feat(tts): 支持外部音频源配置并优化音频会话管理

添加对蓝牙等外部音频源的支持，扩展TTS初始化接口以接收外部音频源标志
优化音频会话配置逻辑，根据输入源类型和当前音频状态动态调整参数
增加音频流处理稳定性，改进WAV头探测和缓冲区管理

local_plugins/agent_service/ios/agent_service/Sources/agent_service/AgentServiceImpl.swift

@@ -342,12 +342,12 @@ class AgentServiceImpl: NSObject {
             sendError("初始化语音识别服务失败", code: "ASR_INIT_ERROR")
             return false
         }
-        
         guard let ttsSuccess = azureTtsHelper?.initialize(
             ttsAppId: "",
             ttsAppToken: azureSpeechKey,
             ttsResource: azureSpeechRegion,
-            language: primaryLanguage
+            language: primaryLanguage,
+            isExtaudioSource: BleService.shared.isConnected()
         ), ttsSuccess else {
             sendError("初始化语音合成服务失败", code: "TTS_INIT_ERROR")
             return false
@@ -619,7 +619,7 @@ class AgentServiceImpl: NSObject {
         return true
     }
     
-    func speakText(sessionid sessionid:String,_ text: String) -> Bool {
+    func speakText(sessionid: String, _ text: String) -> Bool {
         if !isInitialized {
             sendError("服务未初始化", code: "NOT_INITIALIZED")
             return false
@@ -630,6 +630,7 @@ class AgentServiceImpl: NSObject {
         }
         
         _ = azureTtsHelper?.setSpeechParams(rate: ttsRatePercent, pitch: 0, volume: 100)
+        _ = azureTtsHelper?.setAudioSourceType(isExtaudioSource: BleService.shared.isConnected())
         return azureTtsHelper?.speakOnce(sessionid:sessionid,text) ?? false
     }
 
@@ -1788,6 +1789,7 @@ class ChatApiStreamCallback: StreamCallback {
 
             responseBuilder += token
             if speakResponse && reply && broadcast{
+                _ = agentService.azureTtsHelper?.setAudioSourceType(isExtaudioSource: BleService.shared.isConnected())
                 try agentService.azureTtsHelper?.speakStream(sessionid:sessionid,token)
                 // 在开始流式TTS时立即停止气泡音
             }
@@ -1813,6 +1815,7 @@ class ChatApiStreamCallback: StreamCallback {
              }
             
             if speakResponse && reply && broadcast && sessionid == agentService.currsessionId{
+                _ = agentService.azureTtsHelper?.setAudioSourceType(isExtaudioSource: BleService.shared.isConnected())
                 agentService.azureTtsHelper?.flushStream(sessionid:sessionid)
             }
             
@@ -1864,6 +1867,7 @@ class ChatApiStreamCallback: StreamCallback {
                 "message": message
             ])
             if(code == 2001){
+                _ = agentService.azureTtsHelper?.setAudioSourceType(isExtaudioSource: BleService.shared.isConnected())
                 agentService.azureTtsHelper?.speakStream(sessionid: sessionid,agentService.insufficientIntegralText)
             }
             agentService.isAiStreaming = false
@@ -1907,6 +1911,7 @@ class ChatApiStreamCallback: StreamCallback {
          
             if (iscallingTool) {
                 os_log("收到函数调用: callingToolText:%{public}@", log: agentService.logger, type: .info, agentService.callingToolText)
+                _ = agentService.azureTtsHelper?.setAudioSourceType(isExtaudioSource: BleService.shared.isConnected())
                 agentService.azureTtsHelper?.speakStream(sessionid: sessionid,agentService.callingToolText)
                 iscallingTool = false
             }

local_plugins/azure_speech/ios/azure_speech/Sources/azure_speech/AzureTtsHelper.swift

@@ -48,6 +48,7 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
 
     // 自定义音频输出流
     private var customAudioOutputStream: SPXPushAudioOutputStream?
+    private var internalAudioOutputStream: SPXPushAudioOutputStream?
     private var useInternalPlayer = false
     private var micCapture: MicrophoneCapture!
     // 记录最后播放的文本
@@ -67,16 +68,51 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
     private var pcmFormat: AVAudioFormat?
     private var pcmPendingData = Data()
     private var hasStrippedWavHeader = false
+    private var wavHeaderProbeBuffer = Data()
     private var hasNotifiedPlaybackStartedForStream = false
     private var activeStreamSynthesisCount = 0
     private var scheduledBufferCount = 0
-    private let streamChunkBytes = 1600
+    private let streamChunkBytes = 3200
     private var suppressStreamPlayback = false
+    private var pushStreamChunkCount = 0
+    private var pushStreamCallbackCount = 0
+    private var synthEventAudioCount = 0
+    private var isExtaudioSource = false
+
+
 
