AIModels

AI model presets for selective downloading.

Use these presets with VT.initialize(preload = ...) to control which models are downloaded at initialization. Models not preloaded will be downloaded on-demand when first used.

Usage

// Default: Just pitch model (recommended for most apps)
VT.initialize(proxyEndpoint, context)

// Pitch + Speech VAD
VT.initialize(
    proxyEndpoint,
    context,
    preload = setOf(AIModels.Pitch.REALTIME, AIModels.VAD.SPEECH)
)

// All models
VT.initialize(proxyEndpoint, context, preload = AIModels.ALL)

// No preload (fully lazy, download on first use)
VT.initialize(proxyEndpoint, context, preload = AIModels.NONE)

Model Categories

• Pitch: Models for pitch detection (CalibraPitch)

• VAD: Models for voice activity detection (CalibraVAD)

Functional Presets vs Algorithm Access

Each category provides:

• Functional presets (recommended): Describe the use case (e.g., REALTIME, SPEECH)

• Algorithm access: Direct access to specific algorithms (e.g., SWIFT_F0, SILERO)

Functional presets are recommended as they may be updated to use better algorithms in future SDK versions without breaking your code.