feat: 客户端语音回放 长录音客户端切片，并行转录

app-android/app/src/main/java/com/huaga/life_echo/ui/viewmodel/CreateMemoryViewModel.kt

@@ -10,8 +10,11 @@ import com.huaga.life_echo.data.repository.ChapterRepository
 import com.huaga.life_echo.data.repository.MessageRepository
 import com.huaga.life_echo.feature.conversation.ports.ConversationApiPort
 import com.huaga.life_echo.feature.conversation.ports.ConversationRealtimePort
+import com.huaga.life_echo.feature.conversation.ports.AudioSegmentRequest
 import com.huaga.life_echo.feature.voice.AudioPlayer
+import com.huaga.life_echo.feature.voice.AudioSegmenter
 import com.huaga.life_echo.feature.voice.PlaybackInfo
+import com.huaga.life_echo.feature.voice.PendingVoiceSegmentBatchBuilder
 import com.huaga.life_echo.feature.voice.RecordingCoordinator
 import com.huaga.life_echo.feature.voice.RecordingFinishResult
 import com.huaga.life_echo.feature.voice.RecordingStartResult
@@ -20,7 +23,10 @@ import com.huaga.life_echo.network.MessageType
 import com.huaga.life_echo.model.MessageDto
 import com.huaga.life_echo.data.database.Chapter
 import kotlinx.coroutines.Job
+import kotlinx.coroutines.async
+import kotlinx.coroutines.awaitAll
 import kotlinx.coroutines.channels.Channel
+import kotlinx.coroutines.coroutineScope
 import kotlinx.coroutines.flow.MutableStateFlow
 import kotlinx.coroutines.flow.StateFlow
 import kotlinx.coroutines.flow.asStateFlow
@@ -44,6 +50,7 @@ class CreateMemoryViewModel(
     companion object {
         private const val TAG = "CreateMemoryViewModel"
         private const val MIN_RECORDING_DURATION = 1
+        private const val SEGMENT_DURATION_SECONDS = 60
     }
     
     private val audioPlayer = AudioPlayer(context)
@@ -272,13 +279,11 @@ class CreateMemoryViewModel(
                         Log.w(TAG, "录音时间太短: ${capture.durationSeconds}s")
                         return@launch
                     }
-                    val audioBytes = try {
+                    sendRecordingSegments(
-                        File(capture.session.filePath).readBytes()
+                        filePath = capture.session.filePath,
-                    } catch (e: Exception) {
+                        voiceSessionId = capture.session.voiceSessionId,
-                        Log.e(TAG, "读取录音文件失败: ${e.message}")
+                        durationSeconds = capture.durationSeconds,
-                        return@launch
+                    )
-                    }
-                    sendAudioMessage(audioBytes, capture.session.filePath, capture.durationSeconds)
                 }
                 is RecordingFinishResult.Failed -> {
                     Log.e(TAG, "停止录音失败: ${result.capture.cause.message}")
@@ -345,8 +350,16 @@ class CreateMemoryViewModel(
         }
     }
 
-    private suspend fun sendAudioMessage(audioBytes: ByteArray, filePath: String, durationSeconds: Int) {
+    /**
-        Log.d(TAG, "准备发送音频消息，大小: ${audioBytes.size}, 时长: ${durationSeconds}s")
+     * 将录音文件按 [SEGMENT_DURATION_SECONDS] 切片，逐段通过 [sendAudioSegment] 发送。
+     * 短录音（<= 段时长）不会真正切片，直接整文件复制为单段。
+     */
+    private suspend fun sendRecordingSegments(
+        filePath: String,
+        voiceSessionId: String,
+        durationSeconds: Int,
+    ) {
+        Log.d(TAG, "准备切片并发送，时长: ${durationSeconds}s, 段长: ${SEGMENT_DURATION_SECONDS}s")
 
         if (conversationId.value == null) {
             Log.d(TAG, "对话ID为空，开始创建新对话")
@@ -358,9 +371,9 @@ class CreateMemoryViewModel(
             return
         }
 
