使用NAudio流式输出音频出现间隙的问题排查请求

我在将软件定义无线电（SDR）捕获的双声道音频流式传输到音频设备时遇到间隙问题：

• SDR捕获的音频以两个double数组存储（左、右声道各一组），原始采样率2MHz，每组数组长度固定为32768
• 需要将信号下采样至192kHz后，通过NAudio输出到音频设备
• 测试发现：单次输出完整信号数组时音频连续无间隙；但拆分多个数组依次输出时，每次传输都会出现音频间隙

当前实现的核心问题在于每次调用SkimmerOutput都会创建新的WaveOutEvent和StereoSampleProvider实例，并且在前一个实例播放完成后才会初始化下一个实例。这个过程中存在以下导致间隙的环节：

1. 旧的WaveOutEvent和StereoSampleProvider被销毁时，音频设备会停止输出
2. 新实例初始化、配置设备、启动播放的过程存在明显时间差，这段时间音频设备无数据输出，从而产生间隙
3. 下采样逻辑仅直接抽取采样点，可能导致信号不连续，但这不是间隙的主要原因

核心思路是复用同一个WaveOutEvent和支持动态追加数据的SampleProvider，通过缓冲队列持续给音频设备喂数据，避免重复初始化音频输出组件。

步骤1：实现支持动态追加数据的BufferedStereoSampleProvider #

替换原来一次性的StereoSampleProvider，改用可持续接收新数据的缓冲版本：

using System;
using System.Collections.Concurrent;
using NAudio.Wave;

class BufferedStereoSampleProvider : IWaveProvider
{
    private readonly BlockingCollection<Tuple<float[], float[]>> _bufferQueue = new BlockingCollection<Tuple<float[], float[]>>();
    private float[] _currentLeftBuffer;
    private float[] _currentRightBuffer;
    private int _currentBufferPosition;
    private readonly WaveFormat _waveFormat;

    public BufferedStereoSampleProvider(int sampleRate)
    {
        _waveFormat = WaveFormat.CreateIeeeFloatWaveFormat(sampleRate, 2);
    }

    public WaveFormat WaveFormat => _waveFormat;

    // 对外提供的追加数据方法
    public void AddSamples(double[] left, double[] right)
    {
        var leftFloat = Array.ConvertAll(left, d => (float)d);
        var rightFloat = Array.ConvertAll(right, d => (float)d);
        _bufferQueue.Add(Tuple.Create(leftFloat, rightFloat));
    }

    public int Read(byte[] buffer, int offset, int count)
    {
        int bytesWritten = 0;
        int samplesNeeded = count / 8; // 每个立体声样本占8字节（2*4字节float）

        while (samplesNeeded > 0)
        {
            // 当前缓冲为空时，从队列取新数据
            if (_currentLeftBuffer == null || _currentBufferPosition >= _currentLeftBuffer.Length)
            {
                if (!_bufferQueue.TryTake(out var nextBuffer, 100)) // 等待新数据，超时避免卡死
                {
                    break;
                }
                _currentLeftBuffer = nextBuffer.Item1;
                _currentRightBuffer = nextBuffer.Item2;
                _currentBufferPosition = 0;
            }

            // 计算本次能读取的样本数
            int samplesToCopy = Math.Min(samplesNeeded, _currentLeftBuffer.Length - _currentBufferPosition);
            for (int i = 0; i < samplesToCopy; i++)
            {
                BitConverter.GetBytes(_currentLeftBuffer[_currentBufferPosition]).CopyTo(buffer, offset + bytesWritten);
                BitConverter.GetBytes(_currentRightBuffer[_currentBufferPosition]).CopyTo(buffer, offset + bytesWritten + 4);
                bytesWritten += 8;
                _currentBufferPosition++;
                samplesNeeded--;
            }
        }

        return bytesWritten;
    }

    // 清空缓冲，用于停止播放时重置
    public void ClearBuffer()
    {
        _bufferQueue.TryTake(out _, -1);
        _currentLeftBuffer = null;
        _currentRightBuffer = null;
        _currentBufferPosition = 0;
    }
}

