACD,MAAM;GACJ,OAAO,OAAO;GACd,OAAO,EAAE,WAAW,CAAC,OAAO,aAAa,EAAE;GAC3C,GAAI,OAAO,WAAW,EAAE,YAAY,EAAE,eAAe,OAAO,UAAU,EAAE,GAAG,EAAE;GAC9E;EACD,QAAQ,OAAO;EAChB,CAAC,EAC2B,MAAM;AACnC,KAAI,KAAK,KACP,OAAM,IAAI,MAAM,0BAA0B,KAAK,KAAK,KAAK,KAAK,UAAU;AAE1E,KAAI,CAAC,KAAK,QAAQ,SAAS;AACzB,MAAI,MAAM,EAAE,UAAU,MAAM,EAAE,2CAA2C;AACzE,QAAM,IAAI,MAAM,6CAA6C,KAAK,UAAU,KAAK,GAAG;;AAEtF,KAAI,MAAM,EAAE,QAAQ,KAAK,OAAO,SAAS,EAAE,iBAAiB;AAC5D,QAAO,KAAK,OAAO;;AAGrB,eAAe,SAAS,QAKM;CAC5B,MAAM,YAAY,KAAK,KAAK;CAC5B,MAAM,MAAM,GAAG,eAAe,GAAG,OAAO;AACxC,QAAO,KAAK,KAAK,GAAG,YAAY,aAAa;EAC3C,MAAM,WAAW,MAAM,wBAAwB,KAAK;GAClD,WAAW,OAAO;GAClB,OAAO;GACP,QAAQ,OAAO;GACf,MAAM;IACJ,QAAQ;IACR,SAAS,EAAE,eAAe,UAAU,OAAO,UAAU;IACtD;GACF,CAAC;AACF,MAAI,CAAC,SAAS,IAAI;GAChB,MAAM,YAAY,MAAM,SAAS,MAAM,CAAC,YAAY,GAAG;AACvD,SAAM,IAAI,kBAAkB;IAC1B,OAAO;IACP,QAAQ,SAAS;IACjB,QAAQ,UAAU,MAAM,GAAG,IAAI,IAAI,SAAS;IAC7C,CAAC;;EAEJ,MAAM,OAAQ,MAAM,SAAS,MAAM;EACnC,MAAM,SAAS,KAAK,QAAQ;AAC5B,MAAI,WAAW,aAAa;GAC1B,MAAM,SAAS,KAAK,QAAQ,UAAU;AACtC,OAAI,CAAC,OACH,OAAM,IAAI,MAAM,kDAAkD;AAQpE,UAAO,EAAE,OADQ,MALW,mBAC1B,OAAO,mBACP,OAAO,WACP,OAAO,OACR,EAC8B,YAAY,KAAK,MAAM,EAAE,KAAK,CAAC,KAAK,KAC5C,EAAE;;AAE3B,MAAI,WAAW,SACb,OAAM,IAAI,MAAM,0BAA0B;AAE5C,QAAM,kCAAkC,kBAAkB,OAAO,OAAO;;AAE1E,OAAM,IAAI,MAAM,4CAA4C,YAAY,IAAI;;AAG9E,eAAe,mBACb,KACA,WACA,QAC8B;CAC9B,MAAM,WAAW,MAAM,wBAAwB,KAAK;EAClD;EACA,OAAO;EACP;EACD,CAAC;AACF,KAAI,CAAC,SAAS,GACZ,OAAM,IAAI,kBAAkB;EAC1B,OAAO;EACP,QAAQ,SAAS;EACjB,QAAQ,SAAS;EAClB,CAAC;AAEJ,QAAQ,MAAM,SAAS,MAAM;;AAG/B,eAAe,gBAAgB,KAAmE;CAChG,MAAM,YAAY,KAAK,KAAK;CAC5B,MAAM,QAAQ,IAAI,SAAS;CAC3B,MAAM,UAAU,IAAI,WAAW;CAI/B,MAAM,cAAc,IAAI,OAAO,SAAS,SAAS;CACjD,MAAM,eAAe,QAAQ,IAAI,QAAQ,YAAY,UAAU;AAE/D,KAAI,MACF;EAAE;EAAO,YAAY,IAAI,OAAO;EAAQ,UAAU,IAAI;EAAU,UAAU,IAAI;EAAU,EACxF,gCACD;AAED,KAAI;EACF,MAAM,SAAS,MAAM,WAAW;GAC9B,QAAQ,IAAI;GACZ;GACA;GACA;GACA,GAAI,IAAI,WAAW,EAAE,UAAU,IAAI,UAAU,GAAG,EAAE;GAClD,WAAW,IAAI;GACf,QAAQ,IAAI;GACb,CAAC;EACF,MAAM,SAAS,MAAM,SAAS;GAC5B,QAAQ,IAAI;GACZ;GACA,WAAW,IAAI;GACf,QAAQ,IAAI;GACb,CAAC;EACF,MAAM,mBAAmB,KAAK,KAAK,GAAG,aAAa;AACnD,MAAI,KACF;GAAE,UAAU;GAAW;GAAiB,YAAY,OAAO,KAAK;GAAQ,EACxE,0BACD;AACD,SAAO;GACL,MAAM,OAAO;GACb;GACA,GAAI,IAAI,WAAW,EAAE,UAAU,IAAI,UAAU,GAAG,EAAE;GAClD;GACD;UACM,OAAO;AACd,MAAI,IAAI,QAAQ,QACd,OAAM;EAER,MAAM,WAAW,iBAAiB,QAAQ,MAAM,UAAU,OAAO,MAAM;AACvE,MAAI,MACF;GAAE,KAAK;GAAO,YAAY,IAAI,OAAO;GAAQ;GAAO,EACpD,iCAAiC,WAClC;AACD,QAAM,IAAI,MAAM,uBAAuB,WAAW;;;AAItD,MAAa,+BAA2D;CACtE,IAAI;CACJ,SAAS,CAAC,aAAa,aAAa;CACpC,cAAc,CAAC,QAAQ;CACvB,QAAQ;CACR,eAAe,EAAE,OAAO,eAAe;CAGvC,cAAc,EAAE,OAAO,IAAI;CAC3B;CACD;AAED,mCAAmC,6BAA6B"}

package/dist/src/voice/stt/providers/index.d.ts

@@ -12,3 +12,4 @@
  */
 export { openAiTranscriptionProvider } from './openai-transcription.js';
 export { alibabaTranscriptionProvider } from './alibaba-transcription.js';
+export { localTranscriptionProvider } from './local-transcription.js';

package/dist/src/voice/stt/providers/index.js

@@ -1,3 +1,4 @@
 import { openAiTranscriptionProvider } from "./openai-transcription.js";
 import { alibabaTranscriptionProvider } from "./alibaba-transcription.js";
-export { alibabaTranscriptionProvider, openAiTranscriptionProvider };
+import { localTranscriptionProvider } from "./local-transcription.js";
+export { alibabaTranscriptionProvider, localTranscriptionProvider, openAiTranscriptionProvider };

package/dist/src/voice/stt/providers/local-transcription.d.ts

@@ -0,0 +1,2 @@
+import type { MediaUnderstandingProvider } from '../../../media-understanding/types.js';
+export declare const localTranscriptionProvider: MediaUnderstandingProvider;

package/dist/src/voice/stt/providers/local-transcription.js

@@ -0,0 +1,51 @@
+import { createLogger } from "../../../utils/logger/index.js";
+import { init_logger } from "../../../utils/logger.js";
+import { registerMediaUnderstandingProvider } from "../../../media-understanding/registry.js";
+import { DEFAULT_LOCAL_VOICE_MODEL_ID, getLocalVoiceModel, hasInstalledLocalVoiceModel, isLocalVoiceModelInstalled } from "../../local/models.js";
+import { getLocalVoiceRuntimeClient } from "../../local/runtime-client.js";
+import { decodeWavToMonoFloat32 } from "../../local/wav.js";
+//#region src/voice/stt/providers/local-transcription.ts
+init_logger();
+const log = createLogger("STT:Local");
+async function transcribeAudio(req) {
+	const model = getLocalVoiceModel(req.model);
+	if (!isLocalVoiceModelInstalled(model.id)) throw new Error(`Local voice model "${model.id}" is not installed`);
+	const decoded = decodeWavToMonoFloat32(req.buffer);
+	const audioBytes = Buffer.from(decoded.samples.buffer, decoded.samples.byteOffset, decoded.samples.byteLength);
+	const startedAt = Date.now();
+	const result = await getLocalVoiceRuntimeClient().request("transcribe", {
+		modelId: model.id,
+		audioBase64: audioBytes.toString("base64"),
+		...req.language ? { language: req.language } : {}
+	}, {
+		signal: req.signal,
+		timeoutMs: req.timeoutMs
+	});
+	log.info({
+		model: model.id,
+		audioDurationSeconds: decoded.durationSeconds,
+		latencyMs: Date.now() - startedAt,
+		textLength: result.text.length
+	}, "Local transcription completed");
+	return {
+		text: result.text,
+		model: model.id,
+		...result.language || req.language ? { language: result.language ?? req.language } : {},
+		durationSeconds: decoded.durationSeconds
+	};
+}
+const localTranscriptionProvider = {
+	id: "xopc-local",
+	aliases: ["local"],
+	capabilities: ["audio"],
+	requiresApiKey: false,
+	defaultModels: { audio: DEFAULT_LOCAL_VOICE_MODEL_ID },
+	autoPriority: { audio: 0 },
+	isConfigured: (options) => options?.model ? isLocalVoiceModelInstalled(options.model) : hasInstalledLocalVoiceModel(),
+	transcribeAudio
+};
+registerMediaUnderstandingProvider(localTranscriptionProvider);
+//#endregion
+export { localTranscriptionProvider };
+
+//# sourceMappingURL=local-transcription.js.map

package/dist/src/voice/stt/providers/local-transcription.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"local-transcription.js","names":[],"sources":["../../../../../src/voice/stt/providers/local-transcription.ts"],"sourcesContent":["import { registerMediaUnderstandingProvider } from '../../../media-understanding/registry.js';\nimport type {\n  AudioTranscriptionRequest,\n  AudioTranscriptionResult,\n  MediaUnderstandingProvider,\n} from '../../../media-understanding/types.js';\nimport { createLogger } from '../../../utils/logger.js';\nimport { getLocalVoiceRuntimeClient } from '../../local/runtime-client.js';\nimport { decodeWavToMonoFloat32 } from '../../local/wav.js';\nimport {\n  DEFAULT_LOCAL_VOICE_MODEL_ID,\n  getLocalVoiceModel,\n  hasInstalledLocalVoiceModel,\n  isLocalVoiceModelInstalled,\n} from '../../local/models.js';\n\nconst log = createLogger('STT:Local');\n\nasync function transcribeAudio(req: AudioTranscriptionRequest): Promise<AudioTranscriptionResult> {\n  const model = getLocalVoiceModel(req.model);\n  if (!isLocalVoiceModelInstalled(model.id)) {\n    throw new Error(`Local voice model \"${model.id}\" is not installed`);\n  }\n  const decoded = decodeWavToMonoFloat32(req.buffer);\n  const audioBytes = Buffer.from(\n    decoded.samples.buffer,\n    decoded.samples.byteOffset,\n    decoded.samples.byteLength,\n  );\n  const startedAt = Date.now();\n  const result = await getLocalVoiceRuntimeClient().request<{\n    text: string;\n    modelId: string;\n    language?: string;\n    emotion?: string;\n    event?: string;\n  }>(\n    'transcribe',\n    {\n      modelId: model.id,\n      audioBase64: audioBytes.toString('base64'),\n      ...(req.language ? { language: req.language } : {}),\n    },\n    { signal: req.signal, timeoutMs: req.timeoutMs },\n  );\n  log.info(\n    {\n      model: model.id,\n      audioDurationSeconds: decoded.durationSeconds,\n      latencyMs: Date.now() - startedAt,\n      textLength: result.text.length,\n    },\n    'Local transcription completed',\n  );\n  return {\n    text: result.text,\n    model: model.id,\n    ...(result.language || req.language ? { language: result.language ?? req.language } : {}),\n    durationSeconds: decoded.durationSeconds,\n  };\n}\n\nexport const localTranscriptionProvider: MediaUnderstandingProvider = {\n  id: 'xopc-local',\n  aliases: ['local'],\n  capabilities: ['audio'],\n  requiresApiKey: false,\n  defaultModels: { audio: DEFAULT_LOCAL_VOICE_MODEL_ID },\n  autoPriority: { audio: 0 },\n  isConfigured: (options) => options?.model\n    ? isLocalVoiceModelInstalled(options.model)\n    : hasInstalledLocalVoiceModel(),\n  transcribeAudio,\n};\n\nregisterMediaUnderstandingProvider(localTranscriptionProvider);\n"],"mappings":";;;;;;;aAMwD;AAUxD,MAAM,MAAM,aAAa,YAAY;AAErC,eAAe,gBAAgB,KAAmE;CAChG,MAAM,QAAQ,mBAAmB,IAAI,MAAM;AAC3C,KAAI,CAAC,2BAA2B,MAAM,GAAG,CACvC,OAAM,IAAI,MAAM,sBAAsB,MAAM,GAAG,oBAAoB;CAErE,MAAM,UAAU,uBAAuB,IAAI,OAAO;CAClD,MAAM,aAAa,OAAO,KACxB,QAAQ,QAAQ,QAChB,QAAQ,QAAQ,YAChB,QAAQ,QAAQ,WACjB;CACD,MAAM,YAAY,KAAK,KAAK;CAC5B,MAAM,SAAS,MAAM,4BAA4B,CAAC,QAOhD,cACA;EACE,SAAS,MAAM;EACf,aAAa,WAAW,SAAS,SAAS;EAC1C,GAAI,IAAI,WAAW,EAAE,UAAU,IAAI,UAAU,GAAG,EAAE;EACnD,EACD;EAAE,QAAQ,IAAI;EAAQ,WAAW,IAAI;EAAW,CACjD;AACD,KAAI,KACF;EACE,OAAO,MAAM;EACb,sBAAsB,QAAQ;EAC9B,WAAW,KAAK,KAAK,GAAG;EACxB,YAAY,OAAO,KAAK;EACzB,EACD,gCACD;AACD,QAAO;EACL,MAAM,OAAO;EACb,OAAO,MAAM;EACb,GAAI,OAAO,YAAY,IAAI,WAAW,EAAE,UAAU,OAAO,YAAY,IAAI,UAAU,GAAG,EAAE;EACxF,iBAAiB,QAAQ;EAC1B;;AAGH,MAAa,6BAAyD;CACpE,IAAI;CACJ,SAAS,CAAC,QAAQ;CAClB,cAAc,CAAC,QAAQ;CACvB,gBAAgB;CAChB,eAAe,EAAE,OAAO,8BAA8B;CACtD,cAAc,EAAE,OAAO,GAAG;CAC1B,eAAe,YAAY,SAAS,QAChC,2BAA2B,QAAQ,MAAM,GACzC,6BAA6B;CACjC;CACD;AAED,mCAAmC,2BAA2B"}

