Hololens 2中如何将WindowsMicrophoneStream接入Azure Speech SDK

方案一：自定义PullAudioInputStreamCallback实现（优先推荐） #

Speech SDK的Pull模式是SDK主动从回调拉取音频数据，刚好适配你已经初始化好的Room Capture流，步骤如下：

1. 先确认你的Windows麦克风流（Room Capture类别）格式和Speech SDK要求一致：

   • 采样率：16000Hz（SDK默认强制要求）
   • 位深：16位
   • 声道：单声道
   • 编码：PCM原始格式

2. 实现PullAudioInputStreamCallback核心类，重点完成Read()方法的音频数据读取逻辑：

   using Microsoft.CognitiveServices.Speech.Audio;
   using Windows.Media.Capture;
   using Windows.Media.MediaProperties;
   using Windows.Storage.Streams;
   
   public class RoomCaptureAudioCallback : PullAudioInputStreamCallback
   {
       private readonly MediaCapture _mediaCapture;
       private readonly AudioFrameInputNode _audioInputNode;
       private readonly byte[] _buffer;
       private bool _isStreaming = true;
   
       // 传入你已经初始化好的MediaCapture和对应的AudioFrameInputNode
       public RoomCaptureAudioCallback(MediaCapture mediaCapture, AudioFrameInputNode audioInputNode)
       {
           _mediaCapture = mediaCapture;
           _audioInputNode = audioInputNode;
           // 按10ms音频数据量初始化缓冲区（16000Hz*1声道*2字节/采样 = 320字节）
           _buffer = new byte[320];
       }
   
       public override int Read(byte[] dataBuffer, uint size)
       {
           if (!_isStreaming || _audioInputNode == null)
           {
               return 0; // 返回0告知SDK流已结束
           }
   
           uint bytesRead = 0;
           try
           {
               // 从Room Capture流中获取音频帧并读取PCM数据
               using (var frame = _audioInputNode.GetFrame())
               using (var audioBuffer = frame.LockBuffer(AudioBufferAccessMode.Read))
               using (var bufferRef = audioBuffer.CreateReference())
               {
                   var inputStream = bufferRef as IInputStream;
                   var readResult = inputStream.ReadAsync(_buffer.AsBuffer(), size, InputStreamOptions.None).AsTask().Wait();
                   bytesRead = (uint)_buffer.Length;
                   // 将读取到的数据复制到SDK要求的输出缓冲区
                   System.Buffer.BlockCopy(_buffer, 0, dataBuffer, 0, (int)bytesRead);
               }
           }
           catch
           {
               // 捕获流中断等异常，标记流停止
               _isStreaming = false;
               bytesRead = 0;
           }
   
           return (int)bytesRead;
       }
   
       public override void Close()
       {
           _isStreaming = false;
           _audioInputNode?.Dispose();
           base.Close();
       }
   }
   

3. 关联回调与Speech SDK初始化：

   // 实例化自定义回调（传入你已初始化的MediaCapture和AudioFrameInputNode）
   var audioCallback = new RoomCaptureAudioCallback(yourMediaCapture, yourAudioInputNode);
   // 创建Pull模式音频输入流
   var audioInputStream = AudioInputStream.CreatePullStream(audioCallback);
   // 生成SDK可用的音频配置
   var audioConfig = AudioConfig.FromStreamInput(audioInputStream);
   // 初始化语音识别器
   var speechRecognizer = new SpeechRecognizer(yourSpeechConfig, audioConfig);
   

方案二：Push模式主动推送音频数据 #

如果更倾向于主动将麦克风数据推送给SDK，可以实现后台读取线程，把Room Capture流数据推送到PushAudioInputStream：

1. 实现Push流处理类：

   using Microsoft.CognitiveServices.Speech.Audio;
   using Windows.Media.Capture;
   using Windows.Media.MediaProperties;
   using Windows.Storage.Streams;
   
   public class RoomCapturePushStreamHandler
   {
       private readonly PushAudioInputStream _pushStream;
       private readonly MediaCapture _mediaCapture;
       private readonly AudioFrameInputNode _audioInputNode;
       private bool _isRunning = true;
   
       public RoomCapturePushStreamHandler(MediaCapture mediaCapture, AudioFrameInputNode audioInputNode)
       {
           _mediaCapture = mediaCapture;
           _audioInputNode = audioInputNode;
           _pushStream = AudioInputStream.CreatePushStream();
           // 启动后台线程读取并推送数据
           _ = ReadAndPushAudioData();
       }
   
       private async Task ReadAndPushAudioData()
       {
           var buffer = new byte[320]; // 同10ms音频数据量
           while (_isRunning)
           {
               try
               {
                   using (var frame = _audioInputNode.GetFrame())
                   using (var audioBuffer = frame.LockBuffer(AudioBufferAccessMode.Read))
                   using (var bufferRef = audioBuffer.CreateReference())
                   {
                       var inputStream = bufferRef as IInputStream;
                       var readResult = await inputStream.ReadAsync(buffer.AsBuffer(), (uint)buffer.Length, InputStreamOptions.None);
                       if (readResult.Length > 0)
                       {
                           // 将PCM数据推送给SDK
                           _pushStream.Write(buffer, 0, (int)readResult.Length);
                       }
                   }
                   // 匹配音频采样频率控制读取间隔
                   await Task.Delay(10);
               }
               catch
               {
                   _isRunning = false;
                   break;
               }
           }
       }
   
       // 返回SDK可用的音频配置
       public AudioConfig GetAudioConfig()
       {
           return AudioConfig.FromStreamInput(_pushStream);
       }
   
       // 停止推送并释放资源
       public void Stop()
       {
           _isRunning = false;
           _pushStream.Close();
       }
   }
   

2. 初始化Speech识别器：

   var pushHandler = new RoomCapturePushStreamHandler(yourMediaCapture, yourAudioInputNode);
   var audioConfig = pushHandler.GetAudioConfig();
   var speechRecognizer = new SpeechRecognizer(yourSpeechConfig, audioConfig);
   

关键注意事项 #

• 必须保证Room Capture流的音频格式严格匹配SDK要求，否则会出现识别失败、杂音等问题。
• Hololens 2端需要在Unity Player Settings中开启Microphone权限，同时在Package.appxmanifest中添加microphone和spatialPerception权限（Room Capture依赖空间感知）。
• 场景退出或识别停止时，务必释放MediaCapture、AudioNode、Speech识别器等实例，避免内存泄漏。

内容的提问来源于stack exchange，提问作者Leado