feat: establish v0.70 platform milestone

device-simulator-windows/src/global.d.ts

@@ -4,6 +4,10 @@ type SimulatorConfiguration = {
   deviceSecret: string;
 };
 
+type SimulatorRuntimeDefaults = SimulatorConfiguration & {
+  deviceSecretConfigured: boolean;
+};
+
 type RTCGrant = {
   app_id: string;
   room_id: string;
@@ -17,6 +21,12 @@ type CreatedConversation = {
   conversation_id: string;
   status: string;
   rtc: RTCGrant;
+  privacy: {
+    retention_mode: "none" | "transcript_only" | "transcript_and_audio";
+    record_near: boolean;
+    record_far: boolean;
+    upload_allowed: boolean;
+  };
   session: {
     engine_mode: string;
     heartbeat_interval_sec: number;
@@ -24,10 +34,23 @@ type CreatedConversation = {
   };
 };
 
+type DeviceFinalTurn = {
+  turn_key: string;
+  sequence_no: number;
+  speaker: "subject" | "agent";
+  start_at: string;
+  end_at: string;
+  final_text: string;
+  interrupted: boolean;
+};
+
 interface Window {
   deviceSimulator: {
+    getRuntimeDefaults(): Promise<SimulatorRuntimeDefaults>;
     configure(input: SimulatorConfiguration): Promise<{ apiURL: string; deviceID: string }>;
     probe(): Promise<Record<string, unknown>>;
+    probeRTCNetwork(): Promise<{ endpoints: Array<{ hostname: string; ok: boolean; error?: string }> }>;
+    getAutomationAudio(): Promise<{ data_url: string; source_name: string } | undefined>;
     createConversation(): Promise<CreatedConversation>;
     activateConversation(input: { conversationID: string; roomID: string; userID: string }): Promise<Record<string, unknown>>;
     getConversation(conversationID: string): Promise<{ status: string; action: string; failure_code?: string }>;
@@ -35,6 +58,22 @@ interface Window {
       actions?: Array<{ type: string; rtc?: RTCGrant }>;
       desired_config_revision?: number;
     }>;
-    stopConversation(input: { conversationID: string; reason: string; lastHeartbeatSequence: number }): Promise<Record<string, unknown>>;
+    submitTranscriptBatch(input: { conversationID: string; turns: DeviceFinalTurn[] }): Promise<{
+      conversation_id: string;
+      accepted_count: number;
+      duplicate_count: number;
+    }>;
+    stopConversation(input: { conversationID: string; reason: string; lastHeartbeatSequence: number; nearSegments: number; farSegments: number }): Promise<Record<string, unknown>>;
+    uploadConversationRecording(input: {
+      conversationID: string;
+      segments: Array<{
+        track: "near" | "far";
+        sequence_no: number;
+        codec: "wav_pcm_s16le";
+        content_type: "audio/wav";
+        duration_ms: number;
+        bytes: Uint8Array;
+      }>;
+    }): Promise<{ id: string; status: string; segments: Array<{ id: string; track: "near" | "far"; sequence_no: number; status: string }> }>;
   };
 }

device-simulator-windows/src/main.ts

@@ -1,10 +1,20 @@
 import VERTC, {
+	AudioSourceType,
+  AudioFrameData,
   IRTCEngine,
   LocalAudioPropertiesInfo,
   MediaType,
   RemoteAudioPropertiesInfo,
-  RoomProfileType,
+	RoomProfileType,
+	StreamIndex,
+  SUBTITLE_MODE,
+  SubtitleEvent,
+  SubtitleEventType,
+  SubtitleMessage,
 } from "@volcengine/rtc";
+import { PCMTrackRecorder, RecordedAudioSegment } from "./recording";
+
+VERTC.setLogConfig({ logLevel: "trace", LogfileSize: 20 });
 import "./style.css";
 
 const EngineEventsTypes = VERTC.events;
@@ -25,6 +35,19 @@ const state: {
   agentSpeaking: boolean;
   microphoneVolume: number;
   agentVolume: number;
+  subtitleStarted: boolean;
+  transcriptSequence: number;
+  transcriptFirstSeen: Map<string, number>;
+  collectedTurnKeys: Set<string>;
+  pendingTurns: DeviceFinalTurn[];
+  transcriptUpload?: Promise<void>;
+  subtitleStartWaiter?: { resolve: () => void; reject: (error: Error) => void };
+  automationAudio?: HTMLAudioElement;
+  automationAudioContext?: AudioContext;
+  automationAudioTrack?: MediaStreamTrack;
+  automationAudioObjectURL?: string;
+  nearRecorder?: PCMTrackRecorder;
+  farRecorder?: PCMTrackRecorder;
 } = {
   phase: "idle",
   heartbeatSequence: 0,
@@ -32,6 +55,11 @@ const state: {
   agentSpeaking: false,
   microphoneVolume: 0,
   agentVolume: 0,
+  subtitleStarted: false,
+  transcriptSequence: 0,
+  transcriptFirstSeen: new Map(),
+  collectedTurnKeys: new Set(),
+  pendingTurns: [],
 };
 
 const app = document.querySelector<HTMLElement>("#app")!;
@@ -47,7 +75,7 @@ app.innerHTML = `
     </header>
 