package/dist/src/voice/stt/providers/openai-transcription.js

@@ -25,11 +25,12 @@ async function transcribeAudio(req) {
 	const model = req.model ?? DEFAULT_MODEL;
 	const client = new OpenAI({
 		apiKey: req.apiKey,
+		timeout: req.timeoutMs,
 		...req.baseUrl ? { baseURL: req.baseUrl } : {},
 		...req.headers ? { defaultHeaders: req.headers } : {}
 	});
 	const uint8Array = new Uint8Array(req.buffer.buffer, req.buffer.byteOffset, req.buffer.byteLength);
-	const blob = new Blob([uint8Array], { type: req.mime ?? "audio/ogg" });
+	const file = new File([uint8Array], req.fileName, { type: req.mime ?? "audio/ogg" });
 	log.debug({
 		model,
 		bufferSize: req.buffer.length,
@@ -38,12 +39,12 @@ async function transcribeAudio(req) {
 	}, "Sending to OpenAI Whisper");
 	try {
 		const result = await client.audio.transcriptions.create({
-			file: blob,
+			file,
 			model,
 			...req.language ? { language: req.language } : {},
 			...req.prompt ? { prompt: req.prompt } : {},
 			response_format: "json"
-		});
+		}, req.signal ? { signal: req.signal } : void 0);
 		const durationSeconds = (Date.now() - startTime) / 1e3;
 		log.info({
 			provider: "openai",
@@ -57,6 +58,7 @@ async function transcribeAudio(req) {
 			durationSeconds
 		};
 	} catch (error) {
+		if (req.signal?.aborted) throw error;
 		const errorMsg = error instanceof Error ? error.message : String(error);
 		log.error({
 			err: error,

package/dist/src/voice/stt/providers/openai-transcription.js.map

@@ -1 +1 @@
-{"version":3,"file":"openai-transcription.js","names":[],"sources":["../../../../../src/voice/stt/providers/openai-transcription.ts"],"sourcesContent":["/**\n * OpenAI Whisper STT — implements MediaUnderstandingProvider.transcribeAudio.\n *\n * Uses the official `openai` SDK rather than the raw HTTP path because the SDK\n * already handles multipart/form-data encoding (`file` field), streaming\n * response normalization, and per-call apiKey override via constructor\n * injection. The media-shared HTTP layer is intentionally not used here so we\n * keep parity with the SDK's own fetch (proxy support, retries, etc.); the\n * only egress is api.openai.com which is HTTPS-only and public.\n *\n * Streaming transcription is not implemented (Whisper returns a final text\n * blob). Self-registers with `capabilities: ['audio']` only — the registry's\n * `listProvidersForCapability` filter skips this provider for image/video.\n */\n\nimport OpenAI from 'openai';\n\nimport { registerMediaUnderstandingProvider } from '../../../media-understanding/registry.js';\nimport type {\n  AudioTranscriptionRequest,\n  AudioTranscriptionResult,\n  MediaUnderstandingProvider,\n} from '../../../media-understanding/types.js';\nimport { createLogger } from '../../../utils/logger.js';\n\nconst log = createLogger('STT:OpenAI');\n\nconst DEFAULT_MODEL = 'whisper-1';\n\nasync function transcribeAudio(req: AudioTranscriptionRequest): Promise<AudioTranscriptionResult> {\n  const startTime = Date.now();\n  const model = req.model ?? DEFAULT_MODEL;\n\n  // openai SDK supports baseURL override per-instance.\n  const client = new OpenAI({\n    apiKey: req.apiKey,\n    ...(req.baseUrl ? { baseURL: req.baseUrl } : {}),\n    ...(req.headers ? { defaultHeaders: req.headers } : {}),\n  });\n\n  const uint8Array = new Uint8Array(\n    req.buffer.buffer as ArrayBuffer,\n    req.buffer.byteOffset,\n    req.buffer.byteLength,\n  );\n  // MIME defaults to audio/ogg because Telegram voice notes are the most common\n  // input. The SDK doesn't actually inspect the type — it just uses it for the\n  // multipart Content-Type header. Whisper itself sniffs the file content.\n  const blob = new Blob([uint8Array], { type: req.mime ?? 'audio/ogg' });\n\n  log.debug(\n    { model, bufferSize: req.buffer.length, language: req.language, fileName: req.fileName },\n    'Sending to OpenAI Whisper',\n  );\n\n  try {\n    const result = await client.audio.transcriptions.create({\n      file: blob,\n      model,\n      ...(req.language ? { language: req.language } : {}),\n      ...(req.prompt ? { prompt: req.prompt } : {}),\n      response_format: 'json',\n    });\n    const durationSeconds = (Date.now() - startTime) / 1000;\n    log.info(\n      { provider: 'openai', durationSeconds, textLength: result.text?.length ?? 0 },\n      'Transcription completed',\n    );\n    return {\n      text: result.text || '',\n      model,\n      ...(req.language ? { language: req.language } : {}),\n      durationSeconds,\n    };\n  } catch (error) {\n    const errorMsg = error instanceof Error ? error.message : String(error);\n    log.error(\n      { err: error, bufferSize: req.buffer.length, model },\n      `OpenAI transcription failed: ${errorMsg}`,\n    );\n    throw new Error(`OpenAI STT failed: ${errorMsg}`);\n  }\n}\n\nexport const openAiTranscriptionProvider: MediaUnderstandingProvider = {\n  id: 'openai',\n  capabilities: ['audio'],\n  envKey: 'OPENAI_API_KEY',\n  defaultModels: { audio: DEFAULT_MODEL },\n  autoPriority: { audio: 20 }, // alibaba is preferred when both configured (Chinese-first)\n  transcribeAudio,\n};\n\nregisterMediaUnderstandingProvider(openAiTranscriptionProvider);\n"],"mappings":";;;;;;;;;;;;;;;;;;;aAuBwD;AAExD,MAAM,MAAM,aAAa,aAAa;AAEtC,MAAM,gBAAgB;AAEtB,eAAe,gBAAgB,KAAmE;CAChG,MAAM,YAAY,KAAK,KAAK;CAC5B,MAAM,QAAQ,IAAI,SAAS;CAG3B,MAAM,SAAS,IAAI,OAAO;EACxB,QAAQ,IAAI;EACZ,GAAI,IAAI,UAAU,EAAE,SAAS,IAAI,SAAS,GAAG,EAAE;EAC/C,GAAI,IAAI,UAAU,EAAE,gBAAgB,IAAI,SAAS,GAAG,EAAE;EACvD,CAAC;CAEF,MAAM,aAAa,IAAI,WACrB,IAAI,OAAO,QACX,IAAI,OAAO,YACX,IAAI,OAAO,WACZ;CAID,MAAM,OAAO,IAAI,KAAK,CAAC,WAAW,EAAE,EAAE,MAAM,IAAI,QAAQ,aAAa,CAAC;AAEtE,KAAI,MACF;EAAE;EAAO,YAAY,IAAI,OAAO;EAAQ,UAAU,IAAI;EAAU,UAAU,IAAI;EAAU,EACxF,4BACD;AAED,KAAI;EACF,MAAM,SAAS,MAAM,OAAO,MAAM,eAAe,OAAO;GACtD,MAAM;GACN;GACA,GAAI,IAAI,WAAW,EAAE,UAAU,IAAI,UAAU,GAAG,EAAE;GAClD,GAAI,IAAI,SAAS,EAAE,QAAQ,IAAI,QAAQ,GAAG,EAAE;GAC5C,iBAAiB;GAClB,CAAC;EACF,MAAM,mBAAmB,KAAK,KAAK,GAAG,aAAa;AACnD,MAAI,KACF;GAAE,UAAU;GAAU;GAAiB,YAAY,OAAO,MAAM,UAAU;GAAG,EAC7E,0BACD;AACD,SAAO;GACL,MAAM,OAAO,QAAQ;GACrB;GACA,GAAI,IAAI,WAAW,EAAE,UAAU,IAAI,UAAU,GAAG,EAAE;GAClD;GACD;UACM,OAAO;EACd,MAAM,WAAW,iBAAiB,QAAQ,MAAM,UAAU,OAAO,MAAM;AACvE,MAAI,MACF;GAAE,KAAK;GAAO,YAAY,IAAI,OAAO;GAAQ;GAAO,EACpD,gCAAgC,WACjC;AACD,QAAM,IAAI,MAAM,sBAAsB,WAAW;;;AAIrD,MAAa,8BAA0D;CACrE,IAAI;CACJ,cAAc,CAAC,QAAQ;CACvB,QAAQ;CACR,eAAe,EAAE,OAAO,eAAe;CACvC,cAAc,EAAE,OAAO,IAAI;CAC3B;CACD;AAED,mCAAmC,4BAA4B"}
+{"version":3,"file":"openai-transcription.js","names":[],"sources":["../../../../../src/voice/stt/providers/openai-transcription.ts"],"sourcesContent":["/**\n * OpenAI Whisper STT — implements MediaUnderstandingProvider.transcribeAudio.\n *\n * Uses the official `openai` SDK rather than the raw HTTP path because the SDK\n * already handles multipart/form-data encoding (`file` field), streaming\n * response normalization, and per-call apiKey override via constructor\n * injection. The media-shared HTTP layer is intentionally not used here so we\n * keep parity with the SDK's own fetch (proxy support, retries, etc.); the\n * only egress is api.openai.com which is HTTPS-only and public.\n *\n * Streaming transcription is not implemented (Whisper returns a final text\n * blob). Self-registers with `capabilities: ['audio']` only — the registry's\n * `listProvidersForCapability` filter skips this provider for image/video.\n */\n\nimport OpenAI from 'openai';\n\nimport { registerMediaUnderstandingProvider } from '../../../media-understanding/registry.js';\nimport type {\n  AudioTranscriptionRequest,\n  AudioTranscriptionResult,\n  MediaUnderstandingProvider,\n} from '../../../media-understanding/types.js';\nimport { createLogger } from '../../../utils/logger.js';\n\nconst log = createLogger('STT:OpenAI');\n\nconst DEFAULT_MODEL = 'whisper-1';\n\nasync function transcribeAudio(req: AudioTranscriptionRequest): Promise<AudioTranscriptionResult> {\n  const startTime = Date.now();\n  const model = req.model ?? DEFAULT_MODEL;\n\n  // openai SDK supports baseURL override per-instance.\n  const client = new OpenAI({\n    apiKey: req.apiKey,\n    timeout: req.timeoutMs,\n    ...(req.baseUrl ? { baseURL: req.baseUrl } : {}),\n    ...(req.headers ? { defaultHeaders: req.headers } : {}),\n  });\n\n  const uint8Array = new Uint8Array(\n    req.buffer.buffer as ArrayBuffer,\n    req.buffer.byteOffset,\n    req.buffer.byteLength,\n  );\n  // Preserve the original name and MIME so provider-side format detection and\n  // diagnostics reflect the real upload instead of a synthetic OGG file.\n  const file = new File([uint8Array], req.fileName, { type: req.mime ?? 'audio/ogg' });\n\n  log.debug(\n    { model, bufferSize: req.buffer.length, language: req.language, fileName: req.fileName },\n    'Sending to OpenAI Whisper',\n  );\n\n  try {\n    const result = await client.audio.transcriptions.create(\n      {\n        file,\n        model,\n        ...(req.language ? { language: req.language } : {}),\n        ...(req.prompt ? { prompt: req.prompt } : {}),\n        response_format: 'json',\n      },\n      req.signal ? { signal: req.signal } : undefined,\n    );\n    const durationSeconds = (Date.now() - startTime) / 1000;\n    log.info(\n      { provider: 'openai', durationSeconds, textLength: result.text?.length ?? 0 },\n      'Transcription completed',\n    );\n    return {\n      text: result.text || '',\n      model,\n      ...(req.language ? { language: req.language } : {}),\n      durationSeconds,\n    };\n  } catch (error) {\n    if (req.signal?.aborted) {\n      throw error;\n    }\n    const errorMsg = error instanceof Error ? error.message : String(error);\n    log.error(\n      { err: error, bufferSize: req.buffer.length, model },\n      `OpenAI transcription failed: ${errorMsg}`,\n    );\n    throw new Error(`OpenAI STT failed: ${errorMsg}`);\n  }\n}\n\nexport const openAiTranscriptionProvider: MediaUnderstandingProvider = {\n  id: 'openai',\n  capabilities: ['audio'],\n  envKey: 'OPENAI_API_KEY',\n  defaultModels: { audio: DEFAULT_MODEL },\n  autoPriority: { audio: 20 }, // alibaba is preferred when both configured (Chinese-first)\n  transcribeAudio,\n};\n\nregisterMediaUnderstandingProvider(openAiTranscriptionProvider);\n"],"mappings":";;;;;;;;;;;;;;;;;;;aAuBwD;AAExD,MAAM,MAAM,aAAa,aAAa;AAEtC,MAAM,gBAAgB;AAEtB,eAAe,gBAAgB,KAAmE;CAChG,MAAM,YAAY,KAAK,KAAK;CAC5B,MAAM,QAAQ,IAAI,SAAS;CAG3B,MAAM,SAAS,IAAI,OAAO;EACxB,QAAQ,IAAI;EACZ,SAAS,IAAI;EACb,GAAI,IAAI,UAAU,EAAE,SAAS,IAAI,SAAS,GAAG,EAAE;EAC/C,GAAI,IAAI,UAAU,EAAE,gBAAgB,IAAI,SAAS,GAAG,EAAE;EACvD,CAAC;CAEF,MAAM,aAAa,IAAI,WACrB,IAAI,OAAO,QACX,IAAI,OAAO,YACX,IAAI,OAAO,WACZ;CAGD,MAAM,OAAO,IAAI,KAAK,CAAC,WAAW,EAAE,IAAI,UAAU,EAAE,MAAM,IAAI,QAAQ,aAAa,CAAC;AAEpF,KAAI,MACF;EAAE;EAAO,YAAY,IAAI,OAAO;EAAQ,UAAU,IAAI;EAAU,UAAU,IAAI;EAAU,EACxF,4BACD;AAED,KAAI;EACF,MAAM,SAAS,MAAM,OAAO,MAAM,eAAe,OAC/C;GACE;GACA;GACA,GAAI,IAAI,WAAW,EAAE,UAAU,IAAI,UAAU,GAAG,EAAE;GAClD,GAAI,IAAI,SAAS,EAAE,QAAQ,IAAI,QAAQ,GAAG,EAAE;GAC5C,iBAAiB;GAClB,EACD,IAAI,SAAS,EAAE,QAAQ,IAAI,QAAQ,GAAG,KAAA,EACvC;EACD,MAAM,mBAAmB,KAAK,KAAK,GAAG,aAAa;AACnD,MAAI,KACF;GAAE,UAAU;GAAU;GAAiB,YAAY,OAAO,MAAM,UAAU;GAAG,EAC7E,0BACD;AACD,SAAO;GACL,MAAM,OAAO,QAAQ;GACrB;GACA,GAAI,IAAI,WAAW,EAAE,UAAU,IAAI,UAAU,GAAG,EAAE;GAClD;GACD;UACM,OAAO;AACd,MAAI,IAAI,QAAQ,QACd,OAAM;EAER,MAAM,WAAW,iBAAiB,QAAQ,MAAM,UAAU,OAAO,MAAM;AACvE,MAAI,MACF;GAAE,KAAK;GAAO,YAAY,IAAI,OAAO;GAAQ;GAAO,EACpD,gCAAgC,WACjC;AACD,QAAM,IAAI,MAAM,sBAAsB,WAAW;;;AAIrD,MAAa,8BAA0D;CACrE,IAAI;CACJ,cAAc,CAAC,QAAQ;CACvB,QAAQ;CACR,eAAe,EAAE,OAAO,eAAe;CACvC,cAAc,EAAE,OAAO,IAAI;CAC3B;CACD;AAED,mCAAmC,4BAA4B"}

