audio

doubao/audio_manager.py

@@ -144,30 +144,28 @@ class AudioDeviceManager:
             raise Exception("无法打开任何音频输出设备")
     
     def write_audio_data(self, audio_data: bytes) -> bool:
-        """写入音频数据，支持格式转换和多种播放方式"""
-        # 如果需要格式转换
-        converted_data = audio_data
-        converted_format = None
+        """写入音频数据，性能优化版本"""
+        # 预缓冲：对于较小的音频数据，等待积累到一定大小再播放
+        min_buffer_size = 1024  # 最小缓冲大小
         
-        # 检查是否需要从Float32转换为Int16
-        if (self.output_config.bit_size == pyaudio.paInt16 and 
-            len(audio_data) % 4 == 0):  # 可能是Float32数据
-            
-            try:
-                # 检查是否为Float32数据（通过尝试解析）
-                import struct
-                test_sample = struct.unpack('f', audio_data[:4])[0]
-                if -1.0 <= test_sample <= 1.0:  # 合理的Float32范围
-                    print("检测到Float32数据，转换为Int16格式")
-                    converted_data = self.converter.float32_to_int16(audio_data)
-                    converted_format = 'Int16'
-            except:
-                pass  # 不是Float32数据，不进行转换
+        if not hasattr(self, '_audio_buffer'):
+            self._audio_buffer = b''
         
-        # 尝试直接写入pyaudio流
+        # 累积音频数据
+        self._audio_buffer += audio_data
+        
+        # 如果缓冲区足够大或者数据包较大，直接播放
+        if len(self._audio_buffer) >= min_buffer_size or len(audio_data) > min_buffer_size:
+            buffer_to_play = self._audio_buffer
+            self._audio_buffer = b''
+        else:
+            # 继续等待更多数据
+            return True
+        
+        # 直接写入pyaudio流，不做实时转换
         try:
             if self.output_stream:
-                self.output_stream.write(converted_data)
+                self.output_stream.write(buffer_to_play)
                 return True
         except Exception as e:
             print(f"pyaudio写入失败: {e}")
@@ -176,11 +174,9 @@ class AudioDeviceManager:
         enable_aplay = config.audio_config.get('enable_aplay_fallback', True)
         if enable_aplay and self.platform_config['fallback_to_aplay']:
             print("尝试使用aplay播放...")
-            format_type = converted_format if converted_format else ('Float32' if self.output_config.bit_size == pyaudio.paFloat32 else 'Int16')
-            
             return self.player.play_audio(
-                audio_data=converted_data,
-                format_type=format_type,
+                audio_data=buffer_to_play,
+                format_type='Int16',  # 现在统一使用Int16
                 sample_rate=self.output_config.sample_rate,
                 channels=self.output_config.channels
             )
@@ -257,9 +253,9 @@ class DialogSession:
             self.player_thread.start()
 
     def _audio_player_thread(self):
-        """音频播放线程"""
-        audio_playing_timeout = 1.0  # 1秒没有音频数据认为播放结束
-        queue_check_interval = 0.1  # 每100ms检查一次队列状态
+        """音频播放线程 - 性能优化版本"""
+        audio_playing_timeout = 0.5  # 0.5秒没有音频数据认为播放结束
+        queue_check_interval = 0.05  # 每50ms检查一次队列状态，更低的延迟
         
         while self.is_playing:
             try:
@@ -307,6 +303,14 @@ class DialogSession:
                                 self.say_hello_completed = True
                                 print("say hello 音频播放完成")
                             print("音频播放超时，恢复录音")
+                            
+                            # 刷新音频缓冲区
+                            if hasattr(self.audio_device, '_audio_buffer') and self.audio_device._audio_buffer:
+                                print("刷新剩余音频缓冲区")
+                                remaining_buffer = self.audio_device._audio_buffer
+                                self.audio_device._audio_buffer = b''
+                                self.audio_device.write_audio_data(remaining_buffer)
+                            
                             # 直接发送静音数据，而不是在协程中发送
                             try:
                                 silence_data = b'\x00' * config.input_audio_config["chunk"]

doubao/config.py

@@ -30,7 +30,7 @@ start_session_req = {
         "speaker": "zh_female_vv_jupiter_bigtts",
         # "speaker": "S_XXXXXX",  // 指定自定义的复刻音色,需要填下character_manifest
         # "speaker": "ICL_zh_female_aojiaonvyou_tob" // 指定官方复刻音色，不需要填character_manifest
-        "audio_config": {"channel": 1, "format": "pcm", "sample_rate": 24000},
+        "audio_config": {"channel": 1, "format": "pcm", "sample_rate": 16000},
     },
     "dialog": {
         "bot_name": "豆包",
@@ -58,9 +58,9 @@ input_audio_config = {
 }
 
 output_audio_config = {
-    "chunk": 3200,
+    "chunk": 4096,  # 增加缓冲区大小
     "format": "pcm",
     "channels": 1,
-    "sample_rate": 24000,
-    "bit_size": pyaudio.paFloat32,
+    "sample_rate": 16000,
+    "bit_size": pyaudio.paInt16,
 }