     <section class="status-grid" aria-label="链路状态">
-      <article class="status-card"><span>本地平台</span><strong id="apiStatus">未检测</strong><small id="apiDetail">localhost</small></article>
+      <article class="status-card"><span>云端平台</span><strong id="apiStatus">未检测</strong><small id="apiDetail">121.41.46.238</small></article>
       <article class="status-card"><span>RTC 房间</span><strong id="rtcStatus">未加入</strong><small id="rtcDetail">等待开始</small></article>
       <article class="status-card"><span>豆包智能体</span><strong id="agentStatus">未进房</strong><small id="agentDetail">等待 StartVoiceChat</small></article>
       <article class="status-card"><span>通话时长</span><strong id="duration">00:00</strong><small id="modeDetail">处理方式：—</small></article>
@@ -55,16 +83,16 @@ app.innerHTML = `
 
     <section class="content-grid">
       <article class="panel setup-panel">
-        <div class="panel-heading"><div><p class="eyebrow">LOCAL SETUP</p><h2>本地设备配置</h2></div><span class="safe">密钥仅驻留内存</span></div>
-        <label>本地平台地址<input id="apiURL" value="http://localhost:58080" autocomplete="off" /></label>
+        <div class="panel-heading"><div><p class="eyebrow">DEVICE SETUP</p><h2>测试设备配置</h2></div><span class="safe" id="credentialStatus">正在检查凭据</span></div>
+        <label>云端平台地址<input id="apiURL" value="https://121.41.46.238" autocomplete="off" /></label>
         <label>模拟设备 ID<input id="deviceID" value="10000000-0000-0000-0000-000000000511" autocomplete="off" /></label>
-        <label>设备签名密钥<input id="deviceSecret" type="password" value="local-simulated-device-secret-32-bytes-minimum" autocomplete="off" /></label>
+        <label>设备签名密钥<input id="deviceSecret" type="password" value="" autocomplete="off" /></label>
         <div class="row">
           <label>麦克风<select id="microphone"><option value="">默认麦克风</option></select></label>
           <label>扬声器<select id="speaker"><option value="">默认扬声器</option></select></label>
         </div>
         <button class="secondary" id="checkButton">检测平台和音频设备</button>
-        <p class="hint">RTC AppKey、火山 AK/SK 和豆包 Token 应配置在本地后端，本软件不会读取或保存它们。</p>
+        <p class="hint">RTC AppKey、火山 AK/SK 和豆包 Token 配置在云端后台，本软件不会读取或保存它们。</p>
       </article>
 
       <article class="panel call-panel">
@@ -83,14 +111,20 @@ app.innerHTML = `
           <div><dt>RTC 房间</dt><dd id="roomID">—</dd></div>
           <div><dt>终端用户</dt><dd id="userID">—</dd></div>
           <div><dt>智能体用户</dt><dd id="agentUserID">—</dd></div>
+          <div><dt>聊天语音</dt><dd id="recordingStatus">等待通话</dd></div>
         </dl>
       </article>
     </section>
 
     <section class="panel log-panel">
-      <div class="panel-heading"><div><p class="eyebrow">EVENT STREAM</p><h2>实时事件</h2></div><button class="text-button" id="clearLog">清空</button></div>
+      <div class="panel-heading"><div><p class="eyebrow">EVENT STREAM</p><h2>实时事件</h2></div><div><button class="text-button" id="downloadRtcLog">导出 RTC 日志</button><button class="text-button" id="clearLog">清空</button></div></div>
       <ol id="eventLog" aria-live="polite"><li><time>系统</time><span>模拟器已启动，等待检测本地环境。</span></li></ol>
     </section>
+
+    <section class="panel transcript-panel">
+      <div class="panel-heading"><div><p class="eyebrow">LIVE TRANSCRIPT</p><h2>实时字幕与上传</h2></div><span class="phase-badge" id="transcriptStatus">未启动</span></div>
+      <ol id="liveTranscript" aria-live="polite"><li class="transcript-empty">通话开始后，这里会显示双方的最终字幕。</li></ol>
+    </section>
   </section>
 `;
 
@@ -106,11 +140,25 @@ const controls = {
   stop: $<HTMLButtonElement>("stopButton"),
 };
 
+void window.deviceSimulator.getRuntimeDefaults().then((defaults) => {
+  controls.apiURL.value = defaults.apiURL;
+  controls.deviceID.value = defaults.deviceID;
+  controls.deviceSecret.value = defaults.deviceSecret;
+  if (defaults.deviceSecretConfigured) {
+    controls.deviceSecret.placeholder = "已从Windows安全凭据文件加载";
+    setText("credentialStatus", "凭据已安全加载");
+  } else {
+    controls.deviceSecret.placeholder = "请输入至少32字符的设备密钥";
+    setText("credentialStatus", "凭据未加载");
+  }
+}).catch((error) => log(`读取测试设备启动配置失败：${errorMessage(error)}`, "error"));
+
 function setText(id: string, value: string) {
   $(id).textContent = value;
 }
 