     /**
      * 初始化语音合成服务
+     * - Parameters:
+     *   - ttsAppId: 服务应用ID（Azure 场景未使用，可为空）
+     *   - ttsAppToken: Azure 语音服务订阅密钥
+     *   - ttsResource: Azure 语音服务区域
+     *   - language: 语言代码，如 "zh-CN"
+     * - Returns: 是否初始化成功
+     * - Throws: 无（内部捕获 SDK 异常并记录日志）
      */
     public func initialize(ttsAppId: String, ttsAppToken: String, ttsResource: String, language: String) -> Bool {
+        return initialize(
+            ttsAppId: ttsAppId,
+            ttsAppToken: ttsAppToken,
+            ttsResource: ttsResource,
+            language: language,
+            isExtaudioSource: false
+        )
+    }
+
+    /**
+     * 初始化语音合成服务（支持外部音频源场景）
+     * - Parameters:
+     *   - ttsAppId: 服务应用ID（Azure 场景未使用，可为空）
+     *   - ttsAppToken: Azure 语音服务订阅密钥
+     *   - ttsResource: Azure 语音服务区域
+     *   - language: 语言代码，如 "zh-CN"
+     *   - isExtaudioSource: 是否为外部音频源（例如 BLE 输入）
+     * - Returns: 是否初始化成功
+     * - Throws: 无（内部捕获 SDK 异常并记录日志）
+     */
+    public func initialize(ttsAppId: String, ttsAppToken: String, ttsResource: String, language: String, isExtaudioSource: Bool) -> Bool {
         do {
             micCapture = MicrophoneCapture.shared
             // 创建语音配置
@@ -94,7 +130,7 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
 
             // 创建合成器
             recreateSynthesizer()
-
+            self.isExtaudioSource = isExtaudioSource
             // 设置初始化完成
             isInitialized = true
 
@@ -147,9 +183,10 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
      * @param sessionid 会话ID
      * @return 是否成功启动
      */
-    public func startspeak(sessionid sessionid: String) -> Bool {
+    public func startspeak(sessionid: String) -> Bool {
         self.sessionid = sessionid
         print("startspeak sessionid \(sessionid)")
+        // os_log("调用链: startspeak 设置session=%{public}@", log: log, type: .info, sessionid)
         // 重置当前会话的文本计数
         self.currentSessionTextCount = 0
         self.pendingTextCount = 0
@@ -163,7 +200,7 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
     /**
      * 单次合成并播放
      */
-    public func speakOnce(sessionid sessionid: String, _ text: String) -> Bool {
+    public func speakOnce(sessionid: String, _ text: String) -> Bool {
         if !isInitialized {
             os_log("语音合成未初始化", log: log, type: .error)
             notifyEvent(eventType: .error, params: [
@@ -173,20 +210,23 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
             return false
         }
 
-        if (self.sessionid != sessionid){
+        let effectiveSessionId = sessionid.isEmpty ? self.sessionid : sessionid
-            return false
+        if !sessionid.isEmpty {
+            self.sessionid = sessionid
         }
+        // os_log("调用链: speakOnce 接收 session=%{public}@ effective=%{public}@ speaking=%{public}@ pendingTasks=%{public}d", log: log, type: .info, sessionid, effectiveSessionId, speaking.description, pendingTasks.count)
 
         // 清理文本
         let cleanedText = cleanTextForTTS(text)
         if cleanedText.isEmpty {
+            // os_log("调用链: speakOnce 文本清理后为空 session=%{public}@", log: log, type: .info, effectiveSessionId)
             return false
         }
 
         // 异步处理
-        os_log("调用链: speakOnce 入队合成 session=%{public}@ 文本长度=%{public}d", log: log, type: .info, sessionid, cleanedText.count)
+        // os_log("调用链: speakOnce 入队合成 session=%{public}@ 文本长度=%{public}d", log: log, type: .info, effectiveSessionId, cleanedText.count)
         enqueueSynthesisTask {
-            self.performSynthesis(sessionid: sessionid, text: cleanedText)
+            self.performSynthesis(sessionid: effectiveSessionId, text: cleanedText)
         }
 
