同声翻译-先从录制音频开始

同声翻译的核心：录制音频——调用接口（讯飞，百度，微软都有）语音转文字-调用翻译接口（讯飞，百度，微软都有）。

1：使用Naudio

为什么用这个，因为它可以调用扬声器和麦克风，为什么两个都要兼容，因为有些场景人家喜欢在电脑上放国外的电影，想着实时翻译……我能有什么办法……

NAudio/Docs/WasapiLoopbackCapture.md at master · naudio/NAudio · GitHub

这个技术可以检测声卡和麦克风，但是不能同时检测（虽然我写过，但是后来忘记了……）。

Lots of people ask how they can use NAudio to record the audio being played by another program. The answer is that unfortunately Windows does not provide an API that lets you target the output of one specific program to record. However, with WASAPI loopback capture, you can record all the audio that is being played out of a specific output device.

Since NAudio 2.1.0, the audio can be captured at a sample rate of your choosing, although it will make sense to match the sound card’s format.

Let’s start off by selecting a path to record to, creating an instance of WasapiLoopbackCapture (uses the default system device, but we can pass any rendering MMDevice that we want which we can find with MMDeviceEnumerator). We’ll also create a WaveFileWriter using the capture WaveFormat.

var outputFolder = Path.Combine(Environment.GetFolderPath(Environment.SpecialFolder.Desktop), "NAudio");
Directory.CreateDirectory(outputFolder);
var outputFilePath = Path.Combine(outputFolder, "recorded.wav");
var capture = new WasapiLoopbackCapture();
// optionally we can set the capture waveformat here: e.g. capture.WaveFormat = new WaveFormat(44100, 16,2);
var writer = new WaveFileWriter(outputFilePath, capture.WaveFormat);

We need to handle the DataAvailable event, and it’s very much the same approach here as recording to a WAV file from a regular WaveIn device. We just write BytesRecorded bytes from the Buffer into the WaveFileWriter. And in this example, I am stopping recording when we’ve captured 20 seconds worth of audio, by calling StopRecording.

capture.DataAvailable += (s, a) =>
{
    writer.Write(a.Buffer, 0, a.BytesRecorded);
    if (writer.Position > capture.WaveFormat.AverageBytesPerSecond * 20)
    {
        capture.StopRecording();
    }
};

When the RecordingStopped event fires, we Dispose our WaveFileWriter so we create a valid WAV file, and we’re done recording so we’ll Dispose our capture device as well.

capture.RecordingStopped += (s, a) =>
{
    writer.Dispose();
    writer = null;
    capture.Dispose();
};

All that remains is for us to start recording with StartRecording and wait for recording to finish by monitoring the CaptureState.

capture.StartRecording();
while (capture.CaptureState != NAudio.CoreAudioApi.CaptureState.Stopped)
{
    Thread.Sleep(500);
}

Now there is one gotcha with WasapiLoopbackCapture. If no audio is playing whatsoever, then the DataAvailable event won’t fire. So if you want to record “silence”, one simple trick is to simply use an NAudio playback device to play silence through that device for the duration of time you’re recording. Alternatively, you could insert silence yourself when you detect gaps in the incoming audio.

核心代码如下：

string mainpath = DateTime.Now.ToString("yyyyMMddHHmmss");

var outputFolder = Path.Combine(AppDomain.CurrentDomain.BaseDirectory, "temp", mainpath);

Directory.CreateDirectory(outputFolder);

var outputFilePath = Path.Combine(outputFolder, DateTime.Now.ToString("yyyyMMddHHmmss") + ".wav");

//声卡/扬声器录音
//var capture = new WasapiLoopbackCapture();
//var capture = new WasapiCapture();
//麦克风
var capture = new WasapiCapture();
// optionally we can set the capture waveformat here: e.g. capture.WaveFormat = new WaveFormat(44100, 16,2);
var writer = new WaveFileWriter(outputFilePath, capture.WaveFormat);

capture.DataAvailable += (s, a) =>
{
    writer.Write(a.Buffer, 0, a.BytesRecorded);

    if (writer.Position > capture.WaveFormat.AverageBytesPerSecond * 20)
    {
        var outputFilePath = Path.Combine(outputFolder, DateTime.Now.ToString("yyyyMMddHHmmss") + ".wav");
        //var capture = new WasapiLoopbackCapture();
        writer = new WaveFileWriter(outputFilePath, capture.WaveFormat);
        //capture.StopRecording();
    }
};
capture.RecordingStopped += (s, a) =>
{
    writer.Dispose();
    writer = null;
    capture.Dispose();
};

capture.StartRecording();

while (capture.CaptureState != NAudio.CoreAudioApi.CaptureState.Stopped)
{
    Thread.Sleep(500);
}

Console.WriteLine($"录音完成，文件保存到: {outputFilePath}");

这个代码的意思就是打开麦克风开始检测，每20S一个音频文件，你拿着去做语音转文字就行。