package/dist/src/voice/stt/transcribe-core.js

@@ -18,7 +18,7 @@ function toAttempt(attempt) {
 		provider: attempt.provider ?? "unknown",
 		outcome: attempt.outcome,
 		reasonCode: attempt.outcome === "success" ? "success" : classifyFailureReason(attempt.reason),
-		latencyMs: 0,
+		latencyMs: attempt.latencyMs ?? 0,
 		...attempt.reason ? { error: attempt.reason } : {}
 	};
 }
@@ -35,10 +35,11 @@ async function transcribe(audioBuffer, config, options) {
 		attachments: [{
 			attachmentIndex: 0,
 			buffer: audioBuffer,
-			fileName: `audio-${Date.now()}.ogg`,
-			mime: "audio/ogg"
+			fileName: options?.fileName ?? `audio-${Date.now()}.ogg`,
+			mime: options?.mime ?? "audio/ogg"
 		}],
-		timeoutMs: config.timeoutMs ?? 6e4
+		timeoutMs: config.timeoutMs ?? 6e4,
+		signal: options?.signal
 	});
 	const attachmentDecision = result.decision.attachments[0];
 	const attempts = (attachmentDecision?.attempts ?? []).map(toAttempt);

package/dist/src/voice/stt/transcribe-core.js.map

@@ -1 +1 @@
-{"version":3,"file":"transcribe-core.js","names":[],"sources":["../../../../src/voice/stt/transcribe-core.ts"],"sourcesContent":["/**\n * STT orchestrator — `transcribe()` is the single entry point used by channel\n * adapters (telegram, webchat) and the preflight router.\n *\n * Internally delegates to `runAudioTranscription` from the media-understanding\n * runner. The runner returns a `MediaUnderstandingDecision`; we translate each\n * `decision.attachments[0].attempts[]` entry into the `STTProviderAttempt`\n * shape consumed by downstream telemetry.\n *\n * One-attachment fast path: `transcribe()` is 1-buffer-in / 1-text-out, so we\n * wrap the input as a single CapabilityAttachmentInput and unpack the first\n * output. Multi-attachment cases go through `runAudioTranscription` directly.\n */\n\nimport { createLogger } from '../../utils/logger.js';\n\nimport { runAudioTranscription } from '../../media-understanding/audio-transcription-runner.js';\nimport type {\n  MediaUnderstandingAttachmentDecision,\n  MediaUnderstandingModelDecision,\n} from '../../media-understanding/types.js';\nimport { resolveSTTProviderChain } from './factory.js';\nimport type {\n  STTConfig,\n  STTOptions,\n  STTProviderAttempt,\n  STTProviderFailureReason,\n  STTResultWithTracking,\n} from './types.js';\n\nconst log = createLogger('STT');\n\nfunction classifyFailureReason(reason: string | undefined): STTProviderFailureReason {\n  if (!reason) return 'unknown';\n  const lower = reason.toLowerCase();\n  if (lower.includes('not configured') || lower.includes('api key') || lower.includes('missing')) {\n    return 'not_configured';\n  }\n  if (lower.includes('timeout') || lower.includes('timed out') || lower.includes('aborted')) {\n    return 'timeout';\n  }\n  if (lower.includes('unsupported') || lower.includes('format')) {\n    return 'unsupported_format';\n  }\n  return 'provider_error';\n}\n\nfunction toAttempt(attempt: MediaUnderstandingModelDecision): STTProviderAttempt {\n  return {\n    provider: attempt.provider ?? 'unknown',\n    outcome: attempt.outcome,\n    reasonCode: attempt.outcome === 'success' ? 'success' : classifyFailureReason(attempt.reason),\n    // Latency is not tracked at the runner level today (deferred to future\n    // observability pass); we report 0 to keep the field shape stable.\n    latencyMs: 0,\n    ...(attempt.reason ? { error: attempt.reason } : {}),\n  };\n}\n\nexport async function transcribe(\n  audioBuffer: Buffer,\n  config: STTConfig,\n  options?: STTOptions,\n): Promise<STTResultWithTracking> {\n  if (!config.enabled) {\n    throw new Error('STT is not enabled');\n  }\n  const providers = resolveSTTProviderChain(config);\n  if (providers.length === 0) {\n    throw new Error('No STT providers configured');\n  }\n\n  // Apply caller-supplied language hint to each provider's resolved config.\n  // (We don't bake it into resolveSTTProviderChain because it is per-call.)\n  const providersWithLanguage = options?.language\n    ? providers.map((p) => ({ ...p, language: options.language }))\n    : providers;\n\n  const result = await runAudioTranscription({\n    providers: providersWithLanguage,\n    attachments: [\n      {\n        attachmentIndex: 0,\n        buffer: audioBuffer,\n        fileName: `audio-${Date.now()}.ogg`,\n        // mime is informational for OpenAI's multipart blob; alibaba uses base64 data URL.\n        mime: 'audio/ogg',\n      },\n    ],\n    timeoutMs: config.timeoutMs ?? 60_000,\n  });\n\n  const attachmentDecision: MediaUnderstandingAttachmentDecision | undefined =\n    result.decision.attachments[0];\n  const attempts = (attachmentDecision?.attempts ?? []).map(toAttempt);\n  const attemptedProviders = attempts.map((a) => a.provider);\n\n  if (!attachmentDecision?.chosen) {\n    log.error({ attempts, attemptedProviders }, 'All STT providers failed');\n    const lastError = attempts.find((a) => a.outcome === 'failed')?.error;\n    throw new Error(lastError ?? 'All STT providers failed');\n  }\n\n  const chosen = attachmentDecision.chosen;\n  const output = result.outputs[0];\n  if (!output) {\n    throw new Error('STT runner returned chosen attempt but no output text');\n  }\n  const primaryProvider = providersWithLanguage[0]!.id;\n  const fallbackFrom =\n    chosen.provider && chosen.provider !== primaryProvider ? primaryProvider : undefined;\n\n  return {\n    text: output.text,\n    provider: chosen.provider ?? 'unknown',\n    ...(chosen.model ? { language: options?.language } : {}),\n    attempts,\n    ...(fallbackFrom ? { fallbackFrom } : {}),\n    attemptedProviders,\n  };\n}\n"],"mappings":";;;;;aAcqD;AAgBrD,MAAM,MAAM,aAAa,MAAM;AAE/B,SAAS,sBAAsB,QAAsD;AACnF,KAAI,CAAC,OAAQ,QAAO;CACpB,MAAM,QAAQ,OAAO,aAAa;AAClC,KAAI,MAAM,SAAS,iBAAiB,IAAI,MAAM,SAAS,UAAU,IAAI,MAAM,SAAS,UAAU,CAC5F,QAAO;AAET,KAAI,MAAM,SAAS,UAAU,IAAI,MAAM,SAAS,YAAY,IAAI,MAAM,SAAS,UAAU,CACvF,QAAO;AAET,KAAI,MAAM,SAAS,cAAc,IAAI,MAAM,SAAS,SAAS,CAC3D,QAAO;AAET,QAAO;;AAGT,SAAS,UAAU,SAA8D;AAC/E,QAAO;EACL,UAAU,QAAQ,YAAY;EAC9B,SAAS,QAAQ;EACjB,YAAY,QAAQ,YAAY,YAAY,YAAY,sBAAsB,QAAQ,OAAO;EAG7F,WAAW;EACX,GAAI,QAAQ,SAAS,EAAE,OAAO,QAAQ,QAAQ,GAAG,EAAE;EACpD;;AAGH,eAAsB,WACpB,aACA,QACA,SACgC;AAChC,KAAI,CAAC,OAAO,QACV,OAAM,IAAI,MAAM,qBAAqB;CAEvC,MAAM,YAAY,wBAAwB,OAAO;AACjD,KAAI,UAAU,WAAW,EACvB,OAAM,IAAI,MAAM,8BAA8B;CAKhD,MAAM,wBAAwB,SAAS,WACnC,UAAU,KAAK,OAAO;EAAE,GAAG;EAAG,UAAU,QAAQ;EAAU,EAAE,GAC5D;CAEJ,MAAM,SAAS,MAAM,sBAAsB;EACzC,WAAW;EACX,aAAa,CACX;GACE,iBAAiB;GACjB,QAAQ;GACR,UAAU,SAAS,KAAK,KAAK,CAAC;GAE9B,MAAM;GACP,CACF;EACD,WAAW,OAAO,aAAa;EAChC,CAAC;CAEF,MAAM,qBACJ,OAAO,SAAS,YAAY;CAC9B,MAAM,YAAY,oBAAoB,YAAY,EAAE,EAAE,IAAI,UAAU;CACpE,MAAM,qBAAqB,SAAS,KAAK,MAAM,EAAE,SAAS;AAE1D,KAAI,CAAC,oBAAoB,QAAQ;AAC/B,MAAI,MAAM;GAAE;GAAU;GAAoB,EAAE,2BAA2B;EACvE,MAAM,YAAY,SAAS,MAAM,MAAM,EAAE,YAAY,SAAS,EAAE;AAChE,QAAM,IAAI,MAAM,aAAa,2BAA2B;;CAG1D,MAAM,SAAS,mBAAmB;CAClC,MAAM,SAAS,OAAO,QAAQ;AAC9B,KAAI,CAAC,OACH,OAAM,IAAI,MAAM,wDAAwD;CAE1E,MAAM,kBAAkB,sBAAsB,GAAI;CAClD,MAAM,eACJ,OAAO,YAAY,OAAO,aAAa,kBAAkB,kBAAkB,KAAA;AAE7E,QAAO;EACL,MAAM,OAAO;EACb,UAAU,OAAO,YAAY;EAC7B,GAAI,OAAO,QAAQ,EAAE,UAAU,SAAS,UAAU,GAAG,EAAE;EACvD;EACA,GAAI,eAAe,EAAE,cAAc,GAAG,EAAE;EACxC;EACD"}
+{"version":3,"file":"transcribe-core.js","names":[],"sources":["../../../../src/voice/stt/transcribe-core.ts"],"sourcesContent":["/**\n * STT orchestrator — `transcribe()` is the single entry point used by channel\n * adapters (telegram, webchat) and the preflight router.\n *\n * Internally delegates to `runAudioTranscription` from the media-understanding\n * runner. The runner returns a `MediaUnderstandingDecision`; we translate each\n * `decision.attachments[0].attempts[]` entry into the `STTProviderAttempt`\n * shape consumed by downstream telemetry.\n *\n * One-attachment fast path: `transcribe()` is 1-buffer-in / 1-text-out, so we\n * wrap the input as a single CapabilityAttachmentInput and unpack the first\n * output. Multi-attachment cases go through `runAudioTranscription` directly.\n */\n\nimport { createLogger } from '../../utils/logger.js';\n\nimport { runAudioTranscription } from '../../media-understanding/audio-transcription-runner.js';\nimport type {\n  MediaUnderstandingAttachmentDecision,\n  MediaUnderstandingModelDecision,\n} from '../../media-understanding/types.js';\nimport { resolveSTTProviderChain } from './factory.js';\nimport type {\n  STTConfig,\n  STTOptions,\n  STTProviderAttempt,\n  STTProviderFailureReason,\n  STTResultWithTracking,\n} from './types.js';\n\nconst log = createLogger('STT');\n\nfunction classifyFailureReason(reason: string | undefined): STTProviderFailureReason {\n  if (!reason) return 'unknown';\n  const lower = reason.toLowerCase();\n  if (lower.includes('not configured') || lower.includes('api key') || lower.includes('missing')) {\n    return 'not_configured';\n  }\n  if (lower.includes('timeout') || lower.includes('timed out') || lower.includes('aborted')) {\n    return 'timeout';\n  }\n  if (lower.includes('unsupported') || lower.includes('format')) {\n    return 'unsupported_format';\n  }\n  return 'provider_error';\n}\n\nfunction toAttempt(attempt: MediaUnderstandingModelDecision): STTProviderAttempt {\n  return {\n    provider: attempt.provider ?? 'unknown',\n    outcome: attempt.outcome,\n    reasonCode: attempt.outcome === 'success' ? 'success' : classifyFailureReason(attempt.reason),\n    latencyMs: attempt.latencyMs ?? 0,\n    ...(attempt.reason ? { error: attempt.reason } : {}),\n  };\n}\n\nexport async function transcribe(\n  audioBuffer: Buffer,\n  config: STTConfig,\n  options?: STTOptions,\n): Promise<STTResultWithTracking> {\n  if (!config.enabled) {\n    throw new Error('STT is not enabled');\n  }\n  const providers = resolveSTTProviderChain(config);\n  if (providers.length === 0) {\n    throw new Error('No STT providers configured');\n  }\n\n  // Apply caller-supplied language hint to each provider's resolved config.\n  // (We don't bake it into resolveSTTProviderChain because it is per-call.)\n  const providersWithLanguage = options?.language\n    ? providers.map((p) => ({ ...p, language: options.language }))\n    : providers;\n\n  const result = await runAudioTranscription({\n    providers: providersWithLanguage,\n    attachments: [\n      {\n        attachmentIndex: 0,\n        buffer: audioBuffer,\n        fileName: options?.fileName ?? `audio-${Date.now()}.ogg`,\n        mime: options?.mime ?? 'audio/ogg',\n      },\n    ],\n    timeoutMs: config.timeoutMs ?? 60_000,\n    signal: options?.signal,\n  });\n\n  const attachmentDecision: MediaUnderstandingAttachmentDecision | undefined =\n    result.decision.attachments[0];\n  const attempts = (attachmentDecision?.attempts ?? []).map(toAttempt);\n  const attemptedProviders = attempts.map((a) => a.provider);\n\n  if (!attachmentDecision?.chosen) {\n    log.error({ attempts, attemptedProviders }, 'All STT providers failed');\n    const lastError = attempts.find((a) => a.outcome === 'failed')?.error;\n    throw new Error(lastError ?? 'All STT providers failed');\n  }\n\n  const chosen = attachmentDecision.chosen;\n  const output = result.outputs[0];\n  if (!output) {\n    throw new Error('STT runner returned chosen attempt but no output text');\n  }\n  const primaryProvider = providersWithLanguage[0]!.id;\n  const fallbackFrom =\n    chosen.provider && chosen.provider !== primaryProvider ? primaryProvider : undefined;\n\n  return {\n    text: output.text,\n    provider: chosen.provider ?? 'unknown',\n    ...(chosen.model ? { language: options?.language } : {}),\n    attempts,\n    ...(fallbackFrom ? { fallbackFrom } : {}),\n    attemptedProviders,\n  };\n}\n"],"mappings":";;;;;aAcqD;AAgBrD,MAAM,MAAM,aAAa,MAAM;AAE/B,SAAS,sBAAsB,QAAsD;AACnF,KAAI,CAAC,OAAQ,QAAO;CACpB,MAAM,QAAQ,OAAO,aAAa;AAClC,KAAI,MAAM,SAAS,iBAAiB,IAAI,MAAM,SAAS,UAAU,IAAI,MAAM,SAAS,UAAU,CAC5F,QAAO;AAET,KAAI,MAAM,SAAS,UAAU,IAAI,MAAM,SAAS,YAAY,IAAI,MAAM,SAAS,UAAU,CACvF,QAAO;AAET,KAAI,MAAM,SAAS,cAAc,IAAI,MAAM,SAAS,SAAS,CAC3D,QAAO;AAET,QAAO;;AAGT,SAAS,UAAU,SAA8D;AAC/E,QAAO;EACL,UAAU,QAAQ,YAAY;EAC9B,SAAS,QAAQ;EACjB,YAAY,QAAQ,YAAY,YAAY,YAAY,sBAAsB,QAAQ,OAAO;EAC7F,WAAW,QAAQ,aAAa;EAChC,GAAI,QAAQ,SAAS,EAAE,OAAO,QAAQ,QAAQ,GAAG,EAAE;EACpD;;AAGH,eAAsB,WACpB,aACA,QACA,SACgC;AAChC,KAAI,CAAC,OAAO,QACV,OAAM,IAAI,MAAM,qBAAqB;CAEvC,MAAM,YAAY,wBAAwB,OAAO;AACjD,KAAI,UAAU,WAAW,EACvB,OAAM,IAAI,MAAM,8BAA8B;CAKhD,MAAM,wBAAwB,SAAS,WACnC,UAAU,KAAK,OAAO;EAAE,GAAG;EAAG,UAAU,QAAQ;EAAU,EAAE,GAC5D;CAEJ,MAAM,SAAS,MAAM,sBAAsB;EACzC,WAAW;EACX,aAAa,CACX;GACE,iBAAiB;GACjB,QAAQ;GACR,UAAU,SAAS,YAAY,SAAS,KAAK,KAAK,CAAC;GACnD,MAAM,SAAS,QAAQ;GACxB,CACF;EACD,WAAW,OAAO,aAAa;EAC/B,QAAQ,SAAS;EAClB,CAAC;CAEF,MAAM,qBACJ,OAAO,SAAS,YAAY;CAC9B,MAAM,YAAY,oBAAoB,YAAY,EAAE,EAAE,IAAI,UAAU;CACpE,MAAM,qBAAqB,SAAS,KAAK,MAAM,EAAE,SAAS;AAE1D,KAAI,CAAC,oBAAoB,QAAQ;AAC/B,MAAI,MAAM;GAAE;GAAU;GAAoB,EAAE,2BAA2B;EACvE,MAAM,YAAY,SAAS,MAAM,MAAM,EAAE,YAAY,SAAS,EAAE;AAChE,QAAM,IAAI,MAAM,aAAa,2BAA2B;;CAG1D,MAAM,SAAS,mBAAmB;CAClC,MAAM,SAAS,OAAO,QAAQ;AAC9B,KAAI,CAAC,OACH,OAAM,IAAI,MAAM,wDAAwD;CAE1E,MAAM,kBAAkB,sBAAsB,GAAI;CAClD,MAAM,eACJ,OAAO,YAAY,OAAO,aAAa,kBAAkB,kBAAkB,KAAA;AAE7E,QAAO;EACL,MAAM,OAAO;EACb,UAAU,OAAO,YAAY;EAC7B,GAAI,OAAO,QAAQ,EAAE,UAAU,SAAS,UAAU,GAAG,EAAE;EACvD;EACA,GAAI,eAAe,EAAE,cAAc,GAAG,EAAE;EACxC;EACD"}

