创建正确的 wav 音频格式以发送到 Azure 认知服务语音/翻译 SDK 宽度 MediaRecorder 和 SignalR

Question

Instead of directly using the Azure Cognitive Services JS SDK on my web page, I need to send my recorded sound to my server through SignalR to apply some logic and then send the audio to Translation SDK. 到 stream 我正在使用 JS MediaRecorder 的客户端的音频：

const blobToBase64 = function (blob, callback) {
    const reader = new FileReader();
    reader.onload = function () {
        var dataUrl = reader.result;
        var base64 = dataUrl.split(',')[1];
        callback(base64);
    };
    reader.readAsDataURL(blob);
};

const startUplaoding = async () => {
    const subject = new signalR.Subject();
    const connection = new signalR.HubConnectionBuilder().withUrl("https://.../myHub").build()
    await connection.start();

    //stream to the hub function as chunks of bytes
    await connection.send("UploadStream", subject);

    navigator.mediaDevices.getUserMedia({
        audio: {
            //Azure Speech SDK supported format 
            channelCount: 1,
            sampleSize: 16,
            sampleRate: 16000
        }, video: false
    })
        .then(stream => {
            var audioTrack = stream.getAudioTracks()[0];
            //making sure the constraints are in place
            audioTrack.applyConstraints({
                channelCount: 1,
                sampleSize: 16,
                sampleRate: 16000
            })

            mediaRecorder = new MediaRecorder(stream, {
                mimeType: "audio/webm;codecs=pcm",
            });

            mediaRecorder.addEventListener("dataavailable", e => {
                //convert blob to base64 to send to the SignalR as string
                //Tried sending the blobk directly using "subject.next(e.data)" didn't work
                blobToBase64(e.data, base64 => {
                    subject.next(base64);
                })
            })

            //timeslice = 1000, send every second
            mediaRecorder.start(1000)

            mediaRecorder.addEventListener("stop", () => {
                subject.complete()
            });
        });
}

在我的 SignalR 集线器上，在接收到每个数据块并将其转换为字节数组后，我将块发送到 Azure Translation SDK using in-memory

//SignalR Hub function
public async Task UploadStream(string sessionId, IAsyncEnumerable<string> stream)
{
    await foreach (var base64Str in stream)
    {
        var chunk = Convert.FromBase64String(base64Str);

        TranslationServiceInstance.PushAudio(chunk);
    }
}


//TranslationService Class
public class TranslationService
{
    private PushAudioInputStream audioInputStream { get; set; }
    private AudioConfig audioConfig { get; set; }
    private TranslationRecognizer recognizer { get; set; }

    public event EventHandler<byte[]> AudioReceived;
    public event EventHandler<string> TextReceived;

    public TranslationService()
    {
        var translationConfig = SpeechTranslationConfig.FromSubscription("###", "REGION");
        translationConfig.SetProperty(PropertyId.Speech_LogFilename, @$"PATH\log.txt");
        translationConfig.SpeechRecognitionLanguage = "en-US";

        translationConfig.AddTargetLanguage("fa");
        translationConfig.VoiceName = "fa-IR-FaridNeural";

        audioInputStream = AudioInputStream.CreatePushStream();
        //been trying to convert to a correct format, but not sure what format the audio chunks are being sent
        //audioInputStream = AudioInputStream.CreatePushStream(AudioStreamFormat.GetWaveFormatPCM(16000, 32, 1));
        audioConfig = AudioConfig.FromStreamInput(audioInputStream);
        recognizer = new TranslationRecognizer(translationConfig, audioConfig);
    }

    public async Task Start()
    {
        recognizer.Recognized += Recognizer_Recognized;
        recognizer.Synthesizing += Recognizer_Synthesizing;
        await recognizer.StartContinuousRecognitionAsync();
    }

    public void PushAudio(byte[] audioChunk)
    {
        //To make sure the audio chunk are being sent correctly
        using (var fs = new FileStream(@$"PATH\sending.wav", FileMode.Append))
        {
            fs.Write(audioChunk, 0, audioChunk.Length);
        }

        audioInputStream.Write(audioChunk, audioChunk.Length);
    }

    private void Recognizer_Synthesizing(object sender, TranslationSynthesisEventArgs e)
    {
        var bytes = e.Result.GetAudio();
        AudioReceived?.Invoke(this, bytes);
    }

    private void Recognizer_Recognized(object sender, TranslationRecognitionEventArgs e)
    {
        TextReceived?.Invoke(this, $"Recognizer_Recognized{e.Result.Text}");

        File.AppendAllText(@"PATH\result.txt", e.Result.Text);
    }
}

问题是，识别器没有检测到声音并抱怨音频格式。

问题：

• 如何确保从浏览器向 SignalR 集线器发送正确的格式？
• 我怎么知道 stream 的元数据是什么，所以我可以拒绝或将其转换为语音 SDK 所需的格式？

我的假设是当将原始数据从浏览器发送到识别器时，它会自动将其转换为所需的格式，但似乎我遗漏了一些东西。