         return true
@@ -195,7 +235,7 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
     /**
      * 流式合成文本
      */
-    public func speakStream(sessionid sessionid: String, _ text: String) -> Bool {
+    public func speakStream(sessionid: String, _ text: String) -> Bool {
         if !isInitialized {
             notifyEvent(eventType: .error, params: [
                 "errorCode": "NOT_INITIALIZED",
@@ -203,12 +243,14 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
             ])
             return false
         }
-        if (self.sessionid != sessionid){
+        let effectiveSessionId = sessionid.isEmpty ? self.sessionid : sessionid
-            return false
+        if !sessionid.isEmpty {
-        }
             self.sessionid = sessionid
+        }
+        // os_log("调用链: speakStream 接收 session=%{public}@ effective=%{public}@ 输入长度=%{public}d bufferLen=%{public}d", log: log, type: .info, sessionid, effectiveSessionId, text.count, streamBuffer.count)
 
         if text.isEmpty {
+            // os_log("调用链: speakStream 空输入直接返回 session=%{public}@", log: log, type: .info, effectiveSessionId)
             return true
         }
 
@@ -218,11 +260,17 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
 
         // 清理文本并添加到缓冲区
         let cleanedText = cleanTextForTTS(text)
+        if cleanedText.isEmpty {
+            // os_log("调用链: speakStream 清理后为空 session=%{public}@ rawLen=%{public}d", log: log, type: .info, effectiveSessionId, text.count)
+            return true
+        }
         streamBuffer.append(cleanedText)
+        // os_log("调用链: speakStream 入缓冲 session=%{public}@ cleanedLen=%{public}d bufferLen=%{public}d", log: log, type: .info, effectiveSessionId, cleanedText.count, streamBuffer.count)
 
         // 防抖逻辑 (150ms)
         let currentTime = Date().timeIntervalSince1970
         if currentTime - lastSpeakTime < 0.15 {
+            // os_log("调用链: speakStream 防抖跳过 session=%{public}@ deltaMs=%{public}d", log: log, type: .info, effectiveSessionId, Int((currentTime - lastSpeakTime) * 1000.0))
             return true
         }
         lastSpeakTime = currentTime
@@ -248,17 +296,28 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
             streamBuffer = String(currentText[startIndex...])
 
             if !textToSpeak.isEmpty {
-                return speakOnce(sessionid: sessionid, textToSpeak)
+                // os_log("调用链: speakStream 触发合成 session=%{public}@ speakLen=%{public}d remainLen=%{public}d", log: log, type: .info, effectiveSessionId, textToSpeak.count, streamBuffer.count)
+                return speakOnce(sessionid: effectiveSessionId, textToSpeak)
             }
         }
 
+        let maxCharsBeforeForcedSpeak = 120
+        if currentText.count >= maxCharsBeforeForcedSpeak {
+            let textToSpeak = String(currentText.prefix(maxCharsBeforeForcedSpeak))
+            let startIndex = currentText.index(currentText.startIndex, offsetBy: maxCharsBeforeForcedSpeak)
+            streamBuffer = String(currentText[startIndex...])
+            // os_log("调用链: speakStream 强制分段合成 session=%{public}@ speakLen=%{public}d remainLen=%{public}d", log: log, type: .info, effectiveSessionId, textToSpeak.count, streamBuffer.count)
+            return speakOnce(sessionid: effectiveSessionId, textToSpeak)
+        }
+
+        // os_log("调用链: speakStream 未触发合成 session=%{public}@ bufferLen=%{public}d", log: log, type: .debug, effectiveSessionId, streamBuffer.count)
         return true
     }
 
     /**
      * 刷新并播放流式文本
      */
-    public func flushStream(sessionid sessionid: String) -> Bool {
+    public func flushStream(sessionid: String) -> Bool {
         if !isInitialized {
             notifyEvent(eventType: .error, params: [
                 "errorCode": "NOT_INITIALIZED",
@@ -266,6 +325,7 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
             ])
             return false
         }
+        // os_log("调用链: flushStream session=%{public}@ bufferLen=%{public}d", log: log, type: .info, sessionid, streamBuffer.count)
 
         let remainingText = streamBuffer
         streamBuffer = ""
@@ -372,6 +432,14 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
         }
     }
 
+    public func setAudioSourceType(isExtaudioSource: Bool) -> Bool {
+        self.isExtaudioSource = isExtaudioSource
+        if isInitialized {
+            safeConfigureAudioSessionForTTS()
+        }
+        return true
+    }
+
     /**
      * 设置音频输出设备
      */
@@ -411,6 +479,7 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
         taskLock.lock()
         pendingTasks.append(task)
         taskLock.unlock()
+        // os_log("调用链: enqueueSynthesisTask 入队 queued=%{public}d processing=%{public}@", log: log, type: .debug, queued, isProcessing.description)
 
