@ozaiya/openclaw-channel 0.7.6 → 0.7.8

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -9,7 +9,7 @@ import fs from "node:fs/promises";
9
9
  import path from "node:path";
10
10
  import { registerPluginHttpRoute } from "openclaw/plugin-sdk/webhook-ingress";
11
11
  import { unwrapGroupKey, decryptMessage, encryptMessage, wrapGroupKey } from "./crypto.js";
12
- import { sendMessage, probeApi, fetchGroups, addMember, getUserPublicKeys, toggleReaction, editMessage, deleteMessage, pinMessage, unpinMessage, uploadFile, searchUsers, fetchLinkPreview, joinCall, leaveCall, } from "./api.js";
12
+ import { sendMessage, probeApi, fetchGroups, addMember, getUserPublicKeys, toggleReaction, editMessage, deleteMessage, pinMessage, unpinMessage, uploadFile, searchUsers, fetchLinkPreview, joinCall, leaveCall, startPhoneCall, endPhoneCall, updatePhoneCallStatus, } from "./api.js";
13
13
  import { botCreateDirect, botCreateGroup } from "./botActions.js";
14
14
  import { buildInlineKeyboardSummary, buildLinkPreviewSummary, normalizeMessageText, normalizeToolInlineKeyboardRows, } from "./richContent.js";
15
15
  import { normalizeCallbackQueryPayload } from "./callbackQuery.js";
@@ -19,6 +19,7 @@ import { getOzaiyaRuntime } from "./runtime.js";
19
19
  import { maybeTranscribeInboundAudio, prependVoiceTranscriptToAgentInput, resolveOzaiyaSttConfig, } from "./transcribeAudio.js";
20
20
  import { startGatewayMode } from "./gateway.js";
21
21
  import { VoiceCallSession } from "./voiceCall.js";
22
+ import { PhoneCallSession } from "./phoneCall.js";
22
23
  const DEFAULT_API_BASE_URL = "https://api.ozai.dev";
23
24
  const DEFAULT_WEBHOOK_PATH = "/ozaiya/webhook";
24
25
  const DEFAULT_ACCOUNT_ID = "default";
@@ -34,6 +35,10 @@ const unwrappedKeys = new Map();
34
35
  const groupToBotAccountId = new Map();
35
36
  // Active voice call sessions keyed by callId
36
37
  const activeVoiceCalls = new Map();
38
+ // Active phone call sessions keyed by phoneCallId
39
+ const activePhoneCalls = new Map();
40
+ // Map groupId → phoneCallId for active phone calls (for manual mode message routing)
41
+ const groupToActivePhoneCall = new Map();
37
42
  // Runtime state tracking
38
43
  const runtimeState = new Map();
39
44
  function recordState(accountId, patch) {
@@ -321,13 +326,12 @@ function resolveAttachmentMime(fileName, mime) {
321
326
  return "application/octet-stream";
322
327
  }
323
328
  /**
324
- * Synthesize text to speech via Deepgram Aura and return a WAV buffer.
329
+ * Synthesize text to speech via Deepgram Aura and return an MP3 buffer.
325
330
  * Returns null if TTS is unavailable or fails.
326
331
  */
327
- async function synthesizeSpeechToWav(text, deepgramApiKey, ttsModel, log) {
332
+ async function synthesizeSpeechToMp3(text, deepgramApiKey, ttsModel, log) {
328
333
  try {
329
- const sampleRate = 24000;
330
- const url = `https://api.deepgram.com/v1/speak?model=${encodeURIComponent(ttsModel)}&encoding=linear16&sample_rate=${sampleRate}`;
334
+ const url = `https://api.deepgram.com/v1/speak?model=${encodeURIComponent(ttsModel)}&encoding=mp3`;
331
335
  const res = await fetch(url, {
332
336
  method: "POST",
333
337
  headers: {
@@ -341,33 +345,264 @@ async function synthesizeSpeechToWav(text, deepgramApiKey, ttsModel, log) {
341
345
  log?.warn?.(`ozaiya: TTS error: ${res.status} ${res.statusText}`);
342
346
  return null;
343
347
  }
344
- const pcmBuffer = Buffer.from(await res.arrayBuffer());
345
- // Wrap raw PCM in a WAV container
346
- const numChannels = 1;
347
- const bitsPerSample = 16;
348
- const byteRate = sampleRate * numChannels * (bitsPerSample / 8);
349
- const blockAlign = numChannels * (bitsPerSample / 8);
350
- const header = Buffer.alloc(44);
351
- header.write("RIFF", 0);
352
- header.writeUInt32LE(36 + pcmBuffer.length, 4);
353
- header.write("WAVE", 8);
354
- header.write("fmt ", 12);
355
- header.writeUInt32LE(16, 16);
356
- header.writeUInt16LE(1, 20);
357
- header.writeUInt16LE(numChannels, 22);
358
- header.writeUInt32LE(sampleRate, 24);
359
- header.writeUInt32LE(byteRate, 28);
360
- header.writeUInt16LE(blockAlign, 32);
361
- header.writeUInt16LE(bitsPerSample, 34);
362
- header.write("data", 36);
363
- header.writeUInt32LE(pcmBuffer.length, 40);
364
- return Buffer.concat([header, pcmBuffer]);
348
+ return Buffer.from(await res.arrayBuffer());
365
349
  }
366
350
  catch (err) {
367
351
  log?.warn?.(`ozaiya: TTS synthesis failed: ${String(err)}`);
368
352
  return null;
369
353
  }
370
354
  }
355
+ /**
356
+ * Synthesize text to speech via Volcengine (火山引擎/豆包) TTS v3 streaming API.