package/dist/src/voice/stt/types.d.ts

@@ -20,6 +20,10 @@ export interface STTOptions {
     language?: string;
     /** Model id (provider-specific). */
     model?: string;
+    /** Original upload metadata propagated to provider multipart/data URLs. */
+    mime?: string;
+    fileName?: string;
+    signal?: AbortSignal;
 }
 export type { MediaUnderstandingModelEntry } from '../../media-understanding/types.js';
 import type { MediaUnderstandingModelEntry } from '../../media-understanding/types.js';

package/dist/src/voice/stt/types.js

@@ -1,14 +1,15 @@
 //#region src/voice/stt/types.ts
 const DEFAULT_STT_CONFIG = {
-	enabled: false,
-	provider: "alibaba",
+	enabled: true,
+	provider: "xopc-local",
 	providers: {
+		"xopc-local": { model: "sensevoice-small" },
 		alibaba: { model: "paraformer-v2" },
 		openai: { model: "whisper-1" }
 	},
 	fallback: {
-		enabled: true,
-		order: ["alibaba", "openai"]
+		enabled: false,
+		order: ["xopc-local"]
 	},
 	timeoutMs: 6e4
 };

package/dist/src/voice/stt/types.js.map

@@ -1 +1 @@
-{"version":3,"file":"types.js","names":[],"sources":["../../../../src/voice/stt/types.ts"],"sourcesContent":["/**\n * STT public types — data shapes consumed by downstream code (channels,\n * schema validation, gateway payloads).\n *\n * Provider implementations live behind `MediaUnderstandingProvider.transcribeAudio`\n * (see src/media-understanding/types.ts).\n */\n\nexport interface STTResult {\n  /** Transcribed text. */\n  text: string;\n  /** Provider that performed the transcription (e.g. \"alibaba\", \"openai\"). */\n  provider: string;\n  /** Audio duration in seconds (if reported). */\n  duration?: number;\n  /** Detected language (if reported). */\n  language?: string;\n}\n\nexport interface STTOptions {\n  /** Language hint (e.g. 'zh', 'en'). */\n  language?: string;\n  /** Model id (provider-specific). */\n  model?: string;\n}\n\nexport type { MediaUnderstandingModelEntry } from '../../media-understanding/types.js';\nimport type { MediaUnderstandingModelEntry } from '../../media-understanding/types.js';\n\n/** STTConfig consumed by the schema and downstream wiring. */\nexport interface STTConfig {\n  enabled: boolean;\n  provider: string;\n  /** Capability-local model chain (`tools.media.audio.models`). */\n  models?: MediaUnderstandingModelEntry[];\n  /** Shared model entries merged from `tools.media.models` at runtime. */\n  sharedModels?: MediaUnderstandingModelEntry[];\n  /** Provider settings map (`tools.media.audio.providers.<id>`). */\n  providers?: Record<string, Record<string, unknown>>;\n  fallback?: {\n    enabled: boolean;\n    order: string[];\n  };\n  /** Hard timeout per provider call (ms). Default 60s. */\n  timeoutMs?: number;\n}\n\nexport type STTProviderAttemptOutcome = 'success' | 'skipped' | 'failed';\n\nexport type STTProviderFailureReason =\n  | 'success'\n  | 'not_configured'\n  | 'timeout'\n  | 'provider_error'\n  | 'unsupported_format'\n  | 'unknown';\n\nexport interface STTProviderAttempt {\n  provider: string;\n  outcome: STTProviderAttemptOutcome;\n  reasonCode: STTProviderFailureReason;\n  latencyMs: number;\n  error?: string;\n}\n\nexport interface STTResultWithTracking extends STTResult {\n  attempts: STTProviderAttempt[];\n  fallbackFrom?: string;\n  attemptedProviders: string[];\n}\n\nexport const DEFAULT_STT_CONFIG: STTConfig = {\n  enabled: false,\n  provider: 'alibaba',\n  providers: {\n    alibaba: { model: 'paraformer-v2' },\n    openai: { model: 'whisper-1' },\n  },\n  fallback: {\n    enabled: true,\n    order: ['alibaba', 'openai'],\n  },\n  timeoutMs: 60_000,\n};\n"],"mappings":";AAuEA,MAAa,qBAAgC;CAC3C,SAAS;CACT,UAAU;CACV,WAAW;EACT,SAAS,EAAE,OAAO,iBAAiB;EACnC,QAAQ,EAAE,OAAO,aAAa;EAC/B;CACD,UAAU;EACR,SAAS;EACT,OAAO,CAAC,WAAW,SAAS;EAC7B;CACD,WAAW;CACZ"}
+{"version":3,"file":"types.js","names":[],"sources":["../../../../src/voice/stt/types.ts"],"sourcesContent":["/**\n * STT public types — data shapes consumed by downstream code (channels,\n * schema validation, gateway payloads).\n *\n * Provider implementations live behind `MediaUnderstandingProvider.transcribeAudio`\n * (see src/media-understanding/types.ts).\n */\n\nexport interface STTResult {\n  /** Transcribed text. */\n  text: string;\n  /** Provider that performed the transcription (e.g. \"alibaba\", \"openai\"). */\n  provider: string;\n  /** Audio duration in seconds (if reported). */\n  duration?: number;\n  /** Detected language (if reported). */\n  language?: string;\n}\n\nexport interface STTOptions {\n  /** Language hint (e.g. 'zh', 'en'). */\n  language?: string;\n  /** Model id (provider-specific). */\n  model?: string;\n  /** Original upload metadata propagated to provider multipart/data URLs. */\n  mime?: string;\n  fileName?: string;\n  signal?: AbortSignal;\n}\n\nexport type { MediaUnderstandingModelEntry } from '../../media-understanding/types.js';\nimport type { MediaUnderstandingModelEntry } from '../../media-understanding/types.js';\n\n/** STTConfig consumed by the schema and downstream wiring. */\nexport interface STTConfig {\n  enabled: boolean;\n  provider: string;\n  /** Capability-local model chain (`tools.media.audio.models`). */\n  models?: MediaUnderstandingModelEntry[];\n  /** Shared model entries merged from `tools.media.models` at runtime. */\n  sharedModels?: MediaUnderstandingModelEntry[];\n  /** Provider settings map (`tools.media.audio.providers.<id>`). */\n  providers?: Record<string, Record<string, unknown>>;\n  fallback?: {\n    enabled: boolean;\n    order: string[];\n  };\n  /** Hard timeout per provider call (ms). Default 60s. */\n  timeoutMs?: number;\n}\n\nexport type STTProviderAttemptOutcome = 'success' | 'skipped' | 'failed';\n\nexport type STTProviderFailureReason =\n  | 'success'\n  | 'not_configured'\n  | 'timeout'\n  | 'provider_error'\n  | 'unsupported_format'\n  | 'unknown';\n\nexport interface STTProviderAttempt {\n  provider: string;\n  outcome: STTProviderAttemptOutcome;\n  reasonCode: STTProviderFailureReason;\n  latencyMs: number;\n  error?: string;\n}\n\nexport interface STTResultWithTracking extends STTResult {\n  attempts: STTProviderAttempt[];\n  fallbackFrom?: string;\n  attemptedProviders: string[];\n}\n\nexport const DEFAULT_STT_CONFIG: STTConfig = {\n  enabled: true,\n  provider: 'xopc-local',\n  providers: {\n    'xopc-local': { model: 'sensevoice-small' },\n    alibaba: { model: 'paraformer-v2' },\n    openai: { model: 'whisper-1' },\n  },\n  fallback: {\n    enabled: false,\n    order: ['xopc-local'],\n  },\n  timeoutMs: 60_000,\n};\n"],"mappings":";AA2EA,MAAa,qBAAgC;CAC3C,SAAS;CACT,UAAU;CACV,WAAW;EACT,cAAc,EAAE,OAAO,oBAAoB;EAC3C,SAAS,EAAE,OAAO,iBAAiB;EACnC,QAAQ,EAAE,OAAO,aAAa;EAC/B;CACD,UAAU;EACR,SAAS;EACT,OAAO,CAAC,aAAa;EACtB;CACD,WAAW;CACZ"}

