Models and voices
Models
| Model repository | Parameters | Approx size | Guidance |
|---|---|---|---|
KittenML/kitten-tts-nano-0.8-int8 | 15M | 25 MB | Smallest download; validate output for your use case |
KittenML/kitten-tts-nano-0.8 | 15M | 56 MB | Compact fp32 default family |
KittenML/kitten-tts-micro-0.8 | 40M | 41 MB | Balance of quality and footprint |
KittenML/kitten-tts-mini-0.8 | 80M | 80 MB | Highest-quality v0.8 option |
from kittentts import KittenTTS
tts = KittenTTS("KittenML/kitten-tts-micro-0.8")
Passing a name without / automatically uses the KittenML Hugging Face organization.
Voices
| Display name | Embedding ID | Character |
|---|---|---|
| Bella | expr-voice-2-f | Warm and expressive |
| Jasper | expr-voice-2-m | Clear and conversational |
| Luna | expr-voice-3-f | Calm and smooth |
| Bruno | expr-voice-3-m | Deep and steady |
| Rosie | expr-voice-4-f | Bright and friendly |
| Hugo | expr-voice-4-m | Authoritative |
| Kiki | expr-voice-5-f | Lively and energetic |
| Leo | expr-voice-5-m | Relaxed and natural |
The labels describe intended character, not guaranteed prosody. Evaluate recordings using your own text, punctuation, and speed settings.
Cache location
Hugging Face controls the default cache. Pass cache_dir to keep assets in an application-specific location:
tts = KittenTTS(
"KittenML/kitten-tts-nano-0.8",
cache_dir="./.models",
)