357
+ * Uses X-Api-Key auth with the unidirectional streaming endpoint.
358
+ * Returns audio buffer or null on failure.
359
+ */
360
+ async function synthesizeSpeechVolcengineV3(text, apiKey, voice, encoding, speedRatio, resourceId, log) {
361
+ const headers = {
362
+ "Content-Type": "application/json",
363
+ "X-Api-Key": apiKey,
364
+ "X-Api-Resource-Id": resourceId,
365
+ "X-Api-Connect-Id": `tts-${Date.now()}-${Math.random().toString(36).slice(2, 8)}`,
366
+ };
367
+ const body = {
368
+ user: { uid: "ozaiya-bot" },
369
+ req_params: { text, speaker: voice, speed_ratio: speedRatio },
370
+ audio_params: { format: encoding, sample_rate: 24000 },
371
+ };
372
+ log?.info?.(`ozaiya: Volcengine TTS v3 request: speaker=${voice} encoding=${encoding} resourceId=${resourceId}`);
373
+ const res = await fetch("https://openspeech.bytedance.com/api/v3/tts/unidirectional", {
374
+ method: "POST",
375
+ headers,
376
+ body: JSON.stringify(body),
377
+ signal: AbortSignal.timeout(30_000),
378
+ });
379
+ if (!res.ok) {
380
+ const errText = await res.text().catch(() => "");
381
+ log?.warn?.(`ozaiya: Volcengine TTS v3 HTTP error: ${res.status} ${errText}`);
382
+ return null;
383
+ }
384
+ // Collect streaming response — line-delimited JSON, each line has { data: "<base64>" }
385
+ const responseText = await res.text();
386
+ const audioChunks = [];
387
+ for (const line of responseText.split("\n")) {
388
+ const trimmed = line.trim();
389
+ if (!trimmed)
390
+ continue;
391
+ try {
392
+ const chunk = JSON.parse(trimmed);
393
+ if (chunk.data) {
394
+ audioChunks.push(Buffer.from(chunk.data, "base64"));
395
+ }
396
+ // Non-success, non-final codes indicate errors
397
+ if (chunk.code && chunk.code !== 20000000 && chunk.code !== 0) {
398
+ log?.warn?.(`ozaiya: Volcengine TTS v3 chunk error: code=${chunk.code} message=${chunk.message}`);
399
+ }
400
+ }
401
+ catch {
402
+ // skip non-JSON lines
403
+ }
404
+ }
405
+ if (audioChunks.length === 0) {
406
+ log?.warn?.(`ozaiya: Volcengine TTS v3: no audio data in response (${responseText.length} bytes raw)`);
407
+ return null;
408
+ }
409
+ log?.info?.(`ozaiya: Volcengine TTS v3: collected ${audioChunks.length} chunks`);
410
+ return Buffer.concat(audioChunks);
411
+ }
412
+ /**
413
+ * Synthesize text to speech via Volcengine (火山引擎/豆包) TTS v1 API.
414
+ * Uses Bearer;token auth (legacy).
415
+ * Returns audio buffer or null on failure.
416
+ */
417
+ async function synthesizeSpeechVolcengineV1(text, config, voice, encoding, speedRatio, log) {
418
+ const cluster = config.cluster ?? "volcano_tts";
419
+ const reqId = `tts-${Date.now()}-${Math.random().toString(36).slice(2, 8)}`;
420
+ const payload = {
421
+ app: {
422
+ appid: config.appId ?? "",
423
+ token: config.accessToken ?? "",
424
+ cluster,
425
+ },
426
+ user: { uid: "ozaiya-bot" },
427
+ audio: {
428
+ voice_type: voice,
429
+ encoding,
430
+ speed_ratio: speedRatio,
431
+ },
432
+ request: {
433
+ reqid: reqId,
434
+ text,
435
+ operation: "query",
436
+ },
437
+ };
438
+ const headers = {
439
+ "Content-Type": "application/json",
440
+ Authorization: `Bearer;${config.accessToken}`,
441
+ };
442
+ const res = await fetch("https://openspeech.bytedance.com/api/v1/tts", {
443
+ method: "POST",
444
+ headers,
445
+ body: JSON.stringify(payload),
446
+ signal: AbortSignal.timeout(30_000),
447
+ });
448
+ if (!res.ok) {
449
+ log?.warn?.(`ozaiya: Volcengine TTS v1 HTTP error: ${res.status} ${res.statusText}`);
450
+ return null;
451
+ }
452
+ const result = await res.json();
453
+ if (result.code !== 3000 || !result.data) {
454
+ log?.warn?.(`ozaiya: Volcengine TTS v1 error: code=${result.code} message=${result.message}`);
455
+ return null;
456
+ }
457
+ return Buffer.from(result.data, "base64");
458
+ }
459
+ /**
460
+ * Synthesize text to speech via Volcengine AI Gateway (OpenAI-compatible endpoint).
