fix audio

doubao/audio_manager.py

@@ -35,13 +35,26 @@ class AudioDeviceManager:
         self.input_stream = None
         self.output_stream = None
         self.audio_queue = None
+        self.playback_queue = None  # 播放队列
         self.recording = False
+        self.playing = False
+        
+        # 预缓冲机制
+        self.pre_buffer = []
+        self.pre_buffer_size = 5  # 预缓冲5个音频块
+        self.buffer_threshold = 3  # 缓冲阈值，低于此值开始预缓冲
+        
+        # 静音检测和回声消除
+        self.silence_threshold = 500  # 静音阈值
+        self.echo_suppression_enabled = True
+        self.last_audio_level = 0
+        self.audio_level_history = []
         
     def open_input_stream(self):
         """打开音频输入流"""
         try:
             import queue
-            self.audio_queue = queue.Queue(maxsize=100)  # 音频数据队列
+            self.audio_queue = queue.Queue(maxsize=100)  # 增大队列大小，提供更多缓冲
             
             def audio_callback(indata, frames, time_info, status):
                 """音频数据回调"""
@@ -51,17 +64,23 @@ class AudioDeviceManager:
                     try:
                         # 将numpy数组转换为字节数据
                         audio_bytes = indata.tobytes()
+                        
+                        # 添加音频数据预处理，提高质量
+                        if hasattr(self, '_audio_processor'):
+                            audio_bytes = self._audio_processor(audio_bytes)
+                        
                         self.audio_queue.put_nowait(audio_bytes)
                     except queue.Full:
-                        print("警告: 音频队列已满，丢弃数据")
+                        pass  # 静默丢弃，避免阻塞
             
             self.input_stream = sd.InputStream(
                 samplerate=self.input_config.sample_rate,
                 channels=self.input_config.channels,
-                dtype='int16',  # 16-bit PCM
+                dtype='int16',
                 blocksize=self.input_config.chunk,
                 callback=audio_callback,
-                device=None  # 使用默认设备
+                device=None,
+                latency='low'  # 低延迟模式
             )
             self.input_stream.start()
             self.recording = True
@@ -73,14 +92,53 @@ class AudioDeviceManager:
     def open_output_stream(self):
         """打开音频输出流"""
         try:
+            import queue
+            self.playback_queue = queue.Queue(maxsize=50)  # 增大播放队列，提供更多缓冲
+            
+            def playback_callback(outdata, frames, time_info, status):
+                """音频播放回调"""
+                if status:
+                    print(f"播放状态: {status}")
+                
+                try:
+                    # 从队列获取音频数据
+                    audio_data = self.playback_queue.get_nowait()
+                    
+                    # 转换字节数据为numpy数组
+                    audio_array = np.frombuffer(audio_data, dtype=np.int16)
+                    audio_array = audio_array.reshape(-1, self.output_config.channels)
+                    
+                    # 应用音频淡入淡出效果，减少爆音
+                    if hasattr(self, '_apply_volume_fade'):
+                        audio_array = self._apply_volume_fade(audio_array)
+                    
+                    # 确保数据大小匹配
+                    if len(audio_array) < frames:
+                        # 数据不足，用0填充
+                        padded = np.zeros((frames, self.output_config.channels), dtype=np.int16)
+                        padded[:len(audio_array)] = audio_array
+                        outdata[:] = padded
+                    else:
+                        outdata[:] = audio_array[:frames]
+                        
+                except queue.Empty:
+                    # 队列为空，输出静音
+                    outdata.fill(0)
+                except Exception as e:
+                    print(f"播放回调错误: {e}")
+                    outdata.fill(0)
+            
             self.output_stream = sd.OutputStream(
                 samplerate=self.output_config.sample_rate,
                 channels=self.output_config.channels,
-                dtype='int16',  # 16-bit PCM
+                dtype='int16',
                 blocksize=self.output_config.chunk,
-                device=None  # 使用默认设备
+                callback=playback_callback,
+                device=None,
+                latency='low'  # 低延迟模式
             )
             self.output_stream.start()
+            self.playing = True
             return self.output_stream
         except Exception as e:
             print(f"打开输出流失败: {e}")
@@ -89,13 +147,42 @@ class AudioDeviceManager:
     def play_audio(self, audio_data: bytes) -> None:
         """播放音频数据"""
         try:
-            # 将字节数据转换为numpy数组
-            audio_array = np.frombuffer(audio_data, dtype=np.int16)
-            audio_array = audio_array.reshape(-1, self.output_config.channels)
+            if self.playing and self.playback_queue:
+                # 音频数据预缓冲：将大数据块分成更小的块以获得更流畅的播放
+                chunk_size = self.output_config.chunk * 2  # 每个样本2字节
                 
