Android平台NAudio自定义录音驱动PCM Float高缓冲卡顿问题求助

核心问题定位 #

自定义NAudio录音驱动在PCM Float格式下，大缓冲时长（40/50ms）出现卡顿，但16位PCM格式无此问题，主要由以下几个原因导致：

• 每次循环重复创建缓冲区，引发GC频繁回收：大缓冲时数组内存占用更高，频繁内存分配与回收会阻塞录音线程，导致音频卡顿
• DataAvailable事件传递错误的字节长度：当前传递全量缓冲区长度，但实际有效字节数应为读取的浮点样本数×4，无效数据会导致处理端逻辑异常卡顿
• 缓冲区大小计算不匹配：RecordingLogic中重新计算的缓冲区大小，可能与AudioRecord初始化时的内部缓冲区大小不一致，大缓冲时更容易出现读取同步问题

针对性修复方案 #

方案1：复用缓冲区，避免频繁GC #

将Float格式的缓冲区提前初始化，在循环中复用，减少内存分配次数：

private void RecordingLogic()
{
    //Initialize the wave buffer
    int bufferSize = BufferMilliseconds * WaveFormat.AverageBytesPerSecond / 1000;
    if (bufferSize % WaveFormat.BlockAlign != 0)
    {
        bufferSize -= bufferSize % WaveFormat.BlockAlign;
    }

    WaveBuffer waveBuffer = new WaveBuffer(bufferSize);
    // 提前初始化Float格式缓冲区，循环内复用
    float[] floatBuffer = null;
    byte[] byteBuffer = null;
    if (WaveFormat.Encoding == WaveFormatEncoding.IeeeFloat)
    {
        floatBuffer = new float[bufferSize / 4];
        byteBuffer = new byte[bufferSize];
    }
    captureState = CaptureState.Capturing;

    //Run the record loop
    while (captureState != CaptureState.Stopped)
    {
        if (captureState != CaptureState.Capturing)
        {
            Thread.Sleep(10);
            continue;
        }

        if (WaveFormat.Encoding == WaveFormatEncoding.Pcm)
        {
            var bytesRead = audioRecord.Read(waveBuffer.ByteBuffer, 0, bufferSize);
            if (bytesRead > 0)
            {
                DataAvailable?.Invoke(this, new WaveInEventArgs(waveBuffer.ByteBuffer, bytesRead));
            }
        }
        else if (WaveFormat.Encoding == WaveFormatEncoding.IeeeFloat)
        {
            try
            {
                var floatsRead = audioRecord.Read(floatBuffer, 0, floatBuffer.Length, 1);
                if (floatsRead > 0)
                {
                    Buffer.BlockCopy(floatBuffer, 0, byteBuffer, 0, floatsRead * 4);
                    // 传递实际有效字节数，而非全量缓冲区大小
                    DataAvailable?.Invoke(this, new WaveInEventArgs(byteBuffer, floatsRead * 4));
                }
            }
            catch(Exception ex)
            {
                Console.WriteLine(ex);
            }
        }
    }
}

方案2：统一缓冲区大小计算逻辑 #

直接使用AudioRecord初始化时的缓冲区大小，避免读取时的同步问题：

private void RecordingLogic()
{
    if (audioRecord == null) return;
    
    // 直接复用AudioRecord的缓冲区大小，确保读取逻辑与底层一致
    int bufferSize = audioRecord.BufferSizeInFrames * WaveFormat.BlockAlign;
    WaveBuffer waveBuffer = new WaveBuffer(bufferSize);
    float[] floatBuffer = null;
    byte[] byteBuffer = null;
    if (WaveFormat.Encoding == WaveFormatEncoding.IeeeFloat)
    {
        floatBuffer = new float[bufferSize / 4];
        byteBuffer = new byte[bufferSize];
    }
    captureState = CaptureState.Capturing;

    //Run the record loop
    while (captureState != CaptureState.Stopped)
    {
        if (captureState != CaptureState.Capturing)
        {
            Thread.Sleep(10);
            continue;
        }

        if (WaveFormat.Encoding == WaveFormatEncoding.Pcm)
        {
            var bytesRead = audioRecord.Read(waveBuffer.ByteBuffer, 0, bufferSize);
            if (bytesRead > 0)
            {
                DataAvailable?.Invoke(this, new WaveInEventArgs(waveBuffer.ByteBuffer, bytesRead));
            }
        }
        else if (WaveFormat.Encoding == WaveFormatEncoding.IeeeFloat)
        {
            try
            {
                var floatsRead = audioRecord.Read(floatBuffer, 0, floatBuffer.Length, 1);
                if (floatsRead > 0)
                {
                    Buffer.BlockCopy(floatBuffer, 0, byteBuffer, 0, floatsRead * 4);
                    DataAvailable?.Invoke(this, new WaveInEventArgs(byteBuffer, floatsRead * 4));
                }
            }
            catch(Exception ex)
            {
                Console.WriteLine(ex);
            }
        }
    }
}

方案3：调整读取模式，避免线程占用过高 #

将阻塞读取模式改为非阻塞模式，配合短暂休眠让出CPU：

// 将读取模式参数从1改为0（非阻塞）
var floatsRead = audioRecord.Read(floatBuffer, 0, floatBuffer.Length, 0);
if (floatsRead == 0)
{
    Thread.Sleep(1); // 短暂休眠，避免线程空转占用过高CPU
    continue;
}

验证步骤 #

1. 优先修复DataAvailable事件的字节长度问题，这是导致处理端卡顿的直接原因
2. 再添加缓冲区复用逻辑，解决GC频繁回收问题
3. 最后验证缓冲区大小匹配的调整，确保录音线程与底层AudioRecord同步

内容的提问来源于stack exchange，提问作者SineVector241