 function log(message: string, level: "info" | "success" | "error" = "info") {
+  console.log(`[${level}] ${message}`);
   const item = document.createElement("li");
   item.className = level;
   const time = document.createElement("time");
@@ -154,10 +202,16 @@ async function configureAndProbe() {
     if (!permission.audio) throw permission.audioExceptionError || new Error("未获得麦克风权限");
     await refreshAudioDevices();
     await window.deviceSimulator.probe();
+    const rtcNetwork = await window.deviceSimulator.probeRTCNetwork();
+    const blockedEndpoints = rtcNetwork.endpoints.filter((endpoint) => !endpoint.ok);
+    if (blockedEndpoints.length > 0) {
+      const details = blockedEndpoints.map((endpoint) => `${endpoint.hostname}（${endpoint.error || "连接失败"}）`).join("、");
+      throw new Error(`RTC 网络检测失败：${details}。请关闭代理/VPN、放行 TCP 443，或改用手机热点后重试`);
+    }
     setText("apiStatus", "连接正常");
     setText("apiDetail", controls.apiURL.value);
     setPhase("ready", "可以开始通话");
-    log(`环境检测通过，RTC Web SDK ${VERTC.getSdkVersion()}。`, "success");
+    log(`环境检测通过，RTC Web SDK ${VERTC.getSdkVersion()}，火山 RTC 接入域名可达。`, "success");
   } catch (error) {
     setText("apiStatus", "检测失败");
     setText("apiDetail", errorMessage(error));
@@ -184,6 +238,30 @@ function fillDeviceOptions(select: HTMLSelectElement, devices: MediaDeviceInfo[]
 }
 
 function registerRTCEvents(engine: IRTCEngine) {
+  engine.on(EngineEventsTypes.onSubtitleStateChanged, (event: SubtitleEvent) => {
+    if (event.event === SubtitleEventType.STARTED) {
+      state.subtitleStarted = true;
+      state.subtitleStartWaiter?.resolve();
+      setText("transcriptStatus", "识别中");
+      log("RTC 实时字幕已启动，双方最终话轮将自动上传。", "success");
+      return;
+    }
+    if (event.event === SubtitleEventType.STOPPED) {
+      state.subtitleStarted = false;
+      state.subtitleStartWaiter?.reject(new Error("RTC 实时字幕在启动期间被停止"));
+      setText("transcriptStatus", "已停止");
+      return;
+    }
+    if (event.event === SubtitleEventType.ERROR) {
+      state.subtitleStarted = false;
+      state.subtitleStartWaiter?.reject(new Error(`RTC 实时字幕启动失败：${event.errorCode || "UNKNOWN"}${event.errorMessage ? ` · ${event.errorMessage}` : ""}`));
+      setText("transcriptStatus", "识别失败");
+      log(`RTC 实时字幕错误：${event.errorCode || "UNKNOWN"}${event.errorMessage ? ` · ${event.errorMessage}` : ""}`, "error");
+    }
+  });
+  engine.on(EngineEventsTypes.onSubtitleMessageReceived, (messages: SubtitleMessage[]) => {
+    collectSubtitleMessages(messages);
+  });
   engine.on(EngineEventsTypes.onUserJoined, (event) => {
     const userID = event.userInfo.userId;
     log(`远端用户已进房：${userID}`, "success");
@@ -227,7 +305,17 @@ function registerRTCEvents(engine: IRTCEngine) {
     setPhase(state.phase, state.phase === "active" ? (state.agentSpeaking ? "豆包正在说话" : "请开始说话") : undefined);
   });
   engine.on(EngineEventsTypes.onConnectionStateChanged, (event) => {
-    log(`RTC 连接状态：${String(event.state)}`);
+    const names: Record<number, string> = {
+      0: "准备连接",
+      1: "已断开",
+      2: "连接中",
+      3: "已连接",
+      4: "重连中",
+      5: "已重连",
+      6: "连接丢失",
+    };
+    const level = event.state === 1 || event.state === 6 ? "error" : "info";
+    log(`RTC 连接状态：${String(event.state)}（${names[event.state] || "未知"}）${event.reason !== undefined ? `，原因 ${String(event.reason)}` : ""}`, level);
   });
   engine.on(EngineEventsTypes.onError, (event) => {
     log(`RTC 错误：${event.errorCode}`, "error");
@@ -237,6 +325,106 @@ function registerRTCEvents(engine: IRTCEngine) {
   });
 }
 