package/dist/src/voice/tts/providers/edge-speech.js

@@ -20,8 +20,8 @@ import { mkdtempSync, readFileSync, rmSync, statSync } from "node:fs";
 */
 init_logger();
 const log = createLogger("SpeechProvider:Edge");
-const DEFAULT_VOICE = "en-US-MichelleNeural";
-const DEFAULT_LANG = "en-US";
+const DEFAULT_VOICE = "zh-CN-XiaoxiaoNeural";
+const DEFAULT_LANG = "zh-CN";
 const DEFAULT_OUTPUT_FORMAT = "audio-24khz-48kbitrate-mono-mp3";
 function trimToUndefined(value) {
 	if (typeof value !== "string") return;

package/dist/src/voice/tts/providers/edge-speech.js.map

@@ -1 +1 @@
-{"version":3,"file":"edge-speech.js","names":[],"sources":["../../../../../src/voice/tts/providers/edge-speech.ts"],"sourcesContent":["/**\n * Edge TTS provider — wraps the `node-edge-tts` package. Requires no API key\n * (Microsoft's Edge browser endpoint is unauthenticated).\n *\n * Implementation notes:\n *   - Temp-file dance (mkdtemp → ttsPromise → readFileSync → rm) is required\n *     because `node-edge-tts` does not expose a stream-to-buffer API.\n *   - `synthesizeStream` is implemented as \"synthesize + wrap as single-chunk\n *     stream\" so callers see uniform behavior. This is not a true network\n *     stream; it's buffered-then-wrapped.\n *   - `inferEdgeExtension` derives the file extension from Edge's outputFormat\n *     string.\n */\n\nimport { mkdtempSync, readFileSync, rmSync, statSync } from 'node:fs';\nimport { tmpdir } from 'node:os';\nimport path from 'node:path';\n\nimport { createLogger } from '../../../utils/logger.js';\nimport { registerSpeechProvider } from '../speech-registry.js';\nimport type {\n  SpeechDirectiveTokenParseContext,\n  SpeechDirectiveTokenParseResult,\n  SpeechProviderConfig,\n  SpeechProviderConfiguredContext,\n  SpeechProviderPlugin,\n  SpeechProviderResolveConfigContext,\n  SpeechSynthesisRequest,\n  SpeechSynthesisResult,\n  SpeechSynthesisStreamRequest,\n  SpeechSynthesisStreamResult,\n} from '../speech-provider-types.js';\n\nconst log = createLogger('SpeechProvider:Edge');\n\nconst DEFAULT_VOICE = 'en-US-MichelleNeural';\nconst DEFAULT_LANG = 'en-US';\nconst DEFAULT_OUTPUT_FORMAT = 'audio-24khz-48kbitrate-mono-mp3';\n\ninterface EdgeTtsConfig extends Record<string, unknown> {\n  voice: string;\n  lang: string;\n  outputFormat: string;\n  pitch?: string;\n  rate?: string;\n  volume?: string;\n  proxy?: string;\n  enabled: boolean;\n}\n\nfunction trimToUndefined(value: unknown): string | undefined {\n  if (typeof value !== 'string') {\n    return undefined;\n  }\n  const trimmed = value.trim();\n  return trimmed.length > 0 ? trimmed : undefined;\n}\n\nfunction asObject(value: unknown): Record<string, unknown> | undefined {\n  return typeof value === 'object' && value !== null && !Array.isArray(value)\n    ? (value as Record<string, unknown>)\n    : undefined;\n}\n\nfunction inferEdgeExtension(outputFormat: string): { ext: string; format: string } {\n  const normalized = outputFormat.toLowerCase();\n  if (normalized.includes('webm')) return { ext: 'webm', format: 'webm' };\n  if (normalized.includes('opus')) return { ext: 'opus', format: 'opus' };\n  if (normalized.includes('ogg')) return { ext: 'ogg', format: 'ogg' };\n  if (normalized.includes('wav') || normalized.includes('riff') || normalized.includes('pcm')) {\n    return { ext: 'wav', format: 'wav' };\n  }\n  return { ext: 'mp3', format: 'mp3' };\n}\n\nfunction normalizeConfig(rawConfig: Record<string, unknown>): EdgeTtsConfig {\n  const raw = asObject(rawConfig.edge) ?? rawConfig;\n  return {\n    voice: trimToUndefined(raw.voice ?? raw.voiceId) ?? DEFAULT_VOICE,\n    lang: trimToUndefined(raw.lang) ?? DEFAULT_LANG,\n    outputFormat: trimToUndefined(raw.outputFormat) ?? DEFAULT_OUTPUT_FORMAT,\n    pitch: trimToUndefined(raw.pitch),\n    rate: trimToUndefined(raw.rate),\n    volume: trimToUndefined(raw.volume),\n    proxy: trimToUndefined(raw.proxy),\n    enabled: raw.enabled !== false, // default-on\n  };\n}\n\nfunction readProviderConfig(config: SpeechProviderConfig): EdgeTtsConfig {\n  return {\n    voice: trimToUndefined(config.voice ?? config.voiceId) ?? DEFAULT_VOICE,\n    lang: trimToUndefined(config.lang) ?? DEFAULT_LANG,\n    outputFormat: trimToUndefined(config.outputFormat) ?? DEFAULT_OUTPUT_FORMAT,\n    pitch: trimToUndefined(config.pitch),\n    rate: trimToUndefined(config.rate),\n    volume: trimToUndefined(config.volume),\n    proxy: trimToUndefined(config.proxy),\n    enabled: config.enabled !== false,\n  };\n}\n\nfunction parseDirectiveTokenInternal(\n  ctx: SpeechDirectiveTokenParseContext,\n): SpeechDirectiveTokenParseResult {\n  switch (ctx.key) {\n    case 'voice':\n    case 'voice_id':\n    case 'voiceid':\n    case 'edge_voice':\n    case 'edgevoice':\n      if (!ctx.policy.allowVoice) {\n        return { handled: true };\n      }\n      return { handled: true, overrides: { voice: ctx.value } };\n    default:\n      return { handled: false };\n  }\n}\n\nasync function waitForNonEmptyFile(filePath: string, timeoutMs: number): Promise<void> {\n  const deadline = Date.now() + Math.min(Math.max(timeoutMs, 1000), 5000);\n  while (Date.now() < deadline) {\n    try {\n      if (statSync(filePath).size > 0) return;\n    } catch {\n      // The Edge package may resolve before the file is visible on slower filesystems.\n    }\n    await new Promise((resolve) => setTimeout(resolve, 50));\n  }\n}\n\nasync function synthesizeToBuffer(\n  text: string,\n  config: EdgeTtsConfig,\n  voiceOverride: string | undefined,\n  timeoutMs: number,\n): Promise<{ buffer: Buffer; outputFormat: string; ext: string }> {\n  // Lazy import — node-edge-tts loads ws transitively; skip cost for users that\n  // never enable the edge provider.\n  const { EdgeTTS } = await import('node-edge-tts');\n\n  const voice = voiceOverride ?? config.voice;\n  const tempDir = mkdtempSync(path.join(tmpdir(), 'tts-edge-'));\n  const { ext, format } = inferEdgeExtension(config.outputFormat);\n  const outputPath = path.join(tempDir, `speech-${Date.now()}.${ext}`);\n\n  try {\n    log.debug(\n      { voice, outputFormat: config.outputFormat, textLength: text.length },\n      'Calling Edge TTS',\n    );\n    const tts = new EdgeTTS({\n      voice,\n      lang: config.lang,\n      outputFormat: config.outputFormat,\n      proxy: config.proxy,\n      rate: config.rate,\n      pitch: config.pitch,\n      volume: config.volume,\n      timeout: timeoutMs,\n    });\n    await tts.ttsPromise(text, outputPath);\n    await waitForNonEmptyFile(outputPath, timeoutMs);\n    const buffer = readFileSync(outputPath);\n    if (buffer.length === 0) {\n      throw new Error(`Edge TTS produced an empty ${format} file for voice ${voice}`);\n    }\n    return { buffer, outputFormat: format, ext };\n  } finally {\n    try {\n      rmSync(tempDir, { recursive: true, force: true });\n    } catch (cleanupError) {\n      log.warn({ err: cleanupError, tempDir }, 'Failed to cleanup Edge TTS temp directory');\n    }\n  }\n}\n\nexport const edgeSpeechProvider: SpeechProviderPlugin = {\n  id: 'edge',\n  autoSelectOrder: 100,\n\n  resolveConfig: (ctx: SpeechProviderResolveConfigContext) => normalizeConfig(ctx.rawConfig),\n\n  isConfigured: (ctx: SpeechProviderConfiguredContext) => readProviderConfig(ctx.providerConfig).enabled,\n\n  parseDirectiveToken: parseDirectiveTokenInternal,\n\n  // Edge has hundreds of voices; we don't enumerate them statically. UI should\n  // call upstream Microsoft endpoint for the full catalog.\n  listVoices: async () => [{ id: DEFAULT_VOICE, name: DEFAULT_VOICE, locale: DEFAULT_LANG }],\n\n  synthesize: async (req: SpeechSynthesisRequest): Promise<SpeechSynthesisResult> => {\n    const config = readProviderConfig(req.providerConfig);\n    if (!config.enabled) {\n      throw new Error('Edge TTS is disabled in config (messages.tts.providers.edge.enabled = false)');\n    }\n    const overrides = req.providerOverrides ?? {};\n    const voiceOverride = trimToUndefined(overrides.voice ?? overrides.voiceId);\n    const { buffer, outputFormat, ext } = await synthesizeToBuffer(\n      req.text,\n      config,\n      voiceOverride,\n      req.timeoutMs,\n    );\n    return {\n      audioBuffer: buffer,\n      outputFormat,\n      fileExtension: ext,\n      voiceCompatible: outputFormat === 'opus' || outputFormat === 'ogg',\n    };\n  },\n\n  // True streaming would require a node-edge-tts upstream change. We expose a\n  // \"fake stream\" (single chunk) so callers don't fall through to speak-core's\n  // wrapBufferAsStream branch and can treat all providers uniformly.