novolis-audio / speech-models.md
Speech models (STT / VAD)
SFX, voice, and live music
dotnetaudiottsminiaudionovolis
Bundled speech models live under models/ and ship in Novolis.Audio.Voice.SherpaOnnx as models/{profileId}.zip.
Profiles
| Profile | Engine | Purpose |
|---|---|---|
| `silero-vad` | Silero VAD | Voice activity segmentation before STT |
| `en-whisper-tiny` | Whisper tiny (en) | Offline transcription |
Catalog: generated `SpeechModelCatalog.g.cs` from `NovolisAudioSpeechModelsManifest.cs`.
Fetch (maintainers)
pwsh -File scripts/fetch-speech-model.ps1 -ProfileId all
pwsh -File scripts/pack-all-speech-model-archives.ps1Usage
services.AddNovolisSpeech();
await foreach (var utterance in speech.ListenAsync(new ListenOptions(), ct))
Console.WriteLine(utterance.Text);Without models on disk, Sherpa adapters fall back to null engines (no transcripts, no throw) so CI stays green.
Environment
NOVOLIS_SPEECH_MODEL_DIR— override search root for bundled speech models (same layout asmodels/{profileId}/).