+function subtitleIdentity(message: SubtitleMessage) {
+  const conversation = state.conversation;
+  if (!conversation) return undefined;
+  if (message.userId === conversation.rtc.user_id) return { speaker: "subject" as const, label: "长者" };
+  if (message.userId === conversation.rtc.agent_user_id) return { speaker: "agent" as const, label: "豆包" };
+  return undefined;
+}
+
+function subtitleKey(message: SubtitleMessage) {
+  return `${message.userId}:${message.sequence}`;
+}
+
+function collectSubtitleMessages(messages: SubtitleMessage[]) {
+  const now = Date.now();
+  for (const message of messages) {
+    const identity = subtitleIdentity(message);
+    const text = message.text.trim();
+    if (!identity || !text || message.mode !== SUBTITLE_MODE.ASR_ONLY) continue;
+    const sourceKey = subtitleKey(message);
+    if (!state.transcriptFirstSeen.has(sourceKey)) state.transcriptFirstSeen.set(sourceKey, now);
+    if (!message.definite || state.collectedTurnKeys.has(sourceKey)) continue;
+
+    state.collectedTurnKeys.add(sourceKey);
+    state.transcriptSequence += 1;
+    const turn: DeviceFinalTurn = {
+      turn_key: `rtc-subtitle-${identity.speaker}-${message.sequence}`,
+      sequence_no: state.transcriptSequence,
+      speaker: identity.speaker,
+      start_at: new Date(state.transcriptFirstSeen.get(sourceKey) || now).toISOString(),
+      end_at: new Date(now).toISOString(),
+      final_text: text,
+      interrupted: false,
+    };
+    state.pendingTurns.push(turn);
+    renderTranscriptTurn(turn, identity.label);
+  }
+  void flushTranscriptTurns();
+}
+
+function renderTranscriptTurn(turn: DeviceFinalTurn, label: string) {
+  const list = $("liveTranscript");
+  list.querySelector(".transcript-empty")?.remove();
+  const item = document.createElement("li");
+  item.className = `transcript-turn transcript-turn--${turn.speaker}`;
+  const speaker = document.createElement("strong");
+  speaker.textContent = label;
+  const text = document.createElement("span");
+  text.textContent = turn.final_text;
+  item.append(speaker, text);
+  list.append(item);
+}
+
+async function flushTranscriptTurns() {
+  if (state.transcriptUpload) return state.transcriptUpload;
+  if (!state.conversation || state.pendingTurns.length === 0) return;
+  const conversationID = state.conversation.conversation_id;
+  const batch = state.pendingTurns.slice(0, 500);
+  state.transcriptUpload = (async () => {
+    try {
+      const result = await window.deviceSimulator.submitTranscriptBatch({ conversationID, turns: batch });
+      state.pendingTurns.splice(0, batch.length);
+      const uploaded = state.transcriptSequence - state.pendingTurns.length;
+      setText("transcriptStatus", state.subtitleStarted ? `识别中 · 已上传 ${uploaded}` : `已上传 ${uploaded}`);
+      log(`最终字幕已上传：新增 ${result.accepted_count} 条，重复 ${result.duplicate_count} 条。`, "success");
+    } catch (error) {
+      setText("transcriptStatus", `待重试 ${state.pendingTurns.length} 条`);
+      log(`最终字幕上传失败，将在心跳或结束通话时重试：${errorMessage(error)}`, "error");
+    } finally {
+      state.transcriptUpload = undefined;
+    }
+  })();
+  await state.transcriptUpload;
+  if (state.pendingTurns.length > 0 && state.pendingTurns.length !== batch.length) void flushTranscriptTurns();
+}
+
+async function flushTranscriptTurnsWithRetry(attempts: number) {
+  for (let attempt = 1; attempt <= attempts && state.pendingTurns.length > 0; attempt += 1) {
+    await flushTranscriptTurns();
+    if (state.pendingTurns.length === 0) return;
+    if (attempt < attempts) await new Promise((resolve) => window.setTimeout(resolve, attempt * 500));
+  }
+}
+
+async function startSubtitleAndWait(engine: IRTCEngine) {
+  let timeoutID: number | undefined;
+  const stateEvent = new Promise<void>((resolve, reject) => {
+    state.subtitleStartWaiter = { resolve, reject };
+  });
+  const timeout = new Promise<void>((_resolve, reject) => {
+    timeoutID = window.setTimeout(() => reject(new Error("等待 RTC 实时字幕启动超时")), 10_000);
+  });
+  try {
+    await engine.startSubtitle({ mode: SUBTITLE_MODE.ASR_ONLY });
+    await Promise.race([stateEvent, timeout]);
+  } finally {
+    if (timeoutID) window.clearTimeout(timeoutID);
+    state.subtitleStartWaiter = undefined;
+  }
+}
+
 function updateVolume(kind: "mic" | "agent", raw: number) {
   const value = Math.max(0, Math.min(100, Math.round(raw)));
   if (kind === "mic") state.microphoneVolume = value;
@@ -254,6 +442,15 @@ async function startConversation() {
     state.credentialExpiresAt = created.rtc.expires_at;
     state.heartbeatSequence = 0;
     state.agentJoined = false;
+    state.subtitleStarted = false;
+    state.transcriptSequence = 0;
+    state.transcriptFirstSeen.clear();
+    state.collectedTurnKeys.clear();
+    state.pendingTurns = [];
+    state.transcriptUpload = undefined;
+    state.nearRecorder = undefined;
+    state.farRecorder = undefined;
+    $("liveTranscript").innerHTML = '<li class="transcript-empty">正在等待双方的最终字幕。</li>';
     renderSession(created);
     log(`平台已分配 RTC 房间 ${created.rtc.room_id}。`, "success");
 
