Novolis Docs
novolis-audio / speech-models.md

Speech models (STT / VAD)

dotnetaudiottsminiaudionovolis

Bundled speech models live under models/ and ship in Novolis.Audio.Voice.SherpaOnnx as models/{profileId}.zip.

Profiles

ProfileEnginePurpose
`silero-vad`Silero VADVoice 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.ps1

Usage

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 as models/{profileId}/).