VoxaTrace/com.musicmuni.voxatrace.calibra.model/VADModelProvider

VADModelProvider

Type-safe model provider for VAD backends.

Each subclass carries the appropriate model loading closure for its backend. Backend is inferred from the provider type, enabling unified API.

Zero-Config Usage (Recommended)

After VT.initialize(), models auto-load from bundled assets:

// Auto-loads bundled Silero model
CalibraVAD.create(VADModelProvider.speech())

// Auto-loads bundled SwiftF0 model
CalibraVAD.create(VADModelProvider.singingRealtime())

// No model required
CalibraVAD.create(VADModelProvider.general)

Custom Model Loading

Pass a custom loader to override bundled models:

CalibraVAD.create(VADModelProvider.speech { loadCustomSileroModel() })

Inheritors

General

Speech

SingingRealtime

Types

Companion

object Companion

General

data object General : VADModelProvider

GENERAL backend - no model required. Uses RMS-based heuristics for basic voice detection.

SingingRealtime

data class SingingRealtime(val modelProvider: () -> ByteArray) : VADModelProvider

SINGING_REALTIME backend - single SwiftF0 ONNX model. Low-latency pitch-based singing detection.

Speech

data class Speech(val modelProvider: () -> ByteArray) : VADModelProvider

SPEECH backend - single Silero ONNX model. State-of-the-art accuracy for speech detection.