ElevenLabs

Downloads

Frameworks

.NET

.NET MAUI

Operating Systems

Android

iOS

Windows

Overview

ElevenLabs provides high-quality, AI-powered text-to-speech with natural-sounding multilingual voices, and the Scribe speech-to-text model for transcription. This provider replaces the platform-native ITextToSpeechService and/or ISpeechToTextService registrations with ElevenLabs cloud calls while still using platform-native audio capture / playback for the actual microphone and speakers.

Note

Scribe is a request/response STT API — it is not a streaming recognizer. CloudSpeechToText buffers the captured PCM until you call Stop(), wraps it in a WAV container, and posts it once to /v1/speech-to-text. One final SpeechRecognitionResult is yielded. For continuous partial-result streaming during long sessions, use Azure AI Speech instead.

Setup

• NuGet Packages
• Project File
• MauiProgram.cs
• Package.appxmanifest

Shiny.Speech.ElevenLabs

// In MauiProgram.cs — register both STT (Scribe) and TTS
builder.Services.AddElevenLabsSpeech("your-api-key");

// Or pick one:
builder.Services.AddElevenLabsSpeechToText("your-api-key"); // Scribe only
builder.Services.AddElevenLabsTextToSpeech("your-api-key"); // TTS only

IAudioSource (microphone) is auto-registered when STT is enabled; IAudioPlayer (speakers) is auto-registered when TTS is enabled.

Or with a config object:

builder.Services.AddElevenLabsSpeech(new ElevenLabsConfig
{
    ApiKey = "your-api-key",
    SpeechToTextModel = "scribe_v1",                // default
    TextToSpeechModel = "eleven_multilingual_v2",   // default
    DefaultVoiceId = "21m00Tcm4TlvDq8ikWAM"         // Rachel (default)
});

Configuration

public record ElevenLabsConfig
{
    public required string ApiKey { get; init; }
    public string DefaultVoiceId { get; init; } = "21m00Tcm4TlvDq8ikWAM"; // Rachel
    public string TextToSpeechModel { get; init; } = "eleven_multilingual_v2";
    public string SpeechToTextModel { get; init; } = "scribe_v1";
}

Property	Description	Default
ApiKey	Your ElevenLabs API key	(required)
DefaultVoiceId	Voice ID to use when none specified in TextToSpeechOptions	21m00Tcm4TlvDq8ikWAM (Rachel)
TextToSpeechModel	The TTS model to use	eleven_multilingual_v2
SpeechToTextModel	The Scribe transcription model to use	scribe_v1

TTS Usage

public class MyViewModel(ITextToSpeechService tts)
{
    async Task Speak()
    {
        await tts.SpeakAsync("Hello from ElevenLabs!");

        var voices = await tts.GetVoicesAsync();
        var voice = voices.FirstOrDefault(v => v.Name.Contains("Adam"));
        if (voice != null)
        {
            await tts.SpeakAsync("Hello!", new TextToSpeechOptions
            {
                Voice = voice,
                SpeechRate = 1.0f,
                Volume = 0.9f
            });
        }
    }
}

Audio Tags & Tone

ElevenLabs is the one provider that takes emotion inline, as bracketed audio tags in the text — but only on eleven_v3. Older models (including the default eleven_multilingual_v2, plus turbo and flash) read the tags aloud as words.

Shiny.Speech derives the provider’s tone capabilities from TextToSpeechModel, so switching models automatically switches between performing tags and stripping them:

builder.Services.AddElevenLabsSpeech(new ElevenLabsConfig
{
    ApiKey = "your-api-key",
    TextToSpeechModel = "eleven_v3"   // required for audio tags
});

// Tags perform on eleven_v3, and are stripped on every older model
await tts.SpeakAsync("[excited] We just shipped it. [laughs] Everything is live.");

// A portable SpeechTone is emitted as a leading tag
await tts.SpeakAsync("Don't wake them.", new TextToSpeechOptions
{
    Tone = new SpeechTone { Emotion = SpeechEmotion.Whispering }
});
// -> "[whispers] Don't wake them."

SpeechTone.Intensity has no ElevenLabs equivalent and is ignored. Note that eleven_v3 only accepts a voice_settings.stability of 0.0, 0.5 or 1.0.

STT Usage (Scribe)

Use it like any other ISpeechToTextService. Because Scribe is non-streaming, you’ll get partial results from the audio source only after Stop() is called:

public class MyViewModel(ISpeechToTextService stt)
{
    async Task Listen(CancellationToken ct)
    {
        var access = await stt.RequestAccess();
        if (access != AccessState.Available)
            return;

        stt.ResultReceived += (_, r) => Console.WriteLine($"[final] {r.Text}");

        await stt.Start(new SpeechRecognitionOptions
        {
            Culture = CultureInfo.GetCultureInfo("en-US")
        });

        // ... user speaks ...

        await stt.Stop(); // triggers the single POST + final result
    }

    // Or with the convenience extension method:
    async Task<string?> ListenOnce(CancellationToken ct)
        => await stt.ListenUntilSilence(cancellationToken: ct);
}

Combining ElevenLabs STT with Native TTS (or vice versa)

You can mix-and-match — register only the direction you need, and let the other side stay platform-native:

// Scribe STT + native TTS
builder.Services.AddSpeechServices();
builder.Services.AddElevenLabsSpeechToText("your-api-key");

// Native STT + ElevenLabs TTS
builder.Services.AddSpeechServices();
builder.Services.AddElevenLabsTextToSpeech("your-api-key");

Later registrations win when the same service is registered twice.

Links

• ElevenLabs Documentation
• ElevenLabs Voice Library
• Scribe Speech-to-Text