         // 确保任务被处理
         processTasksIfNeeded()
@@ -430,6 +499,7 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
         isProcessing = true
         let task = pendingTasks.removeFirst()
         taskLock.unlock()
+        // os_log("调用链: processTasksIfNeeded 开始执行 remaining=%{public}d", log: log, type: .info, remaining)
         
         synthesisQueue.async {
             task()
@@ -444,6 +514,7 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
                 self.processTasksIfNeeded()
             } else {
                 self.taskLock.unlock()
+                // os_log("调用链: processTasksIfNeeded 队列清空", log: self.log, type: .info)
             }
         }
     }
@@ -471,10 +542,12 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
      */
     private func performSynthesis(sessionid:String,text: String) {
         if (self.sessionid != sessionid) {
+            // os_log("调用链: performSynthesis 丢弃(会话不匹配) current=%{public}@ task=%{public}@", log: log, type: .info, self.sessionid, sessionid)
             return
         }
         speaking = true
 //        lastSpokenText = text
+        // os_log("调用链: performSynthesis 准备合成 session=%{public}@ pushStream=%{public}@ suppress=%{public}@ textLen=%{public}d", log: log, type: .info, sessionid, isUsingPushStreamCapture.description, suppressStreamPlayback.description, text.count)
         if !applyAudioSessionForTTS() {
             speaking = false
             notifyEvent(eventType: .error, params: [
@@ -488,13 +561,14 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
         
         do {
             if (self.sessionid != sessionid) {
+                // os_log("调用链: performSynthesis 中止(会话变化) current=%{public}@ task=%{public}@", log: log, type: .info, self.sessionid, sessionid)
                 return
             }
-            os_log("开始合成: %{public}@", log: log, type: .debug, ssml)
+            // os_log("调用链: performSynthesis startSpeakingSsml session=%{public}@ ssmlLen=%{public}d", log: log, type: .info, sessionid, ssml.count)
             let result = try synthesizer?.startSpeakingSsml(ssml)
             
             if let result = result {
-                os_log("合成完成，结果: %{public}@", log: log, type: .debug, String(describing: result.reason))
+                // os_log("调用链: performSynthesis 返回 reason=%{public}@ session=%{public}@", log: log, type: .debug, String(describing: result.reason), sessionid)
             }
         } catch {
             speaking = false
@@ -514,9 +588,11 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
      */
     private func recreateSynthesizer() {
         do {
+            // os_log("调用链: recreateSynthesizer 开始 useInternalPlayer=%{public}@ customStream=%{public}@ initialized=%{public}@", log: log, type: .info, useInternalPlayer.description, (customAudioOutputStream != nil).description, isInitialized.description)
             // 强制释放旧的合成器
             synthesizer = nil
             isUsingPushStreamCapture = false
+            internalAudioOutputStream = nil
             
             // 创建音频配置
             let audioConfig: SPXAudioConfiguration?
@@ -535,13 +611,14 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
                         return UInt(data.count)
                     }, closeHandler: { [weak self] in
                         guard let self = self else { return }
-                        os_log("调用链: 输出流关闭 累计字节=%{public}d", log: self.log, type: .info, totalBytes)
+                        // os_log("调用链: 输出流关闭 累计字节=%{public}d", log: self.log, type: .info, totalBytes)
                         totalBytes = 0
                     })
                     // 如果创建失败，抛出以进入 catch
                     guard let nonNilStream = created else {
                         throw NSError(domain: "AzureTtsHelper", code: -1, userInfo: [NSLocalizedDescriptionKey: "创建推送输出流失败"])
                     }
+                    internalAudioOutputStream = nonNilStream
                     stream = nonNilStream
                 }
                 audioConfig = try SPXAudioConfiguration(streamOutput: stream)
@@ -556,6 +633,7 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
                 synthesizer = try SPXSpeechSynthesizer(speechConfig!)
             }
             setupEventListeners()
+            // os_log("调用链: recreateSynthesizer 完成 pushStream=%{public}@", log: log, type: .info, isUsingPushStreamCapture.description)
         } catch {
             os_log("重新创建合成器失败: %{public}@", log: log, type: .error, error.localizedDescription)
             notifyEvent(eventType: .error, params: [
@@ -578,16 +656,27 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
                     self.activeStreamSynthesisCount += 1
                     self.prepareStreamingQueueForNewSynthesisIfNeeded()
                 }
+                self.pushStreamChunkCount = 0
+                self.pushStreamCallbackCount = 0
+                self.synthEventAudioCount = 0
             }
-            os_log("调用链: 合成开始 session=%{public}@", log: self.log, type: .info, self.sessionid)
+            // os_log("调用链: 合成开始 session=%{public}@ pushStream=%{public}@ active=%{public}d", log: self.log, type: .info, self.sessionid, self.isUsingPushStreamCapture.description, self.activeStreamSynthesisCount)
             self.notifyEvent(eventType: .synthesisStarted)
         }
         
