Merge branch 'main' of hf.co:MoYoYoTech/VoiceDialogue

.gitignore

@@ -262,3 +262,4 @@ electron-app/dist
 tests/tts_test_output
 .DS_Store
 package-lock.json
+output

src/voice_dialogue/services/audio/player.py

@@ -59,8 +59,7 @@ class AudioStreamPlayer(BaseThread, TaskStatusMixin, HistoryMixin, PerformanceLo
                     )
                 )
 
-            if is_debug_mode():
-                self.log_task_performance(voice_task, "音频播放")
+            self.log_task_performance(voice_task, "音频播放")
 
             self.update_chat_history(voice_task)
 

src/voice_dialogue/services/mixins.py

@@ -2,7 +2,7 @@ from collections import OrderedDict
 
 from voice_dialogue.core.constants import (
     voice_state_manager, session_manager, dropped_audio_cache,
-    user_still_speaking_event, chat_history_cache
+    user_still_speaking_event, chat_history_cache, is_debug_mode
 )
 from voice_dialogue.models.voice_task import VoiceTask
 from voice_dialogue.utils.logger import logger
@@ -62,6 +62,9 @@ class PerformanceLogMixin:
 
     def log_task_performance(self, voice_task: VoiceTask, task_name: str = "任务"):
         """记录ASR, LLM, TTS各阶段耗时和音频长度"""
+        if not is_debug_mode():
+            return
+
         try:
             from voice_dialogue.services.utils import calculate_audio_duration
 
@@ -89,3 +92,16 @@ class PerformanceLogMixin:
             )
         except Exception as e:
             logger.error(f"记录任务性能信息时出错: {e}")
+
+    def log_task_user_question(self, voice_task: VoiceTask):
+        if not is_debug_mode():
+            return
+
+        from voice_dialogue.config.paths import PROJECT_ROOT
+        output_path = PROJECT_ROOT / "output"
+        if not output_path.exists():
+            output_path.mkdir(parents=True, exist_ok=True)
+
+        import soundfile as sf
+        output_filename = output_path / (voice_task.id + ".wav")
+        sf.write(output_filename.as_posix(), voice_task.user_voice, 16000, subtype="PCM_16")

src/voice_dialogue/services/speech/monitor.py

@@ -30,7 +30,7 @@ class SpeechMonitorConfig:
     QUEUE_TIMEOUT = 0.1  # 队列获取超时时间（秒）
 
     # 时间阈值（毫秒）
-    ACTIVE_FRAME_THRESHOLD = 0.32 * 1000  # 连续活跃帧数阈值
+    ACTIVE_FRAME_THRESHOLD = 0.1 * 1000  # 连续活跃帧数阈值
     USER_SILENCE_THRESHOLD = 1 * 1000  # 用户静音阈值
     SILENCE_THRESHOLD = 0.3 * 1000  # 静音检测阈值
     AUDIO_FRAMES_THRESHOLD = 5 * 1000  # 音频帧时长阈值

src/voice_dialogue/services/speech/recognizer.py

@@ -7,11 +7,12 @@ import numpy as np
 from voice_dialogue.core.base import BaseThread
 from voice_dialogue.core.constants import user_still_speaking_event, voice_state_manager, dropped_audio_cache
 from voice_dialogue.models.voice_task import VoiceTask
+from voice_dialogue.services.mixins import PerformanceLogMixin
 from voice_dialogue.utils.cache import LRUCacheDict
 from .recognizers import asr_manager
 
 
-class ASRWorker(BaseThread):
+class ASRWorker(BaseThread, PerformanceLogMixin):
     def __init__(self, group=None, target=None, name=None, args=(), kwargs=None, *, daemon=None,
                  user_voice_queue: Queue,
                  transcribed_text_queue: Queue,
@@ -37,6 +38,7 @@ class ASRWorker(BaseThread):
             except Empty:
                 continue
 
+
             voice_task.language = self.language
             voice_task.whisper_start_time = time.time()
 
@@ -45,6 +47,8 @@ class ASRWorker(BaseThread):
             if not transcribed_text.strip():
                 voice_state_manager.reset_task_id()
                 continue
+                
+            self.log_task_user_question(voice_task)
 
             voice_task.whisper_end_time = time.time()