\n  synthesizeStream: async (req: SpeechSynthesisStreamRequest): Promise<SpeechSynthesisStreamResult> => {\n    const config = readProviderConfig(req.providerConfig);\n    if (!config.enabled) {\n      throw new Error('Edge TTS is disabled in config (messages.tts.providers.edge.enabled = false)');\n    }\n    const overrides = req.providerOverrides ?? {};\n    const voiceOverride = trimToUndefined(overrides.voice ?? overrides.voiceId);\n    const { buffer, outputFormat, ext } = await synthesizeToBuffer(\n      req.text,\n      config,\n      voiceOverride,\n      req.timeoutMs,\n    );\n    const audioStream = new ReadableStream<Uint8Array>({\n      start(controller) {\n        controller.enqueue(new Uint8Array(buffer));\n        controller.close();\n      },\n    });\n    return {\n      audioStream,\n      outputFormat,\n      fileExtension: ext,\n      voiceCompatible: outputFormat === 'opus' || outputFormat === 'ogg',\n    };\n  },\n};\n\nregisterSpeechProvider(edgeSpeechProvider);\n"],"mappings":";;;;;;;;;;;;;;;;;;;;aAkBwD;AAexD,MAAM,MAAM,aAAa,sBAAsB;AAE/C,MAAM,gBAAgB;AACtB,MAAM,eAAe;AACrB,MAAM,wBAAwB;AAa9B,SAAS,gBAAgB,OAAoC;AAC3D,KAAI,OAAO,UAAU,SACnB;CAEF,MAAM,UAAU,MAAM,MAAM;AAC5B,QAAO,QAAQ,SAAS,IAAI,UAAU,KAAA;;AAGxC,SAAS,SAAS,OAAqD;AACrE,QAAO,OAAO,UAAU,YAAY,UAAU,QAAQ,CAAC,MAAM,QAAQ,MAAM,GACtE,QACD,KAAA;;AAGN,SAAS,mBAAmB,cAAuD;CACjF,MAAM,aAAa,aAAa,aAAa;AAC7C,KAAI,WAAW,SAAS,OAAO,CAAE,QAAO;EAAE,KAAK;EAAQ,QAAQ;EAAQ;AACvE,KAAI,WAAW,SAAS,OAAO,CAAE,QAAO;EAAE,KAAK;EAAQ,QAAQ;EAAQ;AACvE,KAAI,WAAW,SAAS,MAAM,CAAE,QAAO;EAAE,KAAK;EAAO,QAAQ;EAAO;AACpE,KAAI,WAAW,SAAS,MAAM,IAAI,WAAW,SAAS,OAAO,IAAI,WAAW,SAAS,MAAM,CACzF,QAAO;EAAE,KAAK;EAAO,QAAQ;EAAO;AAEtC,QAAO;EAAE,KAAK;EAAO,QAAQ;EAAO;;AAGtC,SAAS,gBAAgB,WAAmD;CAC1E,MAAM,MAAM,SAAS,UAAU,KAAK,IAAI;AACxC,QAAO;EACL,OAAO,gBAAgB,IAAI,SAAS,IAAI,QAAQ,IAAI;EACpD,MAAM,gBAAgB,IAAI,KAAK,IAAI;EACnC,cAAc,gBAAgB,IAAI,aAAa,IAAI;EACnD,OAAO,gBAAgB,IAAI,MAAM;EACjC,MAAM,gBAAgB,IAAI,KAAK;EAC/B,QAAQ,gBAAgB,IAAI,OAAO;EACnC,OAAO,gBAAgB,IAAI,MAAM;EACjC,SAAS,IAAI,YAAY;EAC1B;;AAGH,SAAS,mBAAmB,QAA6C;AACvE,QAAO;EACL,OAAO,gBAAgB,OAAO,SAAS,OAAO,QAAQ,IAAI;EAC1D,MAAM,gBAAgB,OAAO,KAAK,IAAI;EACtC,cAAc,gBAAgB,OAAO,aAAa,IAAI;EACtD,OAAO,gBAAgB,OAAO,MAAM;EACpC,MAAM,gBAAgB,OAAO,KAAK;EAClC,QAAQ,gBAAgB,OAAO,OAAO;EACtC,OAAO,gBAAgB,OAAO,MAAM;EACpC,SAAS,OAAO,YAAY;EAC7B;;AAGH,SAAS,4BACP,KACiC;AACjC,SAAQ,IAAI,KAAZ;EACE,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK;AACH,OAAI,CAAC,IAAI,OAAO,WACd,QAAO,EAAE,SAAS,MAAM;AAE1B,UAAO;IAAE,SAAS;IAAM,WAAW,EAAE,OAAO,IAAI,OAAO;IAAE;EAC3D,QACE,QAAO,EAAE,SAAS,OAAO;;;AAI/B,eAAe,oBAAoB,UAAkB,WAAkC;CACrF,MAAM,WAAW,KAAK,KAAK,GAAG,KAAK,IAAI,KAAK,IAAI,WAAW,IAAK,EAAE,IAAK;AACvE,QAAO,KAAK,KAAK,GAAG,UAAU;AAC5B,MAAI;AACF,OAAI,SAAS,SAAS,CAAC,OAAO,EAAG;UAC3B;AAGR,QAAM,IAAI,SAAS,YAAY,WAAW,SAAS,GAAG,CAAC;;;AAI3D,eAAe,mBACb,MACA,QACA,eACA,WACgE;CAGhE,MAAM,EAAE,YAAY,MAAM,OAAO;CAEjC,MAAM,QAAQ,iBAAiB,OAAO;CACtC,MAAM,UAAU,YAAY,KAAK,KAAK,QAAQ,EAAE,YAAY,CAAC;CAC7D,MAAM,EAAE,KAAK,WAAW,mBAAmB,OAAO,aAAa;CAC/D,MAAM,aAAa,KAAK,KAAK,SAAS,UAAU,KAAK,KAAK,CAAC,GAAG,MAAM;AAEpE,KAAI;AACF,MAAI,MACF;GAAE;GAAO,cAAc,OAAO;GAAc,YAAY,KAAK;GAAQ,EACrE,mBACD;AAWD,QAAM,IAVU,QAAQ;GACtB;GACA,MAAM,OAAO;GACb,cAAc,OAAO;GACrB,OAAO,OAAO;GACd,MAAM,OAAO;GACb,OAAO,OAAO;GACd,QAAQ,OAAO;GACf,SAAS;GACV,CACQ,CAAC,WAAW,MAAM,WAAW;AACtC,QAAM,oBAAoB,YAAY,UAAU;EAChD,MAAM,SAAS,aAAa,WAAW;AACvC,MAAI,OAAO,WAAW,EACpB,OAAM,IAAI,MAAM,8BAA8B,OAAO,kBAAkB,QAAQ;AAEjF,SAAO;GAAE;GAAQ,cAAc;GAAQ;GAAK;WACpC;AACR,MAAI;AACF,UAAO,SAAS;IAAE,WAAW;IAAM,OAAO;IAAM,CAAC;WAC1C,cAAc;AACrB,OAAI,KAAK;IAAE,KAAK;IAAc;IAAS,EAAE,4CAA4C;;;;AAK3F,MAAa,qBAA2C;CACtD,IAAI;CACJ,iBAAiB;CAEjB,gBAAgB,QAA4C,gBAAgB,IAAI,UAAU;CAE1F,eAAe,QAAyC,mBAAmB,IAAI,eAAe,CAAC;CAE/F,qBAAqB;CAIrB,YAAY,YAAY,CAAC;EAAE,IAAI;EAAe,MAAM;EAAe,QAAQ;EAAc,CAAC;CAE1F,YAAY,OAAO,QAAgE;EACjF,MAAM,SAAS,mBAAmB,IAAI,eAAe;AACrD,MAAI,CAAC,OAAO,QACV,OAAM,IAAI,MAAM,+EAA+E;EAEjG,MAAM,YAAY,IAAI,qBAAqB,EAAE;EAC7C,MAAM,gBAAgB,gBAAgB,UAAU,SAAS,UAAU,QAAQ;EAC3E,MAAM,EAAE,QAAQ,cAAc,QAAQ,MAAM,mBAC1C,IAAI,MACJ,QACA,eACA,IAAI,UACL;AACD,SAAO;GACL,aAAa;GACb;GACA,eAAe;GACf,iBAAiB,iBAAiB,UAAU,iBAAiB;GAC9D;;CAMH,kBAAkB,OAAO,QAA4E;EACnG,MAAM,SAAS,mBAAmB,IAAI,eAAe;AACrD,MAAI,CAAC,OAAO,QACV,OAAM,IAAI,MAAM,+EAA+E;EAEjG,MAAM,YAAY,IAAI,qBAAqB,EAAE;EAC7C,MAAM,gBAAgB,gBAAgB,UAAU,SAAS,UAAU,QAAQ;EAC3E,MAAM,EAAE,QAAQ,cAAc,QAAQ,MAAM,mBAC1C,IAAI,MACJ,QACA,eACA,IAAI,UACL;AAOD,SAAO;GACL,aAAA,IAPsB,eAA2B,EACjD,MAAM,YAAY;AAChB,eAAW,QAAQ,IAAI,WAAW,OAAO,CAAC;AAC1C,eAAW,OAAO;MAErB,CAEY;GACX;GACA,eAAe;GACf,iBAAiB,iBAAiB,UAAU,iBAAiB;GAC9D;;CAEJ;AAED,uBAAuB,mBAAmB"}
+{"version":3,"file":"edge-speech.js","names":[],"sources":["../../../../../src/voice/tts/providers/edge-speech.ts"],"sourcesContent":["/**\n * Edge TTS provider — wraps the `node-edge-tts` package. Requires no API key\n * (Microsoft's Edge browser endpoint is unauthenticated).\n *\n * Implementation notes:\n *   - Temp-file dance (mkdtemp → ttsPromise → readFileSync → rm) is required\n *     because `node-edge-tts` does not expose a stream-to-buffer API.\n *   - `synthesizeStream` is implemented as \"synthesize + wrap as single-chunk\n *     stream\" so callers see uniform behavior. This is not a true network\n *     stream; it's buffered-then-wrapped.\n *   - `inferEdgeExtension` derives the file extension from Edge's outputFormat\n *     string.\n */\n\nimport { mkdtempSync, readFileSync, rmSync, statSync } from 'node:fs';\nimport { tmpdir } from 'node:os';\nimport path from 'node:path';\n\nimport { createLogger } from '../../../utils/logger.js';\nimport { registerSpeechProvider } from '../speech-registry.js';\nimport type {\n  SpeechDirectiveTokenParseContext,\n  SpeechDirectiveTokenParseResult,\n  SpeechProviderConfig,\n  SpeechProviderConfiguredContext,\n  SpeechProviderPlugin,\n  SpeechProviderResolveConfigContext,\n  SpeechSynthesisRequest,\n  SpeechSynthesisResult,\n  SpeechSynthesisStreamRequest,\n  SpeechSynthesisStreamResult,\n} from '../speech-provider-types.js';\n\nconst log = createLogger('SpeechProvider:Edge');\n\nconst DEFAULT_VOICE = 'zh-CN-XiaoxiaoNeural';\nconst DEFAULT_LANG = 'zh-CN';\nconst DEFAULT_OUTPUT_FORMAT = 'audio-24khz-48kbitrate-mono-mp3';\n\ninterface EdgeTtsConfig extends Record<string, unknown> {\n  voice: string;\n  lang: string;\n  outputFormat: string;\n  pitch?: string;\n  rate?: string;\n  volume?: string;\n  proxy?: string;\n  enabled: boolean;\n}\n\nfunction trimToUndefined(value: unknown): string | undefined {\n  if (typeof value !== 'string') {\n    return undefined;\n  }\n  const trimmed = value.trim();\n  return trimmed.length > 0 ? trimmed : undefined;\n}\n\nfunction asObject(value: unknown): Record<string, unknown> | undefined {\n  return typeof value === 'object' && value !== null && !Array.isArray(value)\n    ? (value as Record<string, unknown>)\n    : undefined;\n}\n\nfunction inferEdgeExtension(outputFormat: string): { ext: string; format: string } {\n  const normalized = outputFormat.toLowerCase();\n  if (normalized.includes('webm')) return { ext: 'webm', format: 'webm' };\n  if (normalized.includes('opus')) return { ext: 'opus', format: 'opus' };\n  if (normalized.includes('ogg')) return { ext: 'ogg', format: 'ogg' };\n  if (normalized.includes('wav') || normalized.includes('riff') || normalized.includes('pcm')) {\n    return { ext: 'wav', format: 'wav' };\n  }\n  return { ext: 'mp3', format: 'mp3' };\n}\n\nfunction normalizeConfig(rawConfig: Record<string, unknown>): EdgeTtsConfig {\n  const raw = asObject(rawConfig.edge) ?? rawConfig;\n  return {\n    voice: trimToUndefined(raw.voice ?? raw.voiceId) ?? DEFAULT_VOICE,\n    lang: trimToUndefined(raw.lang) ?? DEFAULT_LANG,\n    outputFormat: trimToUndefined(raw.outputFormat) ?? DEFAULT_OUTPUT_FORMAT,\n    pitch: trimToUndefined(raw.pitch),\n    rate: trimToUndefined(raw.rate),\n    volume: trimToUndefined(raw.volume),\n    proxy: trimToUndefined(raw.proxy),\n    enabled: raw.enabled !== false, // default-on\n  };\n}\n\nfunction readProviderConfig(config: SpeechProviderConfig): EdgeTtsConfig {\n  return {\n    voice: trimToUndefined(config.voice ?? config.voiceId) ?? DEFAULT_VOICE,\n    lang: trimToUndefined(config.lang) ?? DEFAULT_LANG,\n    outputFormat: trimToUndefined(config.outputFormat) ?? DEFAULT_OUTPUT_FORMAT,\n    pitch: trimToUndefined(config.pitch),\n    rate: trimToUndefined(config.rate),\n    volume: trimToUndefined(config.volume),\n    proxy: trimToUndefined(config.proxy),\n    enabled: config.enabled !