461
+ * Uses the Ark API Key with Bearer token auth.
462
+ * Returns audio buffer or null on failure.
463
+ */
464
+ async function synthesizeSpeechVolcengineGateway(text, arkApiKey, voice, encoding, speedRatio, log) {
465
+ log?.info?.(`ozaiya: Volcengine AI Gateway TTS request: voice=${voice}`);
466
+ const res = await fetch("https://ai-gateway.vei.volces.com/v1/audio/speech", {
467
+ method: "POST",
468
+ headers: {
469
+ "Content-Type": "application/json",
470
+ Authorization: `Bearer ${arkApiKey}`,
471
+ },
472
+ body: JSON.stringify({
473
+ model: "doubao-tts",
474
+ input: text,
475
+ voice,
476
+ response_format: encoding,
477
+ speed: speedRatio,
478
+ }),
479
+ signal: AbortSignal.timeout(30_000),
480
+ });
481
+ if (!res.ok) {
482
+ const errText = await res.text().catch(() => "");
483
+ log?.warn?.(`ozaiya: Volcengine AI Gateway TTS HTTP error: ${res.status} ${errText}`);
484
+ return null;
485
+ }
486
+ // Response is raw audio binary
487
+ const arrayBuffer = await res.arrayBuffer();
488
+ if (arrayBuffer.byteLength === 0) {
489
+ log?.warn?.(`ozaiya: Volcengine AI Gateway TTS: empty response`);
490
+ return null;
491
+ }
492
+ log?.info?.(`ozaiya: Volcengine AI Gateway TTS: received ${arrayBuffer.byteLength} bytes`);
493
+ return Buffer.from(arrayBuffer);
494
+ }
495
+ /**
496
+ * Synthesize text to speech via Volcengine (火山引擎/豆包) TTS API.
497
+ * Priority:
498
+ * 1. arkApiKey → AI Gateway OpenAI-compatible endpoint (simplest, uses Ark API Key)
499
+ * 2. apiKey → v3 streaming openspeech endpoint (uses speech-specific API Key)
500
+ * 3. accessToken → v1 non-streaming openspeech endpoint (legacy)
501
+ * Returns audio buffer or null on failure.
502
+ */
503
+ async function synthesizeSpeechVolcengine(text, config, log) {
504
+ try {
505
+ const voice = config.voice ?? "zh_female_wanwanxiaohe_moon_bigtts";
506
+ const encoding = config.encoding ?? "mp3";
507
+ const speedRatio = config.speedRatio ?? 1.0;
508
+ if (config.arkApiKey) {
509
+ return await synthesizeSpeechVolcengineGateway(text, config.arkApiKey, voice, encoding, speedRatio, log);
510
+ }
511
+ else if (config.apiKey) {
512
+ const resourceId = config.resourceId ?? "volc.service_type.10029";
513
+ return await synthesizeSpeechVolcengineV3(text, config.apiKey, voice, encoding, speedRatio, resourceId, log);
514
+ }
515
+ else if (config.accessToken) {
516
+ return await synthesizeSpeechVolcengineV1(text, config, voice, encoding, speedRatio, log);
517
+ }
518
+ else {
519
+ log?.warn?.(`ozaiya: Volcengine TTS: no arkApiKey, apiKey, or accessToken configured`);
520
+ return null;
521
+ }
522
+ }
523
+ catch (err) {
524
+ log?.warn?.(`ozaiya: Volcengine TTS failed: ${String(err)}`);
525
+ return null;
526
+ }
527
+ }
528
+ /**
529
+ * Synthesize voice reply. Priority:
530
+ * 1. Volcengine TTS (if configured — supports Chinese)
531
+ * 2. OpenClaw runtime TTS (if available — uses configured provider)
532
+ * 3. Deepgram TTS (fallback — English only)
533
+ */
534
+ async function synthesizeVoiceReply(text,
535
+ // eslint-disable-next-line @typescript-eslint/no-explicit-any
536
+ ctx, voiceOverride) {
537
+ const ozaiyaChannelCfg = (ctx.cfg?.channels?.ozaiya ?? {});
538
+ const account = ctx.account;
539
+ const perBotVoice = account?.voiceConfig;
540
+ // 0. Per-bot voice config from gateway (highest priority)
541
+ if (perBotVoice?.provider === 'volcengine' && perBotVoice.appId && perBotVoice.accessToken) {