步骤2：重构CWSkimmer类，使用全局播放实例 #

将WaveOutEvent和BufferedStereoSampleProvider改为全局实例，避免重复创建：

using System;
using System.Linq;
using System.Windows.Forms;
using NAudio.Wave;
using System.Threading;

class CWSkimmer
{
    private static WaveOutEvent _waveOut;
    private static BufferedStereoSampleProvider _bufferedProvider;
    private static readonly int _desiredSampleRate = 192000;
    private static readonly int _originalSampleRate = 2000000;

    // 初始化播放组件（只执行一次）
    public static void InitializePlayback(int outputDeviceIndex = 2)
    {
        if (_waveOut != null) return;

        _bufferedProvider = new BufferedStereoSampleProvider(_desiredSampleRate);
        _waveOut = new WaveOutEvent
        {
            DeviceNumber = outputDeviceIndex,
            DesiredLatency = 100 // 调整延迟，平衡响应速度和稳定性
        };
        _waveOut.Init(_bufferedProvider);
        _waveOut.Play();
    }

    public static void SkimmerOutput(double[] leftChannel, double[] rightChannel)
    {
        // 确保播放组件已初始化
        if (_waveOut == null)
        {
            InitializePlayback();
        }

        // 下采样信号
        double[] downsampledLeft = Downsample(leftChannel, _originalSampleRate, _desiredSampleRate);
        double[] downsampledRight = Downsample(rightChannel, _originalSampleRate, _desiredSampleRate);

        // 将数据追加到缓冲队列
        _bufferedProvider.AddSamples(downsampledLeft, downsampledRight);
    }

    static double[] Downsample(double[] signal, int originalSampleRate, int desiredSampleRate)
    {
        int downsamplingFactor = originalSampleRate / desiredSampleRate;
        int newLength = signal.Length / downsamplingFactor;

        double[] downsampledSignal = new double[newLength];

        // 可选优化：用窗口均值代替单点抽取，减少混叠失真
        // for (int i = 0, j = 0; i < signal.Length && j < newLength; i += downsamplingFactor, j++)
        // {
        //     double sum = 0;
        //     for (int k = 0; k < downsamplingFactor && i + k < signal.Length; k++)
        //     {
        //         sum += signal[i + k];
        //     }
        //     downsampledSignal[j] = sum / downsamplingFactor;
        // }
        downsampledSignal[j] = signal[i];

        return downsampledSignal;
    }

    // 停止播放并清理资源
    public static void StopPlayback()
    {
        _waveOut?.Stop();
        _bufferedProvider?.ClearBuffer();
        _waveOut?.Dispose();
        _waveOut = null;
    }

    public static WaveOutCapabilities[] GetOutputDevices()
    {
        return Enumerable.Range(0, WaveOut.DeviceCount)
            .Select(i => WaveOut.GetCapabilities(i))
            .ToArray();
    }
}

步骤3：修改测试代码，适配新的播放逻辑 #

测试时只需初始化一次，然后连续输出多个数据包：

// 测试代码（无间隙版本）
private void button1_Click(object sender, EventArgs e)
{
    double frequency = 400.0;
    int sampleRate = 2000000;
    double durationSeconds = 5.0;

    double[] generatedSignal = SignalGenerator.GenerateSineWave(frequency, sampleRate, durationSeconds);
    double[] packet1 = new double[5000000];
    double[] packet2 = new double[5000000];

    Array.Copy(generatedSignal, 0, packet1, 0, 5000000);
    Array.Copy(generatedSignal, 5000000, packet2, 0, 5000000);

    // 初始化播放（只需一次）
    CWSkimmer.InitializePlayback(2);
    // 连续输出两个数据包，无间隙
    CWSkimmer.SkimmerOutput(packet1, packet1);
    CWSkimmer.SkimmerOutput(packet2, packet2);
}

• 下采样优化：当前直接抽取采样点会导致混叠失真，建议添加低通滤波器后再抽取，或使用均值采样提升音频质量
• 延迟调整：WaveOutEvent的DesiredLatency参数可在50-200ms区间测试，平衡延迟和播放稳定性
• 线程安全：若SkimmerOutput从多线程调用（如SDR数据捕获线程），当前的BlockingCollection已保证线程安全
• 异常处理：添加播放过程中的异常捕获，避免设备断开等问题导致程序崩溃

内容的提问来源于stack exchange，提问作者Tom