Begin by craeting a Speech resource in the Azure portal. The free tier is valid for seven days and yields two interchangeable subscription keys plus a region identifier (for trial accounts the region is usually westus).

1. Install the SDK

dotnet add package Microsoft.CognitiveServices.Speech

2. Platform target

Open Configuration Manager and set the active solution platform to x64 (or x86 on 32-bit systems). The native libraries do not support Any CPU.

3. Single-shot recognition from the microphone

using Microsoft.CognitiveServices.Speech;
using Microsoft.CognitiveServices.Speech.Audio;

public static async Task<string> TranscribeOnceAsync(string key, string region)
{
    var cfg = SpeechConfig.FromSubscription(key, region);
    cfg.SpeechRecognitionLanguage = "zh-CN";

    using var reco = new SpeechRecognizer(cfg);
    Console.WriteLine("Speak now...");
    var outcome = await reco.RecognizeOnceAsync();

    switch (outcome.Reason)
    {
        case ResultReason.RecognizedSpeech:
            return outcome.Text;
        case ResultReason.NoMatch:
            return "(no speech detected)";
        case ResultReason.Canceled:
            var c = CancellationDetails.FromResult(outcome);
            return $"[Canceled] {c.ErrorDetails}";
        default:
            return "(unknown)";
    }
}

4. Continuous recognition from a audio file

public static async Task TranscribeFileAsync(string key, string region, string wavPath)
{
    var cfg = SpeechConfig.FromSubscription(key, region);
    cfg.SpeechRecognitionLanguage = "zh-CN";

    var finished = new TaskCompletionSource<int>();

    using var input = AudioConfig.FromWavFileInput(wavPath);
    using var reco = new SpeechRecognizer(cfg, input);

    reco.Recognizing  += (_, e) => Console.WriteLine($"Partial: {e.Result.Text}");
    reco.Recognized     += (_, e) => Console.WriteLine($"Final  : {e.Result.Text}");
    reco.Canceled       += (_, e) => Console.WriteLine($"Canceled: {e.ErrorDetails}");
    reco.SessionStarted   += (_, _) => Console.WriteLine("Session started");
    reco.SessionStopped   += (_, _) => finished.TrySetResult(0);

    await reco.StartContinuousRecognitionAsync();
    await finished.Task;
    await reco.StopContinuousRecognitionAsync();
}

Omit AudioConfig.FromWavFileInput to stream from the default microphone instead of a file.

5. Quick timing helper

var sw = System.Diagnostics.Stopwatch.StartNew();
// ... recognition work ...
sw.Stop();
Console.WriteLine($"Elapsed {sw.ElapsedMilliseconds} ms");

More advanced scenarios—such as integrating with Language Understanding (LUIS) for intent extractino—are covered in the official documentation.