         synthesizer?.addSynthesizingEventHandler { [weak self] _, event in
             if let audioData = event.result.audioData {
                 guard let self = self else { return }
-                os_log("调用链: 合成中 接收音频片段=%{public}dB session=%{public}@", log: self.log, type: .debug, audioData.count, self.sessionid)
+                // os_log("调用链: 合成中 接收音频片段=%{public}dB session=%{public}@", log: self.log, type: .debug, audioData.count, self.sessionid)
-                if !self.isUsingPushStreamCapture {
+                if self.isUsingPushStreamCapture {
+                    if self.pushStreamCallbackCount == 0, !audioData.isEmpty {
+                        self.synthEventAudioCount += 1
+                        if self.synthEventAudioCount == 1 {
+                            // os_log("调用链: 推送流无回调，使用合成事件音频回退 session=%{public}@", log: self.log, type: .info, self.sessionid)
+                        }
+                        self.handleSynthesizedAudioChunk(audioData, source: "synthEvent")
+                    }
+                } else {
                     self.currentSynthesisBuffer.append(audioData)
                     self.notifyAudioData(audioData)
                 }
@@ -602,7 +691,7 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
             }
             // 将当前合成的音频加入播放队列
             if !self.currentSynthesisBuffer.isEmpty {
-                os_log("调用链: 合成完成 入队播放 数据长度=%{public}dB session=%{public}@", log: self.log, type: .info, self.currentSynthesisBuffer.count, self.sessionid)
+                // os_log("调用链: 合成完成 入队播放 数据长度=%{public}dB session=%{public}@", log: self.log, type: .info, self.currentSynthesisBuffer.count, self.sessionid)
                 self.enqueuePlaybackItem(self.currentSynthesisBuffer)
                 self.currentSynthesisBuffer = Data()
             }
@@ -613,6 +702,7 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
                     self.checkStreamPlaybackCompletedIfNeeded()
                 }
             }
+            // os_log("调用链: 合成完成 session=%{public}@ pushStream=%{public}@ pendingPcm=%{public}d scheduled=%{public}d", log: self.log, type: .info, self.sessionid, self.isUsingPushStreamCapture.description, self.pcmPendingData.count, self.scheduledBufferCount)
             // 减少待处理文本计数
             self.pendingTextCount = max(0, self.pendingTextCount - 1)
             // 通知合成完成
@@ -667,16 +757,50 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
      * - Throws: 无
      */
     private func handleSynthesizedAudioChunk(_ data: Data) {
-        if suppressStreamPlayback { return }
+        handleSynthesizedAudioChunk(data, source: "pushStream")
-        if sessionid.isEmpty { return }
+    }
+
+    private func handleSynthesizedAudioChunk(_ data: Data, source: String) {
+        if suppressStreamPlayback {
+            // os_log("调用链: 推送音频丢弃(suppress) bytes=%{public}d", log: log, type: .debug, data.count)
+            return
+        }
+        if sessionid.isEmpty, !speaking {
+            // os_log("调用链: 推送音频丢弃(无会话且未speaking) bytes=%{public}d", log: log, type: .debug, data.count)
+            return
+        }
         notifyAudioData(data)
-        guard isUsingPushStreamCapture else { return }
+        guard isUsingPushStreamCapture else {
+            // os_log("调用链: 推送音频忽略(未启用pushStream) bytes=%{public}d", log: log, type: .debug, data.count)
+            return
+        }
+        if source == "pushStream" {
+            pushStreamCallbackCount += 1
+        } else if source == "synthEvent" {
+            if pushStreamCallbackCount > 0 { return }
+        }
+        pushStreamChunkCount += 1
+        if pushStreamChunkCount == 1 {
+            // os_log("调用链: 推送音频首包 source=%{public}@ session=%{public}@ bytes=%{public}d speaking=%{public}@", log: log, type: .info, source, sessionid, data.count, speaking.description)
+        }
         audioPlaybackQueue.async {
-            if self.sessionid.isEmpty { return }
+            if self.sessionid.isEmpty, !self.speaking {
+                // os_log("调用链: 推送音频异步丢弃(无会话且未speaking) bytes=%{public}d", log: self.log, type: .debug, data.count)
+                return
+            }
             let pcm = self.stripWavHeaderIfNeeded(data)
             if !pcm.isEmpty {
                 self.pcmPendingData.append(pcm)
+                if self.scheduledBufferCount == 0 {
+                    // os_log("调用链: 推送音频入缓冲 pcm=%{public}dB pendingPcm=%{public}d", log: self.log, type: .debug, pcm.count, self.pcmPendingData.count)
+                }
                 self.scheduleAvailablePcmBuffers()
+            } else {
+                if self.pushStreamChunkCount <= 3 {
+                    // os_log("调用链: 推送音频等待WAV头 session=%{public}@ probe=%{public}dB in=%{public}dB", log: self.log, type: .info, self.sessionid, self.wavHeaderProbeBuffer.count, data.count)
+                } else {
+                    // os_log("调用链: 推送音频等待WAV头 probe=%{public}dB in=%{public}dB", log: self.log, type: .debug, self.wavHeaderProbeBuffer.count, data.count)
+                }
             }
         }
     }
@@ -689,6 +813,7 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
     private func prepareStreamingQueueForNewSynthesisIfNeeded() {
         if pcmPendingData.isEmpty, scheduledBufferCount == 0, playerNode?.isPlaying != true {
             hasStrippedWavHeader = false
+            wavHeaderProbeBuffer.removeAll(keepingCapacity: true)
             hasNotifiedPlaybackStartedForStream = false
         }
     }
@@ -700,8 +825,10 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
      */
     private func resetStreamingPlaybackState() {
         audioPlaybackQueue.sync {
+            // os_log("调用链: resetStreamingPlaybackState 清理 before pendingPcm=%{public}d scheduled=%{public}d active=%{public}d", log: self.log, type: .info, self.pcmPendingData.count, self.scheduledBufferCount, self.activeStreamSynthesisCount)
             self.pcmPendingData = Data()
             self.hasStrippedWavHeader = false
+            self.wavHeaderProbeBuffer = Data()
             self.hasNotifiedPlaybackStartedForStream = false
             self.activeStreamSynthesisCount = 0
             self.scheduledBufferCount = 0
@@ -710,6 +837,7 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
             self.audioEngine = nil
             self.playerNode = nil
             self.pcmFormat = nil
+            // os_log("调用链: resetStreamingPlaybackState 清理完成", log: self.log, type: .info)
         }
     }
 
