feat(语音识别): 优化按住说话模式的识别结果处理

修改AzureAsrHelper中会话结束事件的条件判断，移除外部音频源限制
在AgentService中添加识别结果缓存，改进按住说话模式下的结果拼接和事件发送

local_plugins/agent_service/android/src/main/kotlin/com/yunqiinnovation/agent_service/AgentService.kt

@@ -149,6 +149,8 @@ object AgentService : CoroutineScope {
     private var isplaymusic = false;
     //是否保留最终结果
     private var isKeepResult = true;
+    // 识别结果
+    private var recognitionResult = ""
 
     // 打断防抖相关
     private var lastInterruptTime = 0L
@@ -666,20 +668,33 @@ object AgentService : CoroutineScope {
             val success = azureAsrHelper?.setupEventListeners(object :
                 AzureAsrHelper.ContinuousRecognizeCallback {
                 override fun onRecognizing(sessionid:String, recognizing: String, detectedLanguage: String) {
+                    Log.d(TAG, "liwei-------- onRecognizing: $sessionid")
                     if (recognizing.isNotEmpty()&&isKeepResult==true) {
                         // 检测到语音，更新状态
                         _hasSpeechDetected.set(true)
                         Log.d(TAG, "ai识别中重置定时器")
                         restartIdleCheck()
-                       
+                        var text = ""
-
+                        if (currentRecognitionMode == "push_to_talk") {
+                            text = recognitionResult + recognizing
+                            sendEvent(
+                                "recognizing", mapOf(
+                                    "sessionid" to currsessionId,
+                                    "text" to  text,
+                                    "language" to detectedLanguage
+                                )
+                            )
+                        }else{
+                            text = recognizing
                             sendEvent(
                                     "recognizing", mapOf(
                                     "sessionid" to sessionid,
-                                "text" to recognizing,
+                                    "text" to  text,
                                     "language" to detectedLanguage
                                 )
                             )
+                        }
+                       
 
                         // 如果TTS正在播放或AI正在生成，检查是否需要打断
                         if (isTtsSpeaking || isAiStreaming) {
@@ -709,7 +724,12 @@ object AgentService : CoroutineScope {
                 }
 
                 override fun onResult(sessionid:String,result: String, detectedLanguage: String) {
+                    Log.d(TAG, "liwei-------- onResult: $sessionid ")
                     if (result.isNotEmpty()&&isKeepResult==true) {//加入是否结束要输出最后的结果判读
+                        // 保存识别结果
+                        if (currentRecognitionMode == "push_to_talk") { //按住说话
+                            recognitionResult += result
+                        }else{
                             sendEvent(
                                 "recognition_result", mapOf(
                                     "sessionid" to sessionid,
@@ -717,10 +737,9 @@ object AgentService : CoroutineScope {
                                     "language" to detectedLanguage
                                 )
                             )
-
                             processWithChatApi(sessionid,result)
                         }
-
+                    }
 
                     _hasSpeechDetected.set(false)
 
@@ -729,25 +748,42 @@ object AgentService : CoroutineScope {
                 }
 
                 override fun onSessionStarted(sessionid:String) {
+                    Log.d(TAG, "liwei-------- onSessionStarted: $sessionid")
                     sendEvent("recognition_started", mapOf("sessionid" to sessionid, "status" to "started"))
                     // 使用模式感知的空闲检测启动方法
+                    recognitionResult = ""
+                    currsessionId = sessionid
                     audioPlayer?.playAudio(R.raw.start)
-
                 }
 
             
 
                 override fun onSessionStopped(sessionid:String) {
+                    Log.d(TAG, "liwei-------- onSessionStopped: $sessionid")
+                    if (currentRecognitionMode == "push_to_talk") { //按住说话
+                        sendEvent(
+                            "recognition_result", mapOf(
+                                "sessionid" to currsessionId,
+                                "text" to recognitionResult
+                            )
+                        )
+                        processWithChatApi(currsessionId,recognitionResult)
+
+                        sendEvent("recognition_stopped", mapOf("sessionid" to currsessionId, "status" to "stopped"))
+                    }else{
                         sendEvent("recognition_stopped", mapOf("sessionid" to sessionid, "status" to "stopped"))
-                    _isRecognitionActive.set(false)
+
-                    stopIdleCheck()
                         audioPlayer?.playAudio(R.raw.stop)
                     }
+                    stopIdleCheck()
+                    _isRecognitionActive.set(false)
+             
+                }
 
                 override fun onCanceled(sessionid: String,reason: String, errorDetails: String) {
+                     Log.d(TAG, "liwei-------- onCanceled: $sessionid $reason $errorDetails")
                     _isRecognitionActive.set(false)
                     stopIdleCheck()
-                    Log.d(TAG, "recognition_canceled:")
                     BleService.closeCodec()
                     sendEvent(
                         "recognition_canceled", mapOf(
@@ -759,6 +795,7 @@ object AgentService : CoroutineScope {
                 }
 
                 override fun onError(sessionid:String, code: Int, error: String) {
+                    Log.d(TAG, "liwei-------- onError: $sessionid $code $error")
                     _isRecognitionActive.set(false)
                     stopIdleCheck()
                     BleService.closeCodec()

local_plugins/azure_speech/android/src/main/kotlin/com/yunqiinnovation/azure_speech/AzureAsrHelper.kt

@@ -421,10 +421,10 @@ class AzureAsrHelper(private val context: Context) {
         EventHandler<SessionEventArgs> { _, _ ->
           // 直接在当前线程调用回调
           Log.d(tag, "会话结束事件")
-          if (audioSourceType == AudioSourceType.EXTERNAL) {
+          // if (audioSourceType == AudioSourceType.EXTERNAL) {
             callback.onSessionStopped(currsessionid)
             audioStream?.isContinuousRecognitionActive = false
-          }
+          // }
         }
       )