Speech-to-Text (STT) Providers

Speech-to-Text (STT) Providers

The DiktaMe.Core architecture treats all external AI services as hot-swappable plugins via strictly defined interfaces.

If you want to add support for a new Speech-to-Text service (like Azure Speech or Google Cloud Speech-to-Text), you simply need to create a class that conforms to our ISTTProvider interface and register it inside the Dependency Injection framework.

The Interfaces

There are two primary interfaces detailing what an STT service is capable of doing.

1. ISTTProvider (Batch Recognition)

This is the foundation. Any class implementing this is capable of receiving a monolithic byte array of WAV audio and returning a single, finalized transcript.

public interface ISTTProvider
{
    // The human-readable name of the provider (e.g. "Whisper.net Local")
    string Name { get; }
    
    // Checks if API keys or offline models are successfully acquired
    Task<bool> IsReadyAsync();

    // The core execution loop
    Task<string> TranscribeAsync(byte[] wavData, string language, CancellationToken cancellationToken);
}

Examples in codebase: WhisperProvider.cs

2. IStreamingSTTProvider (Live Recognition)

If a provider supports real-time WebSocket ingestion (where transcriptions arrive sequentially while the user is still speaking), it should implement this interface which extends the base ISTTProvider.

public interface IStreamingSTTProvider : ISTTProvider
{
    // C# 8.0 Async Streams for yielding text natively as it arrives over the wire
    IAsyncEnumerable<string> TranscribeStreamAsync(
        IAsyncEnumerable<byte[]> audioStream, 
        string language, 
        CancellationToken cancellationToken);
}

Examples in codebase: DeepgramProvider.cs, GeminiAudioProvider.cs

The STT Router

dIKta.me does not hardcode provider implementations into the view logic. Instead, ViewModels request the singleton STTRouter.

The STTRouter acts as a traffic director. Its job is to read the user's $App.Settings.ActiveSttProvider configuration, reach into the DI container via the ISTTProviderFactory, and return the correct provider dynamically.

Fallback Mechanism

Because we support Bring-Your-Own-Key (BYOK), the STTRouter has a vital safety mechanism built into its routing logic.

If a user selects Deepgram as their provider but they have not actually saved an API key, the IsReadyAsync() check will fail. When this happens, the Router will silently fall back to the WhisperProvider (Local execution) so that the user's dictate hotkey doesn't just crash the app.

Adding a New Provider

  1. Create MyCustomSttProvider.cs in src/DiktaMe.Core/STT/.
  2. Implement ISTTProvider.
  3. Add your provider to the SttProviderType selection enum.
  4. Register it logically inside the STTProviderFactory.cs switch statement.
  5. (Dependency Injection): Register your new class as a Transient service in App.xaml.cs.