Spaces:

---

KittenML

/

KittenTTS-Demo

Like 101

Running

This moves the demo to Kitten TTS 2 and keeps the open-source 0.8 models next to it, in a redesigned page that matches kittenml.com.

Models

• Kitten TTS 2 through the KittenML API: every built-in voice the API lists, with a language filter, one-click emotion and sound tags ([joyful], <laugh>, …) and a speed control. Cloned voices are never offered.
• 0.8 Mini, Micro and Nano, loaded when the Space starts and run on its CPU with the 0.8.1 release, as before.

Limits (the same as the demo on kittenml.com)

• 4 requests a minute per visitor and 12 per network, for every model.
• 500 characters per request.
• Kitten TTS 2 also has an audio allowance: 5 minutes per visitor and 20 per network every six hours.

Before merging: add the secret KITTENML_API_KEY under Settings → Variables and secrets. Without it, the 0.8 models still work and Kitten TTS 2 shows a clear message.

Tested locally on Gradio 6.6.0 and Python 3.12: Kitten TTS 2 (4.8 s of audio in 2.3 s) and 0.8 Mini (4.2 s in 0.2 s), model switching, tags and the language filter, at desktop and phone widths.