@@ -720,6 +848,9 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
      */
     private func scheduleAvailablePcmBuffers() {
         guard ensureStreamingEngineIfNeeded() else { return }
+        if !hasNotifiedPlaybackStartedForStream, pcmPendingData.count >= streamChunkBytes {
+            // os_log("调用链: scheduleAvailablePcmBuffers 准备调度 pendingPcm=%{public}d chunk=%{public}d active=%{public}d", log: log, type: .info, pcmPendingData.count, streamChunkBytes, activeStreamSynthesisCount)
+        }
         while pcmPendingData.count >= streamChunkBytes {
             let chunk = pcmPendingData.prefix(streamChunkBytes)
             pcmPendingData.removeFirst(streamChunkBytes)
@@ -727,7 +858,7 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
             scheduledBufferCount += 1
             if !hasNotifiedPlaybackStartedForStream {
                 hasNotifiedPlaybackStartedForStream = true
-                os_log("调用链: 流式播放开始 session=%{public}@", log: log, type: .info, sessionid)
+                // os_log("调用链: 流式播放开始 session=%{public}@", log: log, type: .info, sessionid)
                 notifyEvent(eventType: .playbackStarted)
             }
             playerNode?.scheduleBuffer(buffer, completionHandler: { [weak self] in
@@ -747,7 +878,7 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
                 scheduledBufferCount += 1
                 if !hasNotifiedPlaybackStartedForStream {
                     hasNotifiedPlaybackStartedForStream = true
-                    os_log("调用链: 流式播放开始 session=%{public}@", log: log, type: .info, sessionid)
+                    // os_log("调用链: 流式播放开始 session=%{public}@", log: log, type: .info, sessionid)
                     notifyEvent(eventType: .playbackStarted)
                 }
                 playerNode?.scheduleBuffer(buffer, completionHandler: { [weak self] in
@@ -759,7 +890,8 @@ public class AzureTtsHelper: NSObject, ITtsService, AVAudioPlayerDelegate {
                 })
             }
         }
-        if playerNode?.isPlaying == false {
+        if playerNode?.isPlaying == false, scheduledBufferCount > 0 {
+            // os_log("调用链: playerNode.play scheduled=%{public}d pendingPcm=%{public}d", log: log, type: .debug, scheduledBufferCount, pcmPendingData.count)
             playerNode?.play()
         }
     }
@@ -773,7 +905,7 @@ private func checkStreamPlaybackCompletedIfNeeded() {
         if activeStreamSynthesisCount == 0, pcmPendingData.isEmpty, scheduledBufferCount == 0 {
             speaking = false
             hasNotifiedPlaybackStartedForStream = false
-            os_log("调用链: 流式播放完成 session=%{public}@", log: log, type: .info, sessionid)
+            // os_log("调用链: 流式播放完成 session=%{public}@", log: log, type: .info, sessionid)
             notifyEvent(eventType: .playbackCompleted)
             deactivateAudioSessionAfterTTSIfIdle()
         }
@@ -862,25 +994,36 @@ private func checkStreamPlaybackCompletedIfNeeded() {
      */
     private func stripWavHeaderIfNeeded(_ data: Data) -> Data {
         if hasStrippedWavHeader { return data }
-        if data.count >= 12,
+        wavHeaderProbeBuffer.append(data)
-           String(data: data.subdata(in: 0..<4), encoding: .ascii) == "RIFF",
+        if wavHeaderProbeBuffer.count < 12 { return Data() }
-           String(data: data.subdata(in: 8..<12), encoding: .ascii) == "WAVE" {
+        if wavHeaderProbeBuffer.count >= 12,
+           String(data: wavHeaderProbeBuffer.subdata(in: 0..<4), encoding: .ascii) == "RIFF",
+           String(data: wavHeaderProbeBuffer.subdata(in: 8..<12), encoding: .ascii) == "WAVE" {
             let marker = Data("data".utf8)
-            if let range = data.range(of: marker, options: [], in: 0..<min(data.count, 512)) {
+            if let range = wavHeaderProbeBuffer.range(of: marker, options: [], in: 0..<min(wavHeaderProbeBuffer.count, 512)) {
                 let start = range.lowerBound + 8
-                if start <= data.count {
+                if start <= wavHeaderProbeBuffer.count {
                     hasStrippedWavHeader = true
-                    return data.subdata(in: start..<data.count)
+                    let out = wavHeaderProbeBuffer.subdata(in: start..<wavHeaderProbeBuffer.count)
+                    wavHeaderProbeBuffer.removeAll(keepingCapacity: true)
+                    // os_log("调用链: stripWavHeader dataChunk 命中 start=%{public}d out=%{public}d", log: log, type: .debug, start, out.count)
+                    return out
                 }
             }
-            if data.count > 44 {
+            if wavHeaderProbeBuffer.count > 44 {
                 hasStrippedWavHeader = true
-                return data.subdata(in: 44..<data.count)
+                let out = wavHeaderProbeBuffer.subdata(in: 44..<wavHeaderProbeBuffer.count)
+                wavHeaderProbeBuffer.removeAll(keepingCapacity: true)
+                // os_log("调用链: stripWavHeader fallback44 out=%{public}d", log: log, type: .debug, out.count)
+                return out
             }
             return Data()
         }
         hasStrippedWavHeader = true
-        return data
+        let out = wavHeaderProbeBuffer
+        wavHeaderProbeBuffer.removeAll(keepingCapacity: true)
+        // os_log("调用链: stripWavHeader 非WAV 直接透传 out=%{public}d", log: log, type: .debug, out.count)
+        return out
     }
 
