Azure Speech SDK Speech-to-Text 流式传输音频片段

Question

我一直在使用 Azure 的 Speech-To-Text 服务，发现 here，使用从内存流中识别的方法。基本上我打算做的只是将音频的某些片段流式传输到服务，但我不完全确定如何做到这一点。假设我有一个长度为 5 分钟的视频，我的目标是仅流式传输前 30 秒，甚至只是从音频文件中的 1 分钟标记到 3 分钟标记，我需要在以下代码中启用或更改什么这样做？

我尝试使用 CreatePullStream() 而不是 CreatePushStream() 在几秒钟内提供标记，但它没有产生我上面描述的目标。如果有人知道，请告诉我如何实现这一点，非常感谢！

using System;
using System.IO;
using System.Threading.Tasks;
using Microsoft.CognitiveServices.Speech;
using Microsoft.CognitiveServices.Speech.Audio;

class Program 
{
    async static Task FromStream(SpeechConfig speechConfig)
    {
        var reader = new BinaryReader(File.OpenRead("audioFile.wav"));
        using var audioInputStream = AudioInputStream.CreatePushStream();
        using var audioConfig = AudioConfig.FromStreamInput(audioInputStream);
        using var recognizer = new SpeechRecognizer(speechConfig, audioConfig);

        byte[] readBytes;
        do
        {
            readBytes = reader.ReadBytes(1024);
            audioInputStream.Write(readBytes, readBytes.Length);
        } while (readBytes.Length > 0);

        var result = await recognizer.RecognizeOnceAsync();
        Console.WriteLine($"RECOGNIZED: Text={result.Text}");
    }

    async static Task Main(string[] args)
    {
        var speechConfig = SpeechConfig.FromSubscription("<paste-your-subscription-key>", "<paste-your-region>");
        await FromStream(speechConfig);
    }
}

Answer 1

您可以使用NAudio.Wave 来剪切您的源.wav 文件。例如，如果您想识别 .wav 文件的 1 分钟 - 3 分钟内容，请尝试以下代码：

using System;
using System.IO;
using System.Threading.Tasks;
using Microsoft.CognitiveServices.Speech;
using Microsoft.CognitiveServices.Speech.Audio;
using NAudio.Wave;

class Program
{
    async static Task FromStream(SpeechConfig speechConfig)
    {

        var inputAudioPath = @"<path>";
        var outputAudioPath = @"<path>";
        var startAt = new TimeSpan(0, 1, 0); //start at 1 min
        var duration = new TimeSpan(0,2,0); //cut 1-3 min audio, it lasts 2 mins

        CutAudio(inputAudioPath, outputAudioPath, startAt, duration);

        var reader = new BinaryReader(File.OpenRead(outputAudioPath));
        var audioInputStream = AudioInputStream.CreatePushStream();
        var audioConfig = AudioConfig.FromStreamInput(audioInputStream);
        var recognizer = new SpeechRecognizer(speechConfig, audioConfig);

        byte[] readBytes;
        do
        {
            readBytes = reader.ReadBytes(1024);
            audioInputStream.Write(readBytes, readBytes.Length);
        } while (readBytes.Length > 0);

        var result = await recognizer.RecognizeOnceAsync();
        Console.WriteLine($"RECOGNIZED: Text={result.Text}");
    }


    public static void CutAudio(String inputPath,String destPath,TimeSpan startAt, TimeSpan duration) {

        using (var reader = new AudioFileReader(inputPath))
        {
            reader.CurrentTime = startAt; // jump forward to the position we want to start from
            WaveFileWriter.CreateWaveFile16(destPath, reader.Take(duration));
        }
    }


    async static Task Main(string[] args)
    {


        var speechConfig = SpeechConfig.FromSubscription("<key>", "<region>");
        await FromStream(speechConfig);
    }
}

结果：

顺便说一句，如果你想识别长音频，请参阅this official doc 和我之前的帖子here。