542
+ ctx.log?.info?.(`ozaiya: using per-bot Volcengine TTS`);
543
+ const volcCfg = {
544
+ appId: perBotVoice.appId,
545
+ accessToken: perBotVoice.accessToken,
546
+ ...(voiceOverride ? { voice: voiceOverride } : {}),
547
+ };
548
+ const buf = await synthesizeSpeechVolcengine(text, volcCfg, ctx.log);
549
+ if (buf) {
550
+ return { data: buf, ext: `.${volcCfg.encoding ?? "mp3"}` };
551
+ }
552
+ ctx.log?.warn?.(`ozaiya: per-bot Volcengine TTS failed, trying fallbacks`);
553
+ }
554
+ if (perBotVoice?.provider === 'deepgram' && perBotVoice.apiKey) {
555
+ ctx.log?.info?.(`ozaiya: using per-bot Deepgram TTS`);
556
+ const ttsModel = ozaiyaChannelCfg?.voiceCall?.tts?.model ?? "aura-asteria-en";
557
+ const mp3Buffer = await synthesizeSpeechToMp3(text, perBotVoice.apiKey, ttsModel, ctx.log);
558
+ if (mp3Buffer)
559
+ return { data: mp3Buffer, ext: ".mp3" };
560
+ ctx.log?.warn?.(`ozaiya: per-bot Deepgram TTS failed, trying fallbacks`);
561
+ }
562
+ // 1. Volcengine TTS (preferred for Chinese)
563
+ if (ozaiyaChannelCfg?.volcengineTts?.arkApiKey || ozaiyaChannelCfg?.volcengineTts?.apiKey || ozaiyaChannelCfg?.volcengineTts?.appId) {
564
+ const effectiveVolcCfg = voiceOverride
565
+ ? { ...ozaiyaChannelCfg.volcengineTts, voice: voiceOverride }
566
+ : ozaiyaChannelCfg.volcengineTts;
567
+ ctx.log?.info?.(`ozaiya: using Volcengine TTS (voice=${effectiveVolcCfg.voice ?? 'default'})`);
568
+ const buf = await synthesizeSpeechVolcengine(text, effectiveVolcCfg, ctx.log);
569
+ if (buf) {
570
+ const ext = `.${ozaiyaChannelCfg.volcengineTts.encoding ?? "mp3"}`;
571
+ return { data: buf, ext };
572
+ }
573
+ ctx.log?.warn?.(`ozaiya: Volcengine TTS failed, trying fallbacks`);
574
+ }
575
+ // 2. OpenClaw runtime TTS (edge, openai, elevenlabs etc.)
576
+ try {
577
+ const runtime = getOzaiyaRuntime();
578
+ if (runtime.tts?.textToSpeech) {
579
+ ctx.log?.info?.(`ozaiya: using runtime TTS`);
580
+ const result = await runtime.tts.textToSpeech({
581
+ text,
582
+ cfg: ctx.cfg,
583
+ channel: "ozaiya",
584
+ });
585
+ if (result.success && result.audioPath) {
586
+ const audioData = await fs.readFile(result.audioPath);
587
+ const ext = path.extname(result.audioPath) || ".mp3";
588
+ ctx.log?.info?.(`ozaiya: runtime TTS succeeded (provider=${result.provider}, ${result.latencyMs}ms, ${ext})`);
589
+ return { data: Buffer.from(audioData), ext };
590
+ }
591
+ ctx.log?.warn?.(`ozaiya: runtime TTS failed: ${result.error}`);
592
+ }
593
+ }
594
+ catch (err) {
595
+ ctx.log?.warn?.(`ozaiya: runtime TTS unavailable: ${String(err)}`);
596
+ }
597
+ // 3. Deepgram direct API (English fallback)
598
+ const ttsApiKey = ozaiyaChannelCfg?.voiceCall?.deepgramApiKey || process.env.DEEPGRAM_API_KEY || "";
599
+ const ttsModel = ozaiyaChannelCfg?.voiceCall?.tts?.model ?? "aura-asteria-en";
600
+ if (!ttsApiKey)
601
+ return null;
602
+ ctx.log?.info?.(`ozaiya: falling back to Deepgram TTS`);
603
+ const mp3Buffer = await synthesizeSpeechToMp3(text, ttsApiKey, ttsModel, ctx.log);
604
+ return mp3Buffer ? { data: mp3Buffer, ext: ".mp3" } : null;
605
+ }
371
606
  async function getGroupKeyOrThrow(account, groupId) {
372
607
  let groupKey = unwrappedKeys.get(groupId);
373
608
  if (!groupKey) {
@@ -700,6 +935,8 @@ export const ozaiyaPlugin = {
700
935
  createPinMessageTool(account),
701
936
  createSearchUsersTool(account),
702
937
  createListGroupsTool(account),
938
+ createMakePhoneCallTool(account, cfg),
939