     /**
@@ -908,7 +1051,7 @@ private func checkStreamPlaybackCompletedIfNeeded() {
             audioPlayer?.delegate = self
             audioPlayer?.prepareToPlay()
             isPlaying = true
-            os_log("调用链: 开始播放 队首长度=%{public}dB session=%{public}@", log: log, type: .info, next.count, sessionid)
+            // os_log("调用链: 开始播放 队首长度=%{public}dB session=%{public}@", log: log, type: .info, next.count, sessionid)
             notifyEvent(eventType: .playbackStarted)
             audioPlayer?.play()
         } catch {
@@ -931,7 +1074,7 @@ private func checkStreamPlaybackCompletedIfNeeded() {
         if isWavData(data) {
             return data
         }
-        os_log("调用链: 音频无WAV头，补WAV头后播放 bytes=%{public}d", log: log, type: .info, data.count)
+        // os_log("调用链: 音频无WAV头，补WAV头后播放 bytes=%{public}d", log: log, type: .info, data.count)
         let header = makeWavHeader(pcmDataSize: UInt32(data.count), sampleRate: 16000, channels: 1, bitsPerSample: 16)
         var out = Data()
         out.append(header)
@@ -999,7 +1142,7 @@ private func checkStreamPlaybackCompletedIfNeeded() {
         let fileURL = tmpDir.appendingPathComponent("azure_tts_\(UUID().uuidString).wav")
         do {
             try data.write(to: fileURL, options: .atomic)
-            os_log("调用链: 写入临时文件 path=%{public}@", log: log, type: .debug, fileURL.path)
+            // os_log("调用链: 写入临时文件 path=%{public}@", log: log, type: .debug, fileURL.path)
             return fileURL
         } catch {
             os_log("写入临时文件失败: %{public}@", log: log, type: .error, error.localizedDescription)
@@ -1026,7 +1169,7 @@ public func audioPlayerDidFinishPlaying(_ player: AVAudioPlayer, successfully fl
             currentTempFileURL = nil
         }
         isPlaying = false
-        os_log("调用链: 播放完成 成功=%{public}@ session=%{public}@", log: log, type: .info, flag.description, sessionid)
+        // os_log("调用链: 播放完成 成功=%{public}@ session=%{public}@", log: log, type: .info, flag.description, sessionid)
         notifyEvent(eventType: .playbackCompleted)
         // 如果还有剩余，继续播放
         startPlaybackIfNeeded()
@@ -1208,7 +1351,11 @@ private func applyAudioSessionForTTS() -> Bool {
         let audioSession = AVAudioSession.sharedInstance()
         let outputs: [AVAudioSessionPortDescription] = audioSession.currentRoute.outputs
         let hasBluetoothHFP = outputs.contains(where: { $0.portType == .bluetoothHFP })
+        let hasWiredHeadphones = outputs.contains(where: { $0.portType == .headphones || $0.portType == .headsetMic })
+        let hasBluetoothA2DP = outputs.contains(where: { $0.portType == .bluetoothA2DP })
+        let hasHeadphones = hasWiredHeadphones || hasBluetoothA2DP
         let otherAudioPlaying = audioSession.isOtherAudioPlaying
+        let isRecording = (self.micCapture?.isCapturing == true)
 
