Models and configuration

Models

Enum	Parameters	Approx download
KittenModel.NanoInt8	15M	28 MB including voices
KittenModel.Nano	15M	59 MB including voices
KittenModel.Micro	40M	44 MB including voices
KittenModel.Mini	80M	83 MB including voices

Voices

KittenVoice.Bella, Jasper, Luna, Bruno, Rosie, Hugo, Kiki, and Leo are available.

Configuration

const tts = await KittenTTS.create({
  // Model and voice defaults apply unless a generation overrides them.
  model: KittenModel.NanoInt8,
  defaultVoice: KittenVoice.Luna,
  speed: 1.05,
  downloadRetries: 4,
  ortNumThreads: 4,
  // Bound sentence-group size for long-text inference.
  maxTokensPerChunk: 400,
  trimTrailingSilence: true,
  silenceThreshold: 0.005,
  maxSilenceTrimMs: 250,
});

Option	Default	Purpose
storageDirectory	app document directory	Cache root
modelBaseURL	Hugging Face	Self-hosted mirror directory
modelFiles	none	Direct local paths or bytes
downloadRetries	4	Total attempts per asset
ortNumThreads	4	ONNX intra-op threads
maxTokensPerChunk	400	Long-text inference limit
trimTrailingSilence	true	Remove near-silent chunk tails
silenceThreshold	0.005	Silence amplitude threshold
maxSilenceTrimMs	250	Maximum removal per chunk
phonemizer	CE	Custom text-to-IPA implementation
forceRedownload	false	Refresh assets during create
ortWasmPath	pinned CDN on web	Self-hosted browser ORT files

Speed is clamped to the supported range of 0.5–2.0.