@@ -261,18 +458,27 @@ async function startConversation() {
     state.engine = engine;
     registerRTCEvents(engine);
     engine.enableAudioPropertiesReport({ interval: 300 });
-    await engine.startAudioCapture(controls.microphone.value || undefined);
+    configureRecording(engine, created);
+    const automatedAudio = await prepareAutomationAudio(engine);
+    if (!automatedAudio) await engine.startAudioCapture(controls.microphone.value || undefined);
     if (controls.speaker.value) await engine.setAudioPlaybackDevice(controls.speaker.value);
-    await engine.joinRoom(created.rtc.token, created.rtc.room_id, { userId: created.rtc.user_id }, {
-      isAutoPublish: true,
-      isAutoSubscribeAudio: true,
-      isAutoSubscribeVideo: false,
-      roomProfileType: RoomProfileType.chat,
-    });
+    log("正在连接火山 RTC 信令并加入房间。", "info");
+    await Promise.race([
+      engine.joinRoom(created.rtc.token, created.rtc.room_id, { userId: created.rtc.user_id }, {
+        isAutoPublish: true,
+        isAutoSubscribeAudio: true,
+        isAutoSubscribeVideo: false,
+        roomProfileType: RoomProfileType.chat,
+      }),
+      new Promise<never>((_resolve, reject) => window.setTimeout(() => reject(new Error("RTC 入房超过 15 秒。请关闭代理/VPN、允许本程序通过 Windows 防火墙，或改用手机热点后重试")), 15_000)),
+    ]);
     setText("rtcStatus", "已加入");
     setText("rtcDetail", created.rtc.room_id);
     log("终端已加入 RTC 房间并发布麦克风。", "success");
 
+    setText("transcriptStatus", "正在启动");
+    await startSubtitleAndWait(engine);
+
     await window.deviceSimulator.activateConversation({
       conversationID: created.conversation_id,
       roomID: created.rtc.room_id,
@@ -285,13 +491,96 @@ async function startConversation() {
     setPhase("active", "请开始说话");
     startTimers(created.session.heartbeat_interval_sec || 15);
     log("豆包语音任务已激活，可以开始双向聊天。", "success");
+    if (automatedAudio) {
+      await new Promise((resolve) => window.setTimeout(resolve, 4_000));
+      await state.automationAudioContext?.resume();
+      await state.automationAudio?.play();
+      log("自动验证音频已通过 RTC 自定义音轨发送。", "success");
+    }
   } catch (error) {
     setPhase("error", "启动失败");
     log(`启动通话失败：${errorMessage(error)}`, "error");
     await releaseRTC();
+    if (state.conversation) {
+      try {
+        await window.deviceSimulator.stopConversation({
+          conversationID: state.conversation.conversation_id,
+          reason: "windows_start_failed",
+          lastHeartbeatSequence: state.heartbeatSequence,
+          nearSegments: 0,
+          farSegments: 0,
+        });
+      } catch (cleanupError) {
+        log(`清理失败通话时发生错误：${errorMessage(cleanupError)}`, "error");
+      }
+    }
+    resetSession();
   }
 }
 
