[ 01 ] // quickstart
[ 02 ] // readme
kitten-tts-2
Kitten TTS 2: a 1.7B speech model that clones a voice from 5-30 seconds of audio, with 47 built-in voices, emotion tags and a 0.95 GB ternary download that runs on the CPU too.
Get it
arcflare gen setup kitten-tts-2 --torch cudaarcflare gen tts "Hello from ArcFlare" -m kitten-tts-2It writes a .wav to the current folder (-o to choose). The model gets its own Python environment, so it can't break another one. The first run downloads the weights. On an AMD GPU under Windows, use --torch-from <python> instead of --torch cuda. See Speech.
Voices and cloning
47 voices: Bruno (the default), Bella, Jasper, Luna, Kiki, Victor and more. Or clone one from 5-30 seconds of a single speaker. No transcript needed.
arcflare gen tts "Hello there" -m kitten-tts-2 --voice Lunaarcflare gen tts "Now it sounds like me." -m kitten-tts-2 --ref me.wavEmotion and expression
--instruct takes one emotion: angry, contemplative, excited, joyful, mundane, nervous, sad, stern, surprised or tender. Tags like <laugh>, <sigh> and (((emphasis))) work inline in the text.
arcflare gen tts "We won <laugh> I can (((hardly))) believe it!" -m kitten-tts-2 --instruct joyfulLanguages
English, plus Arabic, Chinese, French, German, Hindi, Italian, Portuguese, Russian and Spanish through voices named after them. --lang de picks the German voice.
The linear weights are ternary, packed to 1.6 bits each, so the download is 0.95 GB instead of 3.5 GB with identical audio. The licence is free under USD 1M in revenue or funding, with attribution.