== false,\n  };\n}\n\nfunction parseDirectiveTokenInternal(\n  ctx: SpeechDirectiveTokenParseContext,\n): SpeechDirectiveTokenParseResult {\n  switch (ctx.key) {\n    case 'voice':\n    case 'voice_id':\n    case 'voiceid':\n    case 'edge_voice':\n    case 'edgevoice':\n      if (!ctx.policy.allowVoice) {\n        return { handled: true };\n      }\n      return { handled: true, overrides: { voice: ctx.value } };\n    default:\n      return { handled: false };\n  }\n}\n\nasync function waitForNonEmptyFile(filePath: string, timeoutMs: number): Promise<void> {\n  const deadline = Date.now() + Math.min(Math.max(timeoutMs, 1000), 5000);\n  while (Date.now() < deadline) {\n    try {\n      if (statSync(filePath).size > 0) return;\n    } catch {\n      // The Edge package may resolve before the file is visible on slower filesystems.\n    }\n    await new Promise((resolve) => setTimeout(resolve, 50));\n  }\n}\n\nasync function synthesizeToBuffer(\n  text: string,\n  config: EdgeTtsConfig,\n  voiceOverride: string | undefined,\n  timeoutMs: number,\n): Promise<{ buffer: Buffer; outputFormat: string; ext: string }> {\n  // Lazy import — node-edge-tts loads ws transitively; skip cost for users that\n  // never enable the edge provider.\n  const { EdgeTTS } = await import('node-edge-tts');\n\n  const voice = voiceOverride ?? config.voice;\n  const tempDir = mkdtempSync(path.join(tmpdir(), 'tts-edge-'));\n  const { ext, format } = inferEdgeExtension(config.outputFormat);\n  const outputPath = path.join(tempDir, `speech-${Date.now()}.${ext}`);\n\n  try {\n    log.debug(\n      { voice, outputFormat: config.outputFormat, textLength: text.length },\n      'Calling Edge TTS',\n    );\n    const tts = new EdgeTTS({\n      voice,\n      lang: config.lang,\n      outputFormat: config.outputFormat,\n      proxy: config.proxy,\n      rate: config.rate,\n      pitch: config.pitch,\n      volume: config.volume,\n      timeout: timeoutMs,\n    });\n    await tts.ttsPromise(text, outputPath);\n    await waitForNonEmptyFile(outputPath, timeoutMs);\n    const buffer = readFileSync(outputPath);\n    if (buffer.length === 0) {\n      throw new Error(`Edge TTS produced an empty ${format} file for voice ${voice}`);\n    }\n    return { buffer, outputFormat: format, ext };\n  } finally {\n    try {\n      rmSync(tempDir, { recursive: true, force: true });\n    } catch (cleanupError) {\n      log.warn({ err: cleanupError, tempDir }, 'Failed to cleanup Edge TTS temp directory');\n    }\n  }\n}\n\nexport const edgeSpeechProvider: SpeechProviderPlugin = {\n  id: 'edge',\n  autoSelectOrder: 100,\n\n  resolveConfig: (ctx: SpeechProviderResolveConfigContext) => normalizeConfig(ctx.rawConfig),\n\n  isConfigured: (ctx: SpeechProviderConfiguredContext) => readProviderConfig(ctx.providerConfig).enabled,\n\n  parseDirectiveToken: parseDirectiveTokenInternal,\n\n  // Edge has hundreds of voices; we don't enumerate them statically. UI should\n  // call upstream Microsoft endpoint for the full catalog.\n  listVoices: async () => [{ id: DEFAULT_VOICE, name: DEFAULT_VOICE, locale: DEFAULT_LANG }],\n\n  synthesize: async (req: SpeechSynthesisRequest): Promise<SpeechSynthesisResult> => {\n    const config = readProviderConfig(req.providerConfig);\n    if (!config.enabled) {\n      throw new Error('Edge TTS is disabled in config (messages.tts.providers.edge.enabled = false)');\n    }\n    const overrides = req.providerOverrides ?? {};\n    const voiceOverride = trimToUndefined(overrides.voice ?? overrides.voiceId);\n    const { buffer, outputFormat, ext } = await synthesizeToBuffer(\n      req.text,\n      config,\n      voiceOverride,\n      req.timeoutMs,\n    );\n    return {\n      audioBuffer: buffer,\n      outputFormat,\n      fileExtension: ext,\n      voiceCompatible: outputFormat === 'opus' || outputFormat === 'ogg',\n    };\n  },\n\n  // True streaming would require a node-edge-tts upstream change. We expose a\n  // \"fake stream\" (single chunk) so callers don't fall through to speak-core's\n  // wrapBufferAsStream branch and can treat all providers uniformly.\n  synthesizeStream: async (req: SpeechSynthesisStreamRequest): Promise<SpeechSynthesisStreamResult> => {\n    const config = readProviderConfig(req.providerConfig);\n    if (!config.enabled) {\n      throw new Error('Edge TTS is disabled in config (messages.tts.providers.edge.enabled = false)');\n    }\n    const overrides = req.providerOverrides ?? {};\n    const voiceOverride = trimToUndefined(overrides.voice ?? overrides.voiceId);\n    const { buffer, outputFormat, ext } = await synthesizeToBuffer(\n      req.text,\n      config,\n      voiceOverride,\n      req.timeoutMs,\n    );\n    const audioStream = new ReadableStream<Uint8Array>({\n      start(controller) {\n        controller.enqueue(new Uint8Array(buffer));\n        controller.close();\n      },\n    });\n    return {\n      audioStream,\n      outputFormat,\n      fileExtension: ext,\n      voiceCompatible: outputFormat === 'opus' || outputFormat === 'ogg',\n    };\n  },\n};\n\nregisterSpeechProvider(edgeSpeechProvider);\n"],"mappings":";;;;;;;;;;;;;;;;;;;;aAkBwD;AAexD,MAAM,MAAM,aAAa,sBAAsB;AAE/C,MAAM,gBAAgB;AACtB,MAAM,eAAe;AACrB,MAAM,wBAAwB;AAa9B,SAAS,gBAAgB,OAAoC;AAC3D,KAAI,OAAO,UAAU,SACnB;CAEF,MAAM,UAAU,MAAM,MAAM;AAC5B,QAAO,QAAQ,SAAS,IAAI,UAAU,KAAA;;AAGxC,SAAS,SAAS,OAAqD;AACrE,QAAO,OAAO,UAAU,YAAY,UAAU,QAAQ,CAAC,MAAM,QAAQ,MAAM,GACtE,QACD,KAAA;;AAGN,SAAS,mBAAmB,cAAuD;CACjF,MAAM,aAAa,aAAa,aAAa;AAC7C,KAAI,WAAW,SAAS,OAAO,CAAE,QAAO;EAAE,KAAK;EAAQ,QAAQ;EAAQ;AACvE,KAAI,WAAW,SAAS,OAAO,CAAE,QAAO;EAAE,KAAK;EAAQ,QAAQ;EAAQ;AACvE,KAAI,WAAW,SAAS,MAAM,CAAE,QAAO;EAAE,KAAK;EAAO,QAAQ;EAAO;AACpE,KAAI,WAAW,SAAS,MAAM,IAAI,WAAW,SAAS,OAAO,IAAI,WAAW,SAAS,MAAM,CACzF,QAAO;EAAE,KAAK;EAAO,QAAQ;EAAO;AAEtC,QAAO;EAAE,KAAK;EAAO,QAAQ;EAAO;;AAGtC,SAAS,gBAAgB,WAAmD;CAC1E,MAAM,MAAM,SAAS,UAAU,KAAK,IAAI;AACxC,QAAO;EACL,OAAO,gBAAgB,IAAI,SAAS,IAAI,QAAQ,IAAI;EACpD,MAAM,gBAAgB,IAAI,KAAK,IAAI;EACnC,cAAc,gBAAgB,IAAI,aAAa,IAAI;EACnD,OAAO,gBAAgB,IAAI,MAAM;EACjC,MAAM,gBAAgB,IAAI,KAAK;EAC/B,QAAQ,gBAAgB,IAAI,OAAO;EACnC,OAAO,gBAAgB,IAAI,MAAM;EACjC,SAAS,IAAI,YAAY;EAC1B;;AAGH,SAAS,mBAAmB,QAA6C;AACvE,QAAO;EACL,OAAO,gBAAgB,OAAO,SAAS,OAAO,QAAQ,IAAI;EAC1D,MAAM,gBAAgB,OAAO,KAAK,IAAI;EACtC,cAAc,gBAAgB,OAAO,aAAa,IAAI;EACtD,OAAO,gBAAgB,OAAO,MAAM;EACpC,MAAM,gBAAgB,OAAO,KAAK;EAClC,QAAQ,gBAAgB,OAAO,OAAO;EACtC,OAAO,gBAAgB,OAAO,MAAM;EACpC,SAAS,OAAO,YAAY;EAC7B;;AAGH,SAAS,4BACP,KACiC;AACjC,SAAQ,IAAI,KAAZ;EACE,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK;AACH,OAAI,CAAC,IAAI,OAAO,WACd,QAAO,EAAE,SAAS,MAAM;AAE1B,UAAO;IAAE,SAAS;IAAM,WAAW,EAAE,OAAO,IAAI,OAAO;IAAE;EAC3D,QACE,QAAO,EAAE,SAAS,OAAO;;;AAI/B,eAAe,oBAAoB,UAAkB,WAAkC;CACrF,MAAM,WAAW,KAAK,KAAK,GAAG,KAAK,IAAI,KAAK,IAAI,WAAW,IAAK,EAAE,IAAK;AACvE,QAAO,KAAK,KAAK,GAAG,UAAU;AAC5B,MAAI;AACF,OAAI,SAAS,SAAS,CAAC,OAAO,EAAG;UAC3B;AAGR,QAAM,IAAI,SAAS,YAAY,WAAW,SAAS,GAAG,CAAC;;;AAI3D,eAAe,mBACb,MACA,QACA,eACA,WACgE;CAGhE,MAAM,EAAE,YAAY,MAAM,OAAO;CAEjC,MAAM,QAAQ,iBAAiB,OAAO;CACtC,MAAM,UAAU,YAAY,KAAK,KAAK,QAAQ,EAAE,YAAY,CAAC;CAC7D,MAAM,EAAE,KAAK,WAAW,mBAAmB,OAAO,aAAa;CAC/D,MAAM,aAAa,KAAK,KAAK,SAAS,UAAU,KAAK,KAAK,CAAC,GAAG,MAAM;AAEpE,KAAI;AACF,MAAI,MACF;GAAE;GAAO,cAAc,OAAO;GAAc,YAAY,KAAK;GAAQ,EACrE,mBACD;AAWD,QAAM,IAVU,QAAQ;GACtB;GACA,MAAM,OAAO;GACb,cAAc,OAAO;GACrB,OAAO,OAAO;GACd,MAAM,OAAO;GACb,OAAO,OAAO;GACd,QAAQ,OAAO;GACf,SAAS;GACV,CACQ,CAAC,WAAW,MAAM,WAAW;AACtC,QAAM,oBAAoB,YAAY,UAAU;EAChD,MAAM,SAAS,aAAa,WAAW;AACvC,MAAI,OAAO,WAAW,EACpB,OAAM,IAAI,MAAM,8BAA8B,OAAO,kBAAkB,QAAQ;AAEjF,SAAO;GAAE;GAAQ,cAAc;GAAQ;GAAK;WACpC;AACR,MAAI;AACF,UAAO,SAAS;IAAE,WAAW;IAAM,OAAO;IAAM,CAAC;WAC1C,cAAc;AACrB,OAAI,KAAK;IAAE,KAAK;IAAc;IAAS,EAAE,4CAA4C;;;;AAK3F,MAAa,qBAA2C;CACtD,IAAI;CACJ,iBAAiB;CAEjB,gBAAgB,QAA4C,gBAAgB,IAAI,UAAU;CAE1F,eAAe,QAAyC,mBAAmB,IAAI,eAAe,CAAC;CAE/F,qBAAqB;CAIrB,YAAY,YAAY,CAAC;EAAE,IAAI;EAAe,MAAM;EAAe,QAAQ;EAAc,CAAC;CAE1F,YAAY,OAAO,QAAgE;EACjF,MAAM,SAAS,mBAAmB,IAAI,eAAe;AACrD,MAAI,CAAC,OAAO,QACV,OAAM,IAAI,MAAM,+EAA+E;EAEjG,MAAM,YAAY,IAAI,qBAAqB,EAAE;EAC7C,MAAM,gBAAgB,gBAAgB,UAAU,SAAS,UAAU,QAAQ;EAC3E,MAAM,EAAE,QAAQ,cAAc,QAAQ,MAAM,mBAC1C,IAAI,MACJ,QACA,eACA,IAAI,UACL;AACD,SAAO;GACL,aAAa;GACb;GACA,eAAe;GACf,iBAAiB,iBAAiB,UAAU,iBAAiB;GAC9D;;CAMH,kBAAkB,OAAO,QAA4E;EACnG,MAAM,SAAS,mBAAmB,IAAI,eAAe;AACrD,MAAI,CAAC,OAAO,QACV,OAAM,IAAI,MAAM,+EAA+E;EAEjG,MAAM,YAAY,IAAI,qBAAqB,EAAE;EAC7C,MAAM,gBAAgB,gBAAgB,UAAU,SAAS,UAAU,QAAQ;EAC3E,MAAM,EAAE,QAAQ,cAAc,QAAQ,MAAM,mBAC1C,IAAI,MACJ,QACA,eACA,IAAI,UACL;AAOD,SAAO;GACL,aAAA,IAPsB,eAA2B,EACjD,MAAM,YAAY;AAChB,eAAW,QAAQ,IAAI,WAAW,OAAO,CAAC;AAC1C,eAAW,OAAO;MAErB,CAEY;GACX;GACA,eAAe;GACf,iBAAiB,iBAAiB,UAAU,iBAAiB;GAC9D;;CAEJ;AAED,uBAAuB,mBAAmB"}