+function configureRecording(engine: IRTCEngine, conversation: CreatedConversation) {
+  const privacy = conversation.privacy;
+  if (!privacy.upload_allowed || (!privacy.record_near && !privacy.record_far)) {
+    setText("recordingStatus", "本次未授权录音");
+    log("本次通话的隐私策略未授权保存原录音，不采集音频帧。", "info");
+    return;
+  }
+  if (privacy.record_near) {
+    state.nearRecorder = new PCMTrackRecorder();
+    engine.setAudioFrameCallback(StreamIndex.STREAM_INDEX_MAIN, undefined, (frame: AudioFrameData) => state.nearRecorder?.append(frame), 4096);
+  }
+  if (privacy.record_far) {
+    state.farRecorder = new PCMTrackRecorder();
+    engine.setAudioFrameCallback(StreamIndex.STREAM_INDEX_MAIN, conversation.rtc.agent_user_id, (frame: AudioFrameData) => state.farRecorder?.append(frame), 4096);
+  }
+  setText("recordingStatus", "录音中");
+  log(`已按授权开始采集${privacy.record_near ? "长者近端" : ""}${privacy.record_near && privacy.record_far ? "和" : ""}${privacy.record_far ? "智能体远端" : ""} PCM 音频。`, "success");
+}
+
+function finishRecording(engine: IRTCEngine | undefined, conversation: CreatedConversation) {
+  try { engine?.setAudioFrameCallback(StreamIndex.STREAM_INDEX_MAIN, undefined, undefined); } catch { /* no-op */ }
+  try { engine?.setAudioFrameCallback(StreamIndex.STREAM_INDEX_MAIN, conversation.rtc.agent_user_id, undefined); } catch { /* no-op */ }
+  const segments = [
+    ...(state.nearRecorder?.finish("near") ?? []),
+    ...(state.farRecorder?.finish("far") ?? []),
+  ];
+  state.nearRecorder = undefined;
+  state.farRecorder = undefined;
+  return segments;
+}
+
+async function prepareAutomationAudio(engine: IRTCEngine) {
+  const asset = await window.deviceSimulator.getAutomationAudio();
+  if (!asset) return false;
+  const comma = asset.data_url.indexOf(",");
+  if (comma < 0) throw new Error("自动验证音频编码无效");
+  const metadata = asset.data_url.slice(0, comma);
+  const mime = metadata.match(/^data:([^;]+)/)?.[1] || "audio/wav";
+  const bytes = Uint8Array.from(atob(asset.data_url.slice(comma + 1)), (value) => value.charCodeAt(0));
+  const objectURL = URL.createObjectURL(new Blob([bytes], { type: mime }));
+  const audio = new Audio(objectURL);
+  audio.preload = "auto";
+  await new Promise<void>((resolve, reject) => {
+    audio.addEventListener("canplaythrough", () => resolve(), { once: true });
+    audio.addEventListener("error", () => reject(new Error(`无法加载自动验证音频：${asset.source_name}`)), { once: true });
+    audio.load();
+  });
+  const context = new AudioContext({ sampleRate: 48_000 });
+  const source = context.createMediaElementSource(audio);
+  const destination = context.createMediaStreamDestination();
+  source.connect(destination);
+  const track = destination.stream.getAudioTracks()[0];
+  if (!track) throw new Error("自动验证音频没有可发布的音轨");
+  await engine.setAudioSourceType(StreamIndex.STREAM_INDEX_MAIN, AudioSourceType.AUDIO_SOURCE_TYPE_EXTERNAL);
+  await engine.setExternalAudioTrack(StreamIndex.STREAM_INDEX_MAIN, track);
+  state.automationAudio = audio;
+  state.automationAudioContext = context;
+  state.automationAudioTrack = track;
+  state.automationAudioObjectURL = objectURL;
+  log(`已准备自动验证音频：${asset.source_name}。`, "success");
+  return true;
+}
+
 function renderSession(created: CreatedConversation) {
   setText("conversationID", created.conversation_id);
   setText("roomID", created.rtc.room_id);
@@ -340,6 +629,7 @@ async function sendHeartbeat() {
         log("RTC Token 已通过设备心跳完成续签。", "success");
       }
     }
+    await flushTranscriptTurns();
   } catch (error) {
     log(`设备心跳失败：${errorMessage(error)}`, "error");
   }
@@ -350,15 +640,59 @@ async function stopConversation(reason = "windows_user_hangup") {
   setPhase("stopping", "正在结束通话");
   stopTimers();
   const conversation = state.conversation;
+  let recordingResult = "等待通话";
   try {
+    if (state.subtitleStarted) state.engine?.stopSubtitle();
+    await new Promise((resolve) => window.setTimeout(resolve, 800));
+    await flushTranscriptTurnsWithRetry(3);
+    if (state.pendingTurns.length > 0) {
+      log(`仍有 ${state.pendingTurns.length} 条最终字幕未上传，通话暂不结束并继续自动重试。`, "error");
+      try {
+        setText("transcriptStatus", "正在恢复识别");
+        await state.engine?.startSubtitle({ mode: SUBTITLE_MODE.ASR_ONLY });
+      } catch (error) {
+        log(`恢复实时字幕失败：${errorMessage(error)}`, "error");
+      }
+      setPhase("active", "字幕待上传");
+      startTimers(conversation.session.heartbeat_interval_sec || 15);
+      return;
+    }
+    setText("recordingStatus", "正在封装录音");
+    const recordingSegments = finishRecording(state.engine, conversation);
     await releaseRTC();
     await window.deviceSimulator.stopConversation({
       conversationID: conversation.conversation_id,
       reason,
       lastHeartbeatSequence: state.heartbeatSequence,
+      nearSegments: recordingSegments.filter((segment) => segment.track === "near").length,
+      farSegments: recordingSegments.filter((segment) => segment.track === "far").length,
     });
     log("RTC 已离房，平台已提交 StopVoiceChat。", "success");
+    if (recordingSegments.length > 0) {
+      try {
+        setText("recordingStatus", "等待通话结束");
+        await waitUntilTerminal(conversation.conversation_id);
+        setText("recordingStatus", "正在上传");
+        const totalBytes = recordingSegments.reduce((sum, segment) => sum + segment.bytes.byteLength, 0);
+        log(`正在上传 ${recordingSegments.length} 个聊天语音分片（${formatRecordingBytes(totalBytes)}）。`, "info");
+        const batch = await uploadRecordingWithRetry(conversation.conversation_id, recordingSegments, 3);
+        setText("recordingStatus", "已上传并校验");
+        recordingResult = "本次已上传";
+        log(`聊天语音已上传并通过服务器校验，批次 ${batch.id}。`, "success");
+      } catch (uploadError) {
+        setText("recordingStatus", "上传失败");
+        recordingResult = "本次上传失败";
+        log(`聊天语音上传失败：${errorMessage(uploadError)}`, "error");
+      }
+    } else if (conversation.privacy.upload_allowed) {
+      setText("recordingStatus", "没有可上传音频");
+      recordingResult = "本次没有音频";
+      log("本次没有收到可封装的 PCM 音频帧，未创建录音上传批次。", "info");
+    } else {
+      recordingResult = "本次未授权录音";
+    }
     resetSession();
+    setText("recordingStatus", recordingResult);
     setPhase("ready", "可以再次通话");
   } catch (error) {
     log(`结束通话失败：${errorMessage(error)}`, "error");
@@ -367,10 +701,53 @@ async function stopConversation(reason = "windows_user_hangup") {
   }
 }
 