-        conversationId.value?.let { id ->
+        val id = conversationId.value ?: return
-            val tempMessageId = "audio_user_${System.currentTimeMillis()}"
 
+        val tempMessageId = "audio_user_${System.currentTimeMillis()}"
         val tempMessage = MessageDto(
             id = tempMessageId,
             conversationId = id,
@@ -370,23 +383,59 @@ class CreateMemoryViewModel(
             messageType = "audio"
         )
         historyMessages.value = historyMessages.value + tempMessage
-
         _audioFilePaths.value = _audioFilePaths.value + (tempMessageId to filePath)
         _audioDurations.value = _audioDurations.value + (tempMessageId to durationSeconds)
 
+        val segmentFiles = try {
+            AudioSegmenter.split(
+                inputPath = filePath,
+                segmentDurationSeconds = SEGMENT_DURATION_SECONDS,
+                cacheDir = context.cacheDir,
+            )
+        } catch (e: Exception) {
+            Log.e(TAG, "音频切片失败: ${e.message}", e)
+            connectionStatus.value = "音频处理失败: ${e.message}"
+            return
+        }
+
         try {
+            val segments = PendingVoiceSegmentBatchBuilder.build(
+                segmentFiles = segmentFiles,
+                conversationId = id,
+                voiceSessionId = voiceSessionId,
+            )
+
             isTyping.value = true
-                conversationRealtime.sendAudioMessage(audioBytes, id, durationSeconds)
+            Log.d(TAG, "并行发送 ${segments.size} 个音频段，服务端按 segmentIndex 排序拼接")
-                Log.d(TAG, "音频消息发送成功")
+            coroutineScope {
+                segments.map { segment ->
+                    async {
+                        conversationRealtime.sendAudioSegment(
+                            AudioSegmentRequest(
+                                audioBytes = segment.audioBytes,
+                                conversationId = segment.conversationId,
+                                voiceSessionId = segment.voiceSessionId,
+                                segmentIndex = segment.segmentIndex,
+                                duration = segment.durationSeconds,
+                                isLast = segment.isLast,
+                                clientSegmentId = segment.clientSegmentId,
+                            )
+                        )
+                        Log.d(TAG, "已发送段 ${segment.segmentIndex}/${segments.size - 1}, last=${segment.isLast}")
+                    }
+                }.awaitAll()
+            }
+            Log.d(TAG, "全部音频段发送完成")
         } catch (e: Exception) {
             isTyping.value = false
-                Log.e(TAG, "音频消息发送失败: ${e.message}", e)
+            Log.e(TAG, "音频段发送失败: ${e.message}", e)
             connectionStatus.value = "发送失败: ${e.message}"
             errorMessages.value = (errorMessages.value + "发送失败: ${e.message}").takeLast(10)
             historyMessages.value = historyMessages.value.filter { it.id != tempMessageId }
             _audioFilePaths.value = _audioFilePaths.value - tempMessageId
             _audioDurations.value = _audioDurations.value - tempMessageId
-            }
+        } finally {
+            segmentFiles.forEach { it.file.delete() }
         }
     }
 

app-android/app/src/main/java/com/huaga/life_echo/ui/viewmodel/ViewModelFactory.kt

@@ -24,7 +24,7 @@ class ViewModelFactory(private val context: Context) : ViewModelProvider.Factory
     override fun <T : ViewModel> create(modelClass: Class<T>): T {
         return when {
             modelClass.isAssignableFrom(CreateMemoryViewModel::class.java) -> {
-                val recorder = VoiceRecorder(context).apply { recordingLimit = 60 }
+                val recorder = VoiceRecorder(context).apply { recordingLimit = 600 }
                 CreateMemoryViewModel(
                     conversationRepository = container.conversationRepository,
                     chapterRepository = container.chapterRepository,