+ createHangUpCallTool(account),
703
940
  ];
704
941
  }),
705
942
  gateway: {
@@ -712,7 +949,7 @@ export const ozaiyaPlugin = {
712
949
  ctx.log?.info(`[gateway] starting gateway mode`);
713
950
  recordState(account.accountId, { running: true, lastStartAt: Date.now() });
714
951
  const ozaiya = resolveConfig(ctx.cfg);
715
- const stateDir = ctx.runtime?.state?.resolveStateDir?.() ?? process.env.HOME ?? ".";
952
+ const stateDir = process.env.OPENCLAW_STATE_DIR ?? ctx.runtime?.state?.resolveStateDir?.() ?? process.env.HOME ?? ".";
716
953
  // Track per-bot unregister functions for hot-reload
717
954
  const botUnregisters = new Map();
718
955
  const startBotHandler = (botAccount) => {
@@ -840,6 +1077,12 @@ export const ozaiyaPlugin = {
840
1077
  session.disconnect().catch(() => { });
841
1078
  activeVoiceCalls.delete(callId);
842
1079
  }
1080
+ // Disconnect all active phone call sessions
1081
+ for (const [phoneCallId, session] of activePhoneCalls) {
1082
+ session.disconnect().catch(() => { });
1083
+ activePhoneCalls.delete(phoneCallId);
1084
+ }
1085
+ groupToActivePhoneCall.clear();
843
1086
  for (const id of gatewayBotAccounts.keys()) {
844
1087
  recordState(id, { running: false, lastStopAt: Date.now() });
845
1088
  }
@@ -1523,6 +1766,227 @@ function createListGroupsTool(account) {
1523
1766
  },
1524
1767
  };
1525
1768
  }
1769
+ function createMakePhoneCallTool(account, cfg) {
1770
+ return {
1771
+ label: "Make Phone Call",
1772
+ name: "make_phone_call",
1773
+ ownerOnly: false,
1774
+ description: "Initiate an outbound phone call to a PSTN number via SIP. " +
1775
+ "The call is recorded and the recording + transcript are saved to the chat. " +
1776
+ "In 'auto' mode, you speak directly to the callee via STT/TTS. " +
1777
+ "In 'manual' mode, the callee's speech is transcribed to chat for the user to respond.",
1778
+ parameters: {
1779
+ type: "object",
1780
+ properties: {
1781
+ groupId: {
1782
+ type: "string",
1783
+ description: "The group/DM where the call record will be posted.",
1784
+ },
1785
+ phoneNumber: {
1786
+ type: "string",
1787
+ description: "E.164 phone number to call (e.g. +8613800138000).",
1788
+ },
1789
+ mode: {
1790
+ type: "string",
1791
+ enum: ["auto", "manual"],
1792
+ description: "auto = AI handles conversation, manual = transcribe to chat for user to reply. Default: auto.",
1793
+ },
1794
+ purpose: {
1795
+ type: "string",
1796
+ description: "Brief note about why you're making this call (logged for reference).",
1797
+ },
1798
+ },
1799
+ required: ["groupId", "phoneNumber"],
1800
+ },
1801
+ execute: async (_toolCallId, rawArgs) => {
1802
+ const args = rawArgs;
1803
+ try {
1804
+ const mode = args.mode ?? "auto";
1805
+ const result = await startPhoneCall(account.apiBaseUrl, account.botToken, args.groupId, args.phoneNumber, mode, args.purpose);
1806
+ // Report connected status
1807
+ updatePhoneCallStatus(account.apiBaseUrl, account.botToken, result.phoneCallId, "connected").catch(() => { });
1808
+ const ozaiyaCfg = (cfg?.channels?.ozaiya ?? {});
1809
+ const voiceCallCfg = ozaiyaCfg.voiceCall ?? {};
1810
+ // Override per-bot voice config and pass volcengineTts from channel config
1811
+ const effectiveVoiceCallCfg = { ...voiceCallCfg };
1812
+ if (account.voiceConfig?.provider === "deepgram" && account.voiceConfig.apiKey) {
1813
+ effectiveVoiceCallCfg.deepgramApiKey = account.voiceConfig.apiKey;
1814
+ }
1815