+async function waitUntilTerminal(conversationID: string) {
+  const deadline = Date.now() + 60_000;
+  while (Date.now() < deadline) {
+    const result = await window.deviceSimulator.getConversation(conversationID);
+    if (result.status === "COMPLETED" || result.status === "FAILED") return;
+    if (result.status === "EXPIRED") throw new Error("通话已过期，平台不接受录音上传");
+    await new Promise((resolve) => window.setTimeout(resolve, 750));
+  }
+  throw new Error("等待通话进入结束状态超时");
+}
+
+async function uploadRecordingWithRetry(conversationID: string, segments: RecordedAudioSegment[], attempts: number) {
+  let lastError: unknown;
+  for (let attempt = 1; attempt <= attempts; attempt += 1) {
+    try {
+      return await window.deviceSimulator.uploadConversationRecording({ conversationID, segments });
+    } catch (error) {
+      lastError = error;
+      if (attempt < attempts) {
+        log(`录音上传第 ${attempt} 次尝试失败，正在重试：${errorMessage(error)}`, "error");
+        await new Promise((resolve) => window.setTimeout(resolve, attempt * 1000));
+      }
+    }
+  }
+  throw lastError;
+}
+
+function formatRecordingBytes(value: number) {
+  if (value < 1024 * 1024) return `${(value / 1024).toFixed(1)} KiB`;
+  return `${(value / 1024 / 1024).toFixed(1)} MiB`;
+}
+
 async function releaseRTC() {
+	state.automationAudio?.pause();
+	state.automationAudioTrack?.stop();
+	if (state.automationAudioObjectURL) URL.revokeObjectURL(state.automationAudioObjectURL);
+	if (state.automationAudioContext) {
+		try { await state.automationAudioContext.close(); } catch { /* no-op */ }
+	}
+	state.automationAudio = undefined;
+	state.automationAudioTrack = undefined;
+	state.automationAudioContext = undefined;
+	state.automationAudioObjectURL = undefined;
   const engine = state.engine;
   state.engine = undefined;
   if (!engine) return;
+  try { engine.stopSubtitle(); } catch { /* no-op */ }
   try { await engine.unpublishStream(MediaType.AUDIO); } catch { /* no-op */ }
   try { await engine.leaveRoom(); } catch { /* no-op */ }
   try { await engine.stopAudioCapture(); } catch { /* no-op */ }
@@ -392,6 +769,15 @@ function resetSession() {
   state.startedAt = undefined;
   state.agentJoined = false;
   state.agentSpeaking = false;
+  state.subtitleStarted = false;
+  state.transcriptSequence = 0;
+  state.transcriptFirstSeen.clear();
+  state.collectedTurnKeys.clear();
+  state.pendingTurns = [];
+  state.transcriptUpload = undefined;
+  state.subtitleStartWaiter = undefined;
+  state.nearRecorder = undefined;
+  state.farRecorder = undefined;
   setText("rtcStatus", "未加入");
   setText("rtcDetail", "等待开始");
   setText("agentStatus", "未进房");
@@ -399,12 +785,19 @@ function resetSession() {
   setText("duration", "00:00");
   updateVolume("mic", 0);
   updateVolume("agent", 0);
+  setText("transcriptStatus", "未启动");
+  setText("recordingStatus", "等待通话");
+  $("liveTranscript").innerHTML = '<li class="transcript-empty">通话开始后，这里会显示双方的最终字幕。</li>';
 }
 
 controls.check.addEventListener("click", () => void configureAndProbe());
 controls.start.addEventListener("click", () => void startConversation());
 controls.stop.addEventListener("click", () => void stopConversation());
 $("clearLog").addEventListener("click", () => $("eventLog").replaceChildren());
+$("downloadRtcLog").addEventListener("click", () => {
+  VERTC.downloadLog();
+  log("已导出火山 RTC SDK 日志，请在 Windows“下载”目录查找最新日志文件。", "success");
+});
 window.addEventListener("beforeunload", () => {
   stopTimers();
   void releaseRTC();