package/dist/src/voice/tts/types.js

@@ -1,8 +1,8 @@
 //#region src/voice/tts/types.ts
 const DEFAULT_TTS_CONFIG = {
-	enabled: false,
+	enabled: true,
 	provider: "edge",
-	trigger: "always",
+	trigger: "inbound",
 	fallback: {
 		enabled: true,
 		order: [
@@ -13,7 +13,7 @@ const DEFAULT_TTS_CONFIG = {
 		]
 	},
 	maxTextLength: 512,
-	timeoutMs: 3e4,
+	timeoutMs: 6e4,
 	summarization: { enabled: true },
 	modelOverrides: {
 		enabled: true,

package/dist/src/voice/tts/types.js.map

@@ -1 +1 @@
-{"version":3,"file":"types.js","names":[],"sources":["../../../../src/voice/tts/types.ts"],"sourcesContent":["/**\n * TTS (Text-to-Speech) Types\n */\n\n/** Built-in provider ids shipped in core. Extensions add more at runtime. */\nexport type BuiltinTTSProvider = 'openai' | 'alibaba' | 'edge' | 'minimax' | 'tts-local-cli';\n\n/** Any registered SpeechProviderPlugin id (built-in or extension). */\nexport type TTSProvider = string;\n\nexport type TTSAutoMode = 'off' | 'always' | 'inbound' | 'tagged';\n\nexport interface TTSOptions {\n  /** Voice ID */\n  voice?: string;\n  /** Speed (0.25 - 4.0) */\n  speed?: number;\n  /** Output format */\n  format?: 'opus' | 'mp3' | 'wav';\n  /** Model ID (for provider-specific model selection) */\n  model?: string;\n  /** Request timeout in milliseconds */\n  timeoutMs?: number;\n}\n\nexport interface TTSResult {\n  /** Audio buffer */\n  audio: Buffer;\n  /** Format */\n  format: string;\n  /** Duration in seconds (if available) */\n  duration?: number;\n  /** Provider that generated the speech */\n  provider: string;\n}\n\nexport interface TTSProviderInterface {\n  /** Provider name */\n  name: string;\n  /** Convert text to speech */\n  speak(text: string, options?: TTSOptions): Promise<TTSResult>;\n  /** Check if provider is configured */\n  isConfigured(): boolean;\n}\n\nexport interface TTSModelOverrideConfig {\n  /** Enable model-provided overrides for TTS */\n  enabled?: boolean;\n  /** Allow model-provided TTS text blocks */\n  allowText?: boolean;\n  /** Allow model-provided provider override (default: false) */\n  allowProvider?: boolean;\n  /** Allow model-provided voice/voiceId override */\n  allowVoice?: boolean;\n  /** Allow model-provided modelId override */\n  allowModelId?: boolean;\n  /** Allow model-provided voice settings override */\n  allowVoiceSettings?: boolean;\n  /** Allow model-provided normalization or language overrides */\n  allowNormalization?: boolean;\n  /** Allow model-provided seed override */\n  allowSeed?: boolean;\n}\n\nexport interface TTSConfig {\n  enabled: boolean;\n  provider: TTSProvider;\n  /** Trigger mode: auto = reply with voice when user sends voice */\n  trigger: TTSAutoMode;\n  /** Fallback configuration */\n  fallback?: {\n    enabled: boolean;\n    order: TTSProvider[];\n  };\n  /** Maximum text length for TTS */\n  maxTextLength?: number;\n  /** API request timeout (ms) */\n  timeoutMs?: number;\n  /** Long-text summarization before TTS */\n  summarization?: TTSSummarizationConfig;\n  /** Allow model to override TTS parameters */\n  modelOverrides?: TTSModelOverrideConfig;\n  /** Provider settings map (`messages.tts.providers.<id>`). */\n  providers?: Record<string, Record<string, unknown>>;\n}\n\nexport interface TTSSummarizationConfig {\n  /** When true (default), long text is summarized via LLM before TTS */\n  enabled?: boolean;\n  targetLength?: number;\n  threshold?: number;\n  model?: string;\n}\n\nexport const DEFAULT_TTS_CONFIG: TTSConfig = {\n  enabled: false,\n  provider: 'edge',\n  trigger: 'always',\n  fallback: {\n    enabled: true,\n    order: ['openai', 'alibaba', 'minimax', 'edge'],\n  },\n  maxTextLength: 512, // Conservative default to accommodate all providers (Alibaba limit is 512)\n  timeoutMs: 30000,\n  summarization: {\n    enabled: true,\n  },\n  modelOverrides: {\n    enabled: true,\n    allowText: true,\n    allowProvider: false,\n    allowVoice: true,\n    allowModelId: true,\n    allowVoiceSettings: false,\n    allowNormalization: false,\n    allowSeed: false,\n  },\n  providers: {\n    alibaba: { model: 'qwen-tts', voice: 'Cherry' },\n    openai: { model: 'tts-1', voice: 'alloy' },\n    edge: {\n      enabled: true,\n      voice: 'en-US-MichelleNeural',\n      lang: 'en-US',\n      outputFormat: 'audio-24khz-48kbitrate-mono-mp3',\n    },\n    minimax: { model: 'speech-2.8-hd', voice: 'male-qn-qingse' },\n  },\n};\n\n/** TTS directive parse result */\nexport interface TtsDirectiveParseResult {\n  cleanedText: string;\n  ttsText?: string;\n  hasDirective: boolean;\n  overrides: TtsDirectiveOverrides;\n  warnings: string[];\n}\n\n/** TTS directive overrides */\nexport interface TtsDirectiveOverrides {\n  ttsText?: string;\n  provider?: TTSProvider;\n  openai?: {\n    voice?: string;\n    model?: string;\n  };\n  alibaba?: {\n    voice?: string;\n    model?: string;\n  };\n  edge?: {\n    voice?: string;\n  };\n  minimax?: {\n    voice?: string;\n    model?: string;\n  };\n}\n\n/** Provider attempt result (TTS fallback chain) */\nexport type ProviderAttemptOutcome = 'success' | 'skipped' | 'failed';\n\n/** Provider failure reason classification */\nexport type ProviderFailureReason =\n  | 'success'\n  | 'not_configured'\n  | 'timeout'\n  | 'provider_error'\n  | 'text_too_long'\n  | 'unknown';\n\n/** Single provider attempt in a TTS fallback chain */\nexport interface ProviderAttempt {\n  provider: string;\n  outcome: ProviderAttemptOutcome;\n  reasonCode: ProviderFailureReason;\n  latencyMs: number;\n  error?: string;\n}\n\n/** TTS result including fallback / preprocessing metadata */\nexport interface TTSResultWithTracking extends TTSResult {\n  attempts: ProviderAttempt[];\n  fallbackFrom?: string;\n  attemptedProviders: string[];\n  wasPreprocessed?: boolean;\n  ttsText?: string;\n  wasSummarized?: boolean;\n}\n\n/** Last TTS call diagnostics (memory-only) */\nexport interface TtsStatusEntry {\n  timestamp: number;\n  success: boolean;\n  provider?: string;\n  latencyMs?: number;\n  error?: string;\n  textLength?: number;\n  audioSize?: number;\n  audioFormat?: string;\n  usedFallback?: boolean;\n  wasSummarized?: boolean;\n}\n\nexport interface TtsRuntimeStatus {\n  lastAttempt?: TtsStatusEntry;\n  recentSuccessRate?: number;\n  totalCalls: number;\n  totalSuccesses: number;\n  totalFailures: number;\n}\n"],"mappings":";AA8FA,MAAa,qBAAgC;CAC3C,SAAS;CACT,UAAU;CACV,SAAS;CACT,UAAU;EACR,SAAS;EACT,OAAO;GAAC;GAAU;GAAW;GAAW;GAAO;EAChD;CACD,eAAe;CACf,WAAW;CACX,eAAe,EACb,SAAS,MACV;CACD,gBAAgB;EACd,SAAS;EACT,WAAW;EACX,eAAe;EACf,YAAY;EACZ,cAAc;EACd,oBAAoB;EACpB,oBAAoB;EACpB,WAAW;EACZ;CACD,WAAW;EACT,SAAS;GAAE,OAAO;GAAY,OAAO;GAAU;EAC/C,QAAQ;GAAE,OAAO;GAAS,OAAO;GAAS;EAC1C,MAAM;GACJ,SAAS;GACT,OAAO;GACP,MAAM;GACN,cAAc;GACf;EACD,SAAS;GAAE,OAAO;GAAiB,OAAO;GAAkB;EAC7D;CACF"}
+{"version":3,"file":"types.js","names":[],"sources":["../../../../src/voice/tts/types.ts"],"sourcesContent":["/**\n * TTS (Text-to-Speech) Types\n */\n\n/** Built-in provider ids shipped in core. Extensions add more at runtime. */\nexport type BuiltinTTSProvider = 'openai' | 'alibaba' | 'edge' | 'minimax' | 'tts-local-cli';\n\n/** Any registered SpeechProviderPlugin id (built-in or extension). */\nexport type TTSProvider = string;\n\nexport type TTSAutoMode = 'off' | 'always' | 'inbound' | 'tagged';\n\nexport interface TTSOptions {\n  /** Voice ID */\n  voice?: string;\n  /** Speed (0.25 - 4.0) */\n  speed?: number;\n  /** Output format */\n  format?: 'opus' | 'mp3' | 'wav';\n  /** Model ID (for provider-specific model selection) */\n  model?: string;\n  /** Request timeout in milliseconds */\n  timeoutMs?: number;\n}\n\nexport interface TTSResult {\n  /** Audio buffer */\n  audio: Buffer;\n  /** Format */\n  format: string;\n  /** Duration in seconds (if available) */\n  duration?: number;\n  /** Provider that generated the speech */\n  provider: string;\n}\n\nexport interface TTSProviderInterface {\n  /** Provider name */\n  name: string;\n  /** Convert text to speech */\n  speak(text: string, options?: TTSOptions): Promise<TTSResult>;\n  /** Check if provider is configured */\n  isConfigured(): boolean;\n}\n\nexport interface TTSModelOverrideConfig {\n  /** Enable model-provided overrides for TTS */\n  enabled?: boolean;\n  /** Allow model-provided TTS text blocks */\n  allowText?: boolean;\n  /** Allow model-provided provider override (default: false) */\n  allowProvider?: boolean;\n  /** Allow model-provided voice/voiceId override */\n  allowVoice?: boolean;\n  /** Allow model-provided modelId override */\n  allowModelId?: boolean;\n  /** Allow model-provided voice settings override */\n  allowVoiceSettings?: boolean;\n  /** Allow model-provided normalization or language overrides */\n  allowNormalization?: boolean;\n  /** Allow model-provided seed override */\n  allowSeed?: boolean;\n}\n\nexport interface TTSConfig {\n  enabled: boolean;\n  provider: TTSProvider;\n  /** Trigger mode: auto = reply with voice when user sends voice */\n  trigger: TTSAutoMode;\n  /** Fallback configuration */\n  fallback?: {\n    enabled: boolean;\n    order: TTSProvider[];\n  };\n  /** Maximum text length for TTS */\n  maxTextLength?: number;\n  /** API request timeout (ms) */\n  timeoutMs?: number;\n  /** Long-text summarization before TTS */\n  summarization?: TTSSummarizationConfig;\n  /** Allow model to override TTS parameters */\n  modelOverrides?: TTSModelOverrideConfig;\n  /** Provider settings map (`messages.tts.providers.<id>`). */\n  providers?: Record<string, Record<string, unknown>>;\n}\n\nexport interface TTSSummarizationConfig {\n  /** When true (default), long text is summarized via LLM before TTS */\n  enabled?: boolean;\n  targetLength?: number;\n  threshold?: number;\n  model?: string;\n}\n\nexport const DEFAULT_TTS_CONFIG: TTSConfig = {\n  enabled: true,\n  provider: 'edge',\n  trigger: 'inbound',\n  fallback: {\n    enabled: true,\n    order: ['openai', 'alibaba', 'minimax', 'edge'],\n  },\n  maxTextLength: 512, // Conservative default to accommodate all providers (Alibaba limit is 512)\n  timeoutMs: 60000,\n  summarization: {\n    enabled: true,\n  },\n  modelOverrides: {\n    enabled: true,\n    allowText: true,\n    allowProvider: false,\n    allowVoice: true,\n    allowModelId: true,\n    allowVoiceSettings: false,\n    allowNormalization: false,\n    allowSeed: false,\n  },\n  providers: {\n    alibaba: { model: 'qwen-tts', voice: 'Cherry' },\n    openai: { model: 'tts-1', voice: 'alloy' },\n    edge: {\n      enabled: true,\n      voice: 'en-US-MichelleNeural',\n      lang: 'en-US',\n      outputFormat: 'audio-24khz-48kbitrate-mono-mp3',\n    },\n    minimax: { model: 'speech-2.8-hd', voice: 'male-qn-qingse' },\n  },\n};\n\n/** TTS directive parse result */\nexport interface TtsDirectiveParseResult {\n  cleanedText: string;\n  ttsText?: string;\n  hasDirective: boolean;\n  overrides: TtsDirectiveOverrides;\n  warnings: string[];\n}\n\n/** TTS directive overrides */\nexport interface TtsDirectiveOverrides {\n  ttsText?: string;\n  provider?: TTSProvider;\n  openai?: {\n    voice?: string;\n    model?: string;\n  };\n  alibaba?: {\n    voice?: string;\n    model?: string;\n  };\n  edge?: {\n    voice?: string;\n  };\n  minimax?: {\n    voice?: string;\n    model?: string;\n  };\n}\n\n/** Provider attempt result (TTS fallback chain) */\nexport type ProviderAttemptOutcome = 'success' | 'skipped' | 'failed';\n\n/** Provider failure reason classification */\nexport type ProviderFailureReason =\n  | 'success'\n  | 'not_configured'\n  | 'timeout'\n  | 'provider_error'\n  | 'text_too_long'\n  | 'unknown';\n\n/** Single provider attempt in a TTS fallback chain */\nexport interface ProviderAttempt {\n  provider: string;\n  outcome: ProviderAttemptOutcome;\n  reasonCode: ProviderFailureReason;\n  latencyMs: number;\n  error?: string;\n}\n\n/** TTS result including fallback / preprocessing metadata */\nexport interface TTSResultWithTracking extends TTSResult {\n  attempts: ProviderAttempt[];\n  fallbackFrom?: string;\n  attemptedProviders: string[];\n  wasPreprocessed?: boolean;\n  ttsText?: string;\n  wasSummarized?: boolean;\n}\n\n/** Last TTS call diagnostics (memory-only) */\nexport interface TtsStatusEntry {\n  timestamp: number;\n  success: boolean;\n  provider?: string;\n  latencyMs?: number;\n  error?: string;\n  textLength?: number;\n  audioSize?: number;\n  audioFormat?: string;\n  usedFallback?: boolean;\n  wasSummarized?: boolean;\n}\n\nexport interface TtsRuntimeStatus {\n  lastAttempt?: TtsStatusEntry;\n  recentSuccessRate?: number;\n  totalCalls: number;\n  totalSuccesses: number;\n  totalFailures: number;\n}\n"],"mappings":";AA8FA,MAAa,qBAAgC;CAC3C,SAAS;CACT,UAAU;CACV,SAAS;CACT,UAAU;EACR,SAAS;EACT,OAAO;GAAC;GAAU;GAAW;GAAW;GAAO;EAChD;CACD,eAAe;CACf,WAAW;CACX,eAAe,EACb,SAAS,MACV;CACD,gBAAgB;EACd,SAAS;EACT,WAAW;EACX,eAAe;EACf,YAAY;EACZ,cAAc;EACd,oBAAoB;EACpB,oBAAoB;EACpB,WAAW;EACZ;CACD,WAAW;EACT,SAAS;GAAE,OAAO;GAAY,OAAO;GAAU;EAC/C,QAAQ;GAAE,OAAO;GAAS,OAAO;GAAS;EAC1C,MAAM;GACJ,SAAS;GACT,OAAO;GACP,MAAM;GACN,cAAc;GACf;EACD,SAAS;GAAE,OAAO;GAAiB,OAAO;GAAkB;EAC7D;CACF"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@xopcai/xopc",
-  "version": "0.0.143",
+  "version": "0.0.145",
   "description": "Local-first AI system that remembers context, coordinates AI, and sustains long-term progress.",
   "type": "module",
   "main": "dist/src/index.js",
@@ -73,6 +73,7 @@
     "@earendil-works/pi-tui": "0.80.10",
     "@grammyjs/runner": "^2.0.3",
     "@hono/node-server": "^2.0.4",
+    "@huggingface/transformers": "3.8.1",
     "@inquirer/prompts": "^8.5.0",
     "@larksuiteoapi/node-sdk": "^1.66.0",
     "@modelcontextprotocol/sdk": "1.29.0",
@@ -80,7 +81,7 @@
     "@sinclair/typebox": "^0.34.49",
     "@vscode/ripgrep": "^1.18.0",
     "abort-controller": "^3.0.0",
-    "adm-zip": "^0.5.17",
+    "adm-zip": "^0.6.0",
     "ajv": "^8.20.0",
     "chalk": "^5.6.2",
     "chokidar": "^5.0.0",
@@ -103,6 +104,7 @@
     "playwright-core": "^1.60.0",
     "proper-lockfile": "^4.1.2",
     "semver": "^7.8.1",
+    "sherpa-onnx-node": "1.13.4",
     "silk-wasm": "^3.7.1",
     "thread-stream": "^4.2.0",
     "undici": "^8.5.0",