Deepgram

Features 🔥

• Fully generated C# SDK based on official Deepgram OpenAPI specification using AutoSDK
• Same day update to support new features
• Updated and supported automatically if there are no breaking changes
• All modern .NET features - nullability, trimming, NativeAOT, etc.
• Support .Net Framework/.Net Standard 2.0

Usage

using Deepgram;

using var client = new DeepgramClient(apiKey);

Generate

Basic example showing how to create a client and make a request.

using var client = new DeepgramClient(apiKey);

Speech To Text Client Get Text Async

using var client = new DeepgramClient(apiKey);
ISpeechToTextClient speechClient = client;

// Transcribe audio using the MEAI ISpeechToTextClient interface.
// Deepgram requires an audio URL via RawRepresentationFactory.
using var dummyStream = new MemoryStream();
var response = await speechClient.GetTextAsync(dummyStream, new SpeechToTextOptions
{
    ModelId = "nova-3",
    RawRepresentationFactory = _ => new ListenV1RequestUrl
    {
        Url = "https://dpgr.am/spacewalk.wav",
    },
});

Console.WriteLine($"Text: {response.Text}");

Speech To Text Client Get Service Metadata

using var client = new DeepgramClient("test-api-key");
ISpeechToTextClient speechClient = client;

// Retrieve metadata about the speech-to-text provider.
var metadata = speechClient.GetService<SpeechToTextClientMetadata>();

Realtime Speech To Text

Real-time speech-to-text streaming using the typed ConnectAsync with query parameters.

var apiKey =
    Environment.GetEnvironmentVariable("DEEPGRAM_API_KEY") is { Length: > 0 } apiKeyValue
        ? apiKeyValue
        : throw new AssertInconclusiveException("DEEPGRAM_API_KEY environment variable is not found.");

// Create a realtime ListenV1 client and authenticate.
await using var realtimeClient = new Realtime.DeepgramListenV1RealtimeClient();
realtimeClient.AuthorizeUsingToken(apiKey);

// Connect with typed query parameters — model, interim results, and language.
using var cts = new CancellationTokenSource(TimeSpan.FromSeconds(30));
await realtimeClient.ConnectAsync(
    model: Realtime.ListenV1Model.Nova3,
    interimResults: Realtime.ListenV1InterimResults.True,
    language: Realtime.ListenV1Language.FromString("en"),
    cancellationToken: cts.Token);

// Download a short audio sample and send it as binary frames.
using var httpClient = new HttpClient();
var audioBytes = await httpClient.GetByteArrayAsync(
    "https://dpgr.am/spacewalk.wav", cts.Token);

// Send audio in 8KB chunks.
const int chunkSize = 8192;
for (var offset = 0; offset < audioBytes.Length; offset += chunkSize)
{
    var length = Math.Min(chunkSize, audioBytes.Length - offset);
    await realtimeClient.SendAsync(
        new ArraySegment<byte>(audioBytes, offset, length),
        System.Net.WebSockets.WebSocketMessageType.Binary,
        endOfMessage: true,
        cts.Token);
}

// Signal end of audio and close the stream.
await realtimeClient.SendListenV1CloseStreamAsync(
    new Realtime.ListenV1ControlMessage
    {
        Type = Realtime.ListenV1ControlMessageType.CloseStream,
    },
    cts.Token);

// Receive transcription events until the connection closes.
var transcripts = new List<string>();
string? responseId = null;

await foreach (var serverEvent in realtimeClient
    .ReceiveUpdatesAsync(cts.Token))
{
    if (serverEvent.IsMetadata && serverEvent.Metadata is { } metadata)
    {
        responseId = metadata.RequestId.ToString();
        Console.WriteLine($"Session started: {responseId}");
    }
    else if (serverEvent.IsResults && serverEvent.Results is { } results)
    {
        if (results.IsFinal == true &&
            results.Channel?.Alternatives is { Count: > 0 } alts &&
            alts[0].Transcript is { Length: > 0 } transcript)
        {
            transcripts.Add(transcript);
            Console.WriteLine($"Final: {transcript}");
        }
    }
}

// Verify we received transcription results.
Console.WriteLine($"Total final transcripts: {transcripts.Count}");

Speech To Text Client Get Streaming Text Async

Real-time streaming speech-to-text via the MEAI ISpeechToTextClient interface. Uses typed ConnectAsync internally for model and language selection.

using var client = new DeepgramClient(apiKey);
ISpeechToTextClient speechClient = client;

// Download a short audio sample to use as streaming input.
using var httpClient = new HttpClient();
var audioBytes = await httpClient.GetByteArrayAsync(
    "https://dpgr.am/spacewalk.wav");
using var audioStream = new MemoryStream(audioBytes);

// Stream audio through the MEAI ISpeechToTextClient interface.
// This uses typed ConnectAsync internally with model and language params.
var updates = new List<SpeechToTextResponseUpdate>();
await foreach (var update in speechClient.GetStreamingTextAsync(
    audioStream,
    new SpeechToTextOptions
    {
        ModelId = "nova-3",
        SpeechLanguage = "en",
    }))
{
    updates.Add(update);

    if (update.Kind == SpeechToTextResponseUpdateKind.TextUpdated)
    {
        Console.WriteLine($"Final: {update.Text}");
    }
}

// Verify we received session events and transcription results.

var finalTranscripts = updates
    .Where(u => u.Kind == SpeechToTextResponseUpdateKind.TextUpdated && u.Text is { Length: > 0 })
    .Select(u => u.Text!)
    .ToList();
Console.WriteLine($"Total final transcripts: {finalTranscripts.Count}");

Support

Priority place for bugs: https://github.com/tryAGI/Deepgram/issues
Priority place for ideas and general questions: https://github.com/tryAGI/Deepgram/discussions
Discord: https://discord.gg/Ca2xhfBf3v

Acknowledgments

This project is supported by JetBrains through the Open Source Support Program.