+ if (!effectiveVoiceCallCfg.volcengineTts && ozaiyaCfg.volcengineTts) {
1816
+ effectiveVoiceCallCfg.volcengineTts = ozaiyaCfg.volcengineTts;
1817
+ }
1818
+ // Create PhoneCallSession
1819
+ const session = new PhoneCallSession({
1820
+ phoneCallId: result.phoneCallId,
1821
+ groupId: args.groupId,
1822
+ livekitToken: result.livekitToken,
1823
+ livekitUrl: result.livekitUrl,
1824
+ mode,
1825
+ voiceCallConfig: effectiveVoiceCallCfg,
1826
+ onTranscript: (text) => {
1827
+ if (mode === "auto") {
1828
+ // Auto mode: dispatch to agent, speak reply
1829
+ void handlePhoneAutoTranscript(text, session, account, cfg, args.groupId);
1830
+ }
1831
+ else {
1832
+ // Manual mode: post to chat as a message
1833
+ void postPhoneTranscriptToChat(text, account, args.groupId);
1834
+ }
1835
+ },
1836
+ onPhoneHangUp: () => {
1837
+ // Phone person hung up — end the call
1838
+ void cleanupPhoneCall(result.phoneCallId, session, account);
1839
+ },
1840
+ });
1841
+ activePhoneCalls.set(result.phoneCallId, session);
1842
+ groupToActivePhoneCall.set(args.groupId, result.phoneCallId);
1843
+ // Connect to LiveKit
1844
+ await session.connect();
1845
+ return {
1846
+ content: [{
1847
+ type: "text",
1848
+ text: `Phone call initiated (ID: ${result.phoneCallId}). ` +
1849
+ `Calling ${args.phoneNumber} in ${mode} mode. ` +
1850
+ `The call is being recorded. Use hang_up_call to end it.`,
1851
+ }],
1852
+ };
1853
+ }
1854
+ catch (err) {
1855
+ const msg = err instanceof Error ? err.message : String(err);
1856
+ return { content: [{ type: "text", text: `Error starting phone call: ${msg}` }] };
1857
+ }
1858
+ },
1859
+ };
1860
+ }
1861
+ function createHangUpCallTool(account) {
1862
+ return {
1863
+ label: "Hang Up Call",
1864
+ name: "hang_up_call",
1865
+ ownerOnly: false,
1866
+ description: "End an active outbound phone call.",
1867
+ parameters: {
1868
+ type: "object",
1869
+ properties: {
1870
+ phoneCallId: {
1871
+ type: "string",
1872
+ description: "The phone call ID returned by make_phone_call.",
1873
+ },
1874
+ },
1875
+ required: ["phoneCallId"],
1876
+ },
1877
+ execute: async (_toolCallId, rawArgs) => {
1878
+ const args = rawArgs;
1879
+ try {
1880
+ const session = activePhoneCalls.get(args.phoneCallId);
1881
+ if (session) {
1882
+ await cleanupPhoneCall(args.phoneCallId, session, account);
1883
+ }
1884
+ else {
1885
+ // No local session, still try API
1886
+ await endPhoneCall(account.apiBaseUrl, account.botToken, args.phoneCallId);
1887
+ }
1888
+ return {
1889
+ content: [{ type: "text", text: `Phone call ${args.phoneCallId} ended.` }],
1890
+ };
1891
+ }
1892
+ catch (err) {
1893
+ const msg = err instanceof Error ? err.message : String(err);
1894
+ return { content: [{ type: "text", text: `Error ending phone call: ${msg}` }] };
1895
+ }
1896
+ },
1897
+ };
1898
+ }
1899
+ /**
1900
+ * Handle phone transcription in auto mode: dispatch to agent, speak the reply.
1901
+ */
1902
+ async function handlePhoneAutoTranscript(text, session, account, cfg, groupId) {
1903
+ if (session.isDisposed)
1904
+ return;
1905
+ const runtime = getOzaiyaRuntime();
1906
+ const ch = runtime.channel;
1907
+ const route = ch.routing.resolveAgentRoute({
1908
+ cfg,
1909
+ channel: "ozaiya",
1910
+ accountId: account.accountId,
1911
+ peer: { kind: "group", id: groupId },
1912
+ });
1913
+ const ozaiyaCfg = (cfg?.channels?.ozaiya ?? {});
1914
+ const voicePrompt = ozaiyaCfg.voiceCall?.agentPrompt ??