         let desiredCategory: AVAudioSession.Category
         let desiredMode: AVAudioSession.Mode
@@ -1218,6 +1365,21 @@ private func applyAudioSessionForTTS() -> Bool {
             desiredCategory = .playback
             desiredMode = .default
             desiredOptions.insert(.duckOthers)
+        } else if isRecording {
+            desiredCategory = .playAndRecord
+            desiredMode = self.isExtaudioSource ? .default : .videoChat
+            desiredOptions.insert(.allowBluetooth)
+            desiredOptions.insert(.allowBluetoothA2DP)
+            if self.isExtaudioSource {
+                if otherAudioPlaying {
+                    desiredOptions.insert(.duckOthers)
+                }
+            } else {
+                desiredOptions.insert(.mixWithOthers)
+                if !hasHeadphones {
+                    desiredOptions.insert(.defaultToSpeaker)
+                }
+            }
         } else {
             desiredCategory = .playback
             desiredMode = .spokenAudio
@@ -1233,6 +1395,9 @@ private func applyAudioSessionForTTS() -> Bool {
                     if audioSession.category != desiredCategory || audioSession.mode != desiredMode || audioSession.categoryOptions != desiredOptions {
                         try audioSession.setCategory(desiredCategory, mode: desiredMode, options: desiredOptions)
                     }
+                    if desiredCategory == .playAndRecord {
+                        _ = try? audioSession.setPreferredIOBufferDuration(0.02)
+                    }
                     try audioSession.setActive(true)
                     break
                 } catch {
@@ -1242,6 +1407,9 @@ private func applyAudioSessionForTTS() -> Bool {
                     Thread.sleep(forTimeInterval: 0.05)
                 }
             }
+        } catch {
+            do {
+                try audioSession.setActive(true)
             } catch {
                 do {
                     try audioSession.setCategory(.playback, mode: .default, options: [])
@@ -1252,6 +1420,7 @@ private func applyAudioSessionForTTS() -> Bool {
                 }
             }
         }
+    }
     
     if DispatchQueue.getSpecific(key: audioSessionQueueKey) != nil {
         applyBody()