device-simulator-windows/src/recording.ts

@@ -0,0 +1,128 @@
+export type AudioFrame = {
+  channels: Float32Array[];
+  channelCount: number;
+  sampleRate: number;
+};
+
+export type RecordedAudioSegment = {
+  track: "near" | "far";
+  sequence_no: number;
+  codec: "wav_pcm_s16le";
+  content_type: "audio/wav";
+  duration_ms: number;
+  bytes: Uint8Array;
+};
+
+const WAV_HEADER_SIZE = 44;
+const MAX_SEGMENT_DURATION_MS = 9 * 60 * 1000;
+const MAX_SEGMENT_SIZE = 64 * 1024 * 1024;
+
+export class PCMTrackRecorder {
+  private sampleRate = 0;
+  private chunks: Float32Array[] = [];
+  private sampleCount = 0;
+
+  append(frame: AudioFrame) {
+    if (!Number.isFinite(frame.sampleRate) || frame.sampleRate < 8_000 || frame.sampleRate > 192_000) return;
+    const availableChannels = frame.channels.slice(0, Math.max(1, frame.channelCount)).filter((channel) => channel.length > 0);
+    if (availableChannels.length === 0) return;
+    const frameLength = Math.min(...availableChannels.map((channel) => channel.length));
+    if (frameLength < 1) return;
+
+    const mono = new Float32Array(frameLength);
+    for (let index = 0; index < frameLength; index += 1) {
+      let value = 0;
+      for (const channel of availableChannels) value += channel[index];
+      mono[index] = value / availableChannels.length;
+    }
+
+    if (this.sampleRate === 0) this.sampleRate = frame.sampleRate;
+    const normalized = frame.sampleRate === this.sampleRate ? mono : resampleLinear(mono, frame.sampleRate, this.sampleRate);
+    if (normalized.length === 0) return;
+    this.chunks.push(normalized);
+    this.sampleCount += normalized.length;
+  }
+
+  finish(track: "near" | "far"): RecordedAudioSegment[] {
+    if (this.sampleRate === 0 || this.sampleCount === 0) return [];
+    const durationLimit = Math.floor(this.sampleRate * MAX_SEGMENT_DURATION_MS / 1000);
+    const sizeLimit = Math.floor((MAX_SEGMENT_SIZE - WAV_HEADER_SIZE) / 2);
+    const samplesPerSegment = Math.max(1, Math.min(durationLimit, sizeLimit));
+    const segments: RecordedAudioSegment[] = [];
+
+    for (let offset = 0, sequence = 0; offset < this.sampleCount; offset += samplesPerSegment, sequence += 1) {
+      const length = Math.min(samplesPerSegment, this.sampleCount - offset);
+      const durationMS = Math.floor(length * 1000 / this.sampleRate);
+      if (durationMS < 1) continue;
+      segments.push({
+        track,
+        sequence_no: sequence,
+        codec: "wav_pcm_s16le",
+        content_type: "audio/wav",
+        duration_ms: durationMS,
+        bytes: encodeMonoWAV(this.chunks, offset, length, this.sampleRate),
+      });
+    }
+
+    this.chunks = [];
+    this.sampleCount = 0;
+    this.sampleRate = 0;
+    return segments;
+  }
+}
+
+function resampleLinear(input: Float32Array, sourceRate: number, targetRate: number) {
+  const outputLength = Math.max(1, Math.round(input.length * targetRate / sourceRate));
+  const output = new Float32Array(outputLength);
+  const ratio = sourceRate / targetRate;
+  for (let index = 0; index < outputLength; index += 1) {
+    const position = index * ratio;
+    const left = Math.min(input.length - 1, Math.floor(position));
+    const right = Math.min(input.length - 1, left + 1);
+    const fraction = position - left;
+    output[index] = input[left] + (input[right] - input[left]) * fraction;
+  }
+  return output;
+}
+
+function encodeMonoWAV(chunks: Float32Array[], startSample: number, sampleCount: number, sampleRate: number) {
+  const dataSize = sampleCount * 2;
+  const bytes = new Uint8Array(WAV_HEADER_SIZE + dataSize);
+  const view = new DataView(bytes.buffer);
+  writeASCII(bytes, 0, "RIFF");
+  view.setUint32(4, bytes.length - 8, true);
+  writeASCII(bytes, 8, "WAVE");
+  writeASCII(bytes, 12, "fmt ");
+  view.setUint32(16, 16, true);
+  view.setUint16(20, 1, true);
+  view.setUint16(22, 1, true);
+  view.setUint32(24, sampleRate, true);
+  view.setUint32(28, sampleRate * 2, true);
+  view.setUint16(32, 2, true);
+  view.setUint16(34, 16, true);
+  writeASCII(bytes, 36, "data");
+  view.setUint32(40, dataSize, true);
+
+  let skipped = 0;
+  let written = 0;
+  for (const chunk of chunks) {
+    if (skipped + chunk.length <= startSample) {
+      skipped += chunk.length;
+      continue;
+    }
+    const chunkStart = Math.max(0, startSample - skipped);
+    const available = Math.min(chunk.length - chunkStart, sampleCount - written);
+    for (let index = 0; index < available; index += 1) {
+      const sample = Math.max(-1, Math.min(1, chunk[chunkStart + index]));
+      view.setInt16(WAV_HEADER_SIZE + (written + index) * 2, sample < 0 ? Math.round(sample * 32768) : Math.round(sample * 32767), true);
+    }
+    written += available;
+    skipped += chunk.length;
+    if (written === sampleCount) break;
+  }
+  return bytes;
+}
+
+function writeASCII(target: Uint8Array, offset: number, value: string) {
+  for (let index = 0; index < value.length; index += 1) target[offset + index] = value.charCodeAt(index);
+}

device-simulator-windows/src/style.css

@@ -68,6 +68,15 @@ input:focus, select:focus { border-color: #d6783f; box-shadow: 0 0 0 3px rgba(21
 #eventLog li.success span { color: #247252; }
 #eventLog li.error span { color: #ae3b2f; }
 
+.transcript-panel { margin-top: 16px; }
+.transcript-panel .panel-heading { margin-bottom: 14px; }
+#liveTranscript { display: grid; gap: 10px; margin: 0; padding: 0; max-height: 260px; overflow: auto; list-style: none; }
+.transcript-empty { color: #829198; font-size: 13px; }
+.transcript-turn { display: grid; gap: 4px; max-width: 80%; padding: 11px 14px; border-radius: 15px; background: #f1f6f4; }
+.transcript-turn--agent { justify-self: end; background: #edf3fb; }
+.transcript-turn strong { color: #315966; font-size: 12px; }
+.transcript-turn span { color: #263f49; font-size: 14px; line-height: 1.55; }
+
 @media (max-width: 1050px) {
   .status-grid { grid-template-columns: repeat(2, 1fr); }
   .content-grid { grid-template-columns: 1fr; }