1915
+ "[Phone Call] You are in a live phone call. Your response will be spoken aloud via TTS. " +
1916
+ "Rules: respond concisely (1-3 sentences), use natural spoken language, " +
1917
+ "never use markdown/code blocks/bullet lists/URLs/emojis. " +
1918
+ "Do not say \"sure\" or \"of course\" — just answer directly.";
1919
+ const bodyForAgent = `${voicePrompt}\n\n${text}`;
1920
+ const body = `[ozaiya] from: phone caller | at: ${new Date().toISOString()}\n---\n${text}`;
1921
+ const msgCtx = ch.reply.finalizeInboundContext({
1922
+ Body: body,
1923
+ BodyForAgent: bodyForAgent,
1924
+ RawBody: text,
1925
+ CommandBody: text,
1926
+ From: `ozaiya:group:${groupId}`,
1927
+ To: `ozaiya:group:${groupId}`,
1928
+ SessionKey: route.sessionKey,
1929
+ AccountId: route.accountId,
1930
+ ChatType: "group",
1931
+ ConversationLabel: `group:${groupId}`,
1932
+ GroupSubject: groupId,
1933
+ SenderId: "phone-caller",
1934
+ SenderName: "Phone Caller",
1935
+ Provider: "ozaiya",
1936
+ Surface: "ozaiya-phone",
1937
+ MessageSid: `phone-${Date.now()}`,
1938
+ Timestamp: Date.now(),
1939
+ NumFiles: 0,
1940
+ NumMedia: 0,
1941
+ HasFiles: false,
1942
+ CommandAuthorized: true,
1943
+ OriginatingChannel: "ozaiya",
1944
+ OriginatingTo: `ozaiya:group:${groupId}`,
1945
+ });
1946
+ await ch.reply.dispatchReplyWithBufferedBlockDispatcher({
1947
+ ctx: msgCtx,
1948
+ cfg,
1949
+ dispatcherOptions: {
1950
+ deliver: async (replyPayload, _info) => {
1951
+ const replyText = replyPayload.text;
1952
+ if (!replyText?.trim() || session.isDisposed)
1953
+ return;
1954
+ await session.speakReply(replyText);
1955
+ },
1956
+ onError: (err) => {
1957
+ // eslint-disable-next-line no-console
1958
+ console.warn(`ozaiya: phone call auto-reply error: ${String(err)}`);
1959
+ },
1960
+ },
1961
+ });
1962
+ }
1963
+ /**
1964
+ * Post phone person's transcribed speech to the chat group (manual mode).
1965
+ */
1966
+ async function postPhoneTranscriptToChat(text, account, groupId) {
1967
+ try {
1968
+ const groupKey = unwrappedKeys.get(groupId);
1969
+ if (!groupKey)
1970
+ return;
1971
+ const content = {
1972
+ text: `📞 对方说: ${text}`,
1973
+ };
1974
+ const encrypted = encryptMessage(content, groupKey);
1975
+ await sendMessage(account.apiBaseUrl, account.botToken, groupId, encrypted);
1976
+ }
1977
+ catch {
1978
+ // Fire-and-forget
1979
+ }
1980
+ }
1981
+ /**
1982
+ * Clean up a phone call session: disconnect, end call via API, remove from maps.
1983
+ */
1984
+ async function cleanupPhoneCall(phoneCallId, session, account) {
1985
+ activePhoneCalls.delete(phoneCallId);
1986
+ groupToActivePhoneCall.delete(session.groupId);
1987
+ await session.disconnect();
1988
+ await endPhoneCall(account.apiBaseUrl, account.botToken, phoneCallId, session.transcript).catch(() => { });
1989
+ }
1526
1990
  /**
1527
1991
  * Handle an inbound webhook message:
1528
1992
  * 1. Decrypt message content
@@ -1534,7 +1998,7 @@ function createListGroupsTool(account) {
1534
1998
  async function handleInboundMessage(payload,
1535
1999
  // eslint-disable-next-line @typescript-eslint/no-explicit-any
1536
2000
  ctx) {
1537
- const { groupId, groupType = "group", message, replyAllowed = true, voiceReply, voiceReplyPrompt, context } = payload;
2001
+ const { groupId, groupType = "group", message, replyAllowed = true, voiceReply, voiceReplyPrompt, voiceReplyVoice, context } = payload;
1538
2002
  const account = ctx.account;
1539
2003
  // Record inbound activity
1540
2004
  recordState(account.accountId, { lastInboundAt: Date.now() });
@@ -1561,6 +2025,20 @@ ctx) {
1561
2025
  return;
1562
2026
  }
1563
2027
  const messageText = normalizeMessageText(content.text);
2028
+ // Manual mode phone call: if this group has an active phone call in manual mode,
2029
+ // speak the user's text message to the phone via TTS instead of dispatching to agent.