-            # 使用sounddevice播放
-            sd.play(audio_array, samplerate=self.output_config.sample_rate)
-            sd.wait()  # 等待播放完成
+                # 预处理音频数据
+                if hasattr(self, '_playback_processor'):
+                    audio_data = self._playback_processor(audio_data)
+                
+                # 预缓冲机制：在播放前积累一些音频块
+                if len(self.pre_buffer) < self.pre_buffer_size:
+                    chunk_size = self.output_config.chunk * 2
+                    for i in range(0, len(audio_data), chunk_size):
+                        chunk = audio_data[i:i+chunk_size]
+                        self.pre_buffer.append(chunk)
+                        if len(self.pre_buffer) >= self.pre_buffer_size:
+                            break
+                    
+                    # 如果预缓冲已满，开始播放
+                    if len(self.pre_buffer) >= self.pre_buffer_size:
+                        self._flush_pre_buffer()
+                
+                # 分块处理音频数据，避免单个数据块过大
+                for i in range(0, len(audio_data), chunk_size):
+                    chunk = audio_data[i:i+chunk_size]
+                    try:
+                        # 使用阻塞式put，确保不丢失数据
+                        self.playback_queue.put(chunk, timeout=0.1)
+                    except queue.Full:
+                        print("警告: 播放队列已满，丢弃音频数据")
+                        # 如果队列满，尝试清空一些旧数据
+                        try:
+                            self.playback_queue.get_nowait()
+                            self.playback_queue.put(chunk, timeout=0.05)
+                        except:
+                            pass
+                        break
         except Exception as e:
             print(f"音频播放失败: {e}")
 
@@ -105,9 +192,9 @@ class AudioDeviceManager:
             if not self.recording or self.audio_queue is None:
                 return b'\x00' * (frames * 2)  # 返回静音数据
             
-            # 从队列获取音频数据
+            # 使用更长的超时时间，提高音频数据获取成功率
             try:
-                audio_data = self.audio_queue.get(timeout=0.1)  # 100ms超时
+                audio_data = self.audio_queue.get(timeout=0.1)  # 增加超时时间
                 return audio_data
             except queue.Empty:
                 # 队列为空，返回静音数据
@@ -121,10 +208,75 @@ class AudioDeviceManager:
         """停止录音"""
         self.recording = False
 
+    def stop_playing(self):
+        """停止播放"""
+        self.playing = False
+        if self.playback_queue:
+            # 清空播放队列
+            while not self.playback_queue.empty():
+                try:
+                    self.playback_queue.get_nowait()
+                except queue.Empty:
+                    break
+
+    def _flush_pre_buffer(self):
+        """刷新预缓冲区到播放队列"""
+        if hasattr(self, 'pre_buffer') and self.pre_buffer:
+            for chunk in self.pre_buffer:
+                try:
+                    self.playback_queue.put(chunk, timeout=0.1)
+                except queue.Full:
+                    print("警告: 播放队列已满，丢弃预缓冲数据")
+                    break
+            self.pre_buffer.clear()
+    
+    def _apply_volume_fade(self, audio_array):
+        """应用音量淡入淡出效果，减少爆音"""
+        try:
+            # 简单的淡入淡出效果
+            fade_samples = min(100, len(audio_array) // 10)  # 淡入淡出样本数
+            
+            # 淡入
+            for i in range(fade_samples):
+                factor = i / fade_samples
+                audio_array[i] = int(audio_array[i] * factor)
+            
+            # 淡出
+            for i in range(fade_samples):
+                factor = (fade_samples - i) / fade_samples
+                audio_array[-(i+1)] = int(audio_array[-(i+1)] * factor)
+            
+            return audio_array
+        except Exception as e:
+            print(f"音量淡入淡出失败: {e}")
+            return audio_array
+    
+    def _detect_silence(self, audio_data):
+        """检测静音"""
+        try:
+            audio_array = np.frombuffer(audio_data, dtype=np.int16)
+            audio_level = np.abs(audio_array).mean()
+            
+            # 更新音频电平历史
+            self.audio_level_history.append(audio_level)
+            if len(self.audio_level_history) > 10:
+                self.audio_level_history.pop(0)
+            
+            # 计算平均音频电平
+            avg_level = np.mean(self.audio_level_history) if self.audio_level_history else 0
+            
+            # 检测静音
+            is_silence = audio_level < self.silence_threshold
+            return is_silence, audio_level, avg_level
+        except Exception as e:
+            print(f"静音检测失败: {e}")
+            return False, 0, 0
+    
     def cleanup(self) -> None:
         """清理音频设备资源"""
         try:
-            self.recording = False
+            self.stop_recording()
+            self.stop_playing()
             if self.input_stream:
                 self.input_stream.stop()
                 self.input_stream.close()
@@ -132,6 +284,9 @@ class AudioDeviceManager:
                 self.output_stream.stop()
                 self.output_stream.close()
             sd.stop()  # 停止所有音频播放
+            # 清空预缓冲区
+            if hasattr(self, 'pre_buffer'):
+                self.pre_buffer.clear()
         except Exception as e:
             print(f"清理音频设备失败: {e}")
 

doubao/config.py

@@ -42,7 +42,7 @@ start_session_req = {
 }
 
 input_audio_config = {
-    "chunk": 3200,
+    "chunk": 6400,  # 增大缓冲区大小，减少处理频率
     "format": "pcm",
     "channels": 1,
     "sample_rate": 16000,
@@ -50,7 +50,7 @@ input_audio_config = {
 }
 
 output_audio_config = {
-    "chunk": 3200,
+    "chunk": 6400,  # 增大缓冲区大小，减少处理频率
     "format": "pcm",
     "channels": 1,
     "sample_rate": 24000,