2030
+ const activePhoneCallId = groupToActivePhoneCall.get(groupId);
2031
+ if (activePhoneCallId && messageText) {
2032
+ const phoneSession = activePhoneCalls.get(activePhoneCallId);
2033
+ if (phoneSession && phoneSession.mode === "manual" && !phoneSession.isDisposed) {
2034
+ // Don't intercept bot's own messages (transcriptions posted by the bot)
2035
+ if (message.senderId !== account.accountId) {
2036
+ ctx.log?.info?.(`ozaiya: routing user message to active phone call ${activePhoneCallId} (manual mode)`);
2037
+ phoneSession.speakReply(messageText).catch(() => { });
2038
+ return; // Don't dispatch to agent
2039
+ }
2040
+ }
2041
+ }
1564
2042
  const inboundAttachments = normalizeAttachments(content.files);
1565
2043
  const attachmentSummary = buildAttachmentSummary(inboundAttachments);
1566
2044
  const linkPreviewSummary = buildLinkPreviewSummary(content.linkPreviews);
@@ -1731,12 +2209,6 @@ ctx) {
1731
2209
  }).catch((err) => {
1732
2210
  ctx.log?.warn?.(`ozaiya: failed recording session: ${String(err)}`);
1733
2211
  });
1734
- // Resolve TTS config when voiceReply is active
1735
- const ozaiyaChannelCfg = voiceReply
1736
- ? (ctx.cfg?.channels?.ozaiya ?? {})
1737
- : null;
1738
- const ttsApiKey = ozaiyaChannelCfg?.voiceCall?.deepgramApiKey || process.env.DEEPGRAM_API_KEY || "";
1739
- const ttsModel = ozaiyaChannelCfg?.voiceCall?.tts?.model ?? "aura-asteria-en";
1740
2212
  // Dispatch to agent with buffered block dispatcher
1741
2213
  await ch.reply.dispatchReplyWithBufferedBlockDispatcher({
1742
2214
  ctx: msgCtx,
@@ -1744,26 +2216,27 @@ ctx) {
1744
2216
  dispatcherOptions: {
1745
2217
  deliver: async (replyPayload, _info) => {
1746
2218
  const replyText = replyPayload.text;
1747
- ctx.log?.info?.(`ozaiya: deliver called, text length=${replyText?.length ?? 0}, empty=${!replyText?.trim()}, voiceReply=${voiceReply}, ttsApiKey=${ttsApiKey ? 'set(' + ttsApiKey.length + ')' : 'empty'}`);
2219
+ ctx.log?.info?.(`ozaiya: deliver called, text length=${replyText?.length ?? 0}, empty=${!replyText?.trim()}, voiceReply=${voiceReply}, voiceReplyVoice=${voiceReplyVoice ?? 'none'}`);
1748
2220
  if (!replyText?.trim())
1749
2221
  return;
1750
2222
  // Voice reply: synthesize TTS audio and send as voice message
1751
- if (voiceReply && ttsApiKey) {
2223
+ if (voiceReply) {
1752
2224
  ctx.log?.info?.(`ozaiya: voice reply — synthesizing TTS for group ${groupId}`);
1753
- const wavBuffer = await synthesizeSpeechToWav(replyText, ttsApiKey, ttsModel, ctx.log);
1754
- if (wavBuffer) {
1755
- const fileInfo = await uploadFile(account.apiBaseUrl, account.botToken, groupId, "voice.wav", "audio/wav", wavBuffer);
2225
+ const audioBuffer = await synthesizeVoiceReply(replyText, ctx, voiceReplyVoice ?? undefined);
2226
+ if (audioBuffer) {
2227
+ const ext = audioBuffer.ext;
2228
+ const mime = ext === ".mp3" ? "audio/mpeg" : ext === ".opus" ? "audio/ogg" : ext === ".wav" ? "audio/wav" : "audio/mpeg";
2229
+ const fileInfo = await uploadFile(account.apiBaseUrl, account.botToken, groupId, `voice${ext}`, mime, audioBuffer.data);
1756
2230
  await sendEncryptedChatContent({
1757
2231
  account,
1758
2232
  groupId,
1759
- content: { files: [fileInfo] },
2233
+ content: { text: replyText, files: [fileInfo] },
1760
2234
  log: ctx.log,
1761
2235
  });
1762
- ctx.log?.info?.(`ozaiya: voice reply sent successfully`);
2236
+ ctx.log?.info?.(`ozaiya: voice reply sent successfully (${ext}, ${audioBuffer.data.length} bytes)`);
1763
2237
  return;
1764
2238
  }
1765
2239
  ctx.log?.warn?.(`ozaiya: TTS failed, falling back to text reply`);
1766
- // Notify the user that voice synthesis failed before sending text fallback
1767
2240
  await sendEncryptedChatContent({
1768
2241
  account,
1769
2242
  groupId,
@@ -1992,12 +2465,20 @@ ctx) {
1992
2465
  id: payload.groupId,
1993
2466
  },
1994
2467
  });
2468
+ // Override per-bot voice config and pass volcengineTts from channel config
2469
+ const effectiveVoiceCallCfg = { ...voiceCallCfg };
2470
+ if (account.voiceConfig?.provider === 'deepgram' && account.voiceConfig.apiKey) {
2471
+ effectiveVoiceCallCfg.deepgramApiKey = account.voiceConfig.apiKey;
2472
+ }
2473
+ if (!effectiveVoiceCallCfg.volcengineTts && ozaiyaCfg.volcengineTts) {
2474
+ effectiveVoiceCallCfg.volcengineTts = ozaiyaCfg.volcengineTts;
2475
+ }
1995
2476
  const session = new VoiceCallSession({
1996
2477
  callId: payload.callId,
1997
2478
  groupId: payload.groupId,
1998
2479
  livekitToken: joinResult.token,
1999
2480
  livekitUrl: joinResult.url,
2000
- voiceCallConfig: voiceCallCfg,
2481
+ voiceCallConfig: effectiveVoiceCallCfg,
2001
2482
  onTranscript: (text) => {
2002
2483
  // Dispatch transcript to agent and speak the reply
2003
2484
  void handleVoiceTranscript(text, session, route, account, ctx);