Karakalpak Voice — text-to-speech with voice cloning
Karakalpak text-to-speech with voice cloning, delivered as an API at karakalpakvoice.uz so other products can simply speak the language.
The challenge
Karakalpak had no usable synthetic voice. Every product that needed to speak the language — accessibility tools, public announcement systems, education software, kiosks — had nothing to call, and the speech data to train on barely existed in public form.
Our approach
Collect the speech data first
Recording and cleaning Karakalpak speech, because no dataset of usable size and quality was available to start from.
Train voice, then cloning
A neural TTS model for natural Karakalpak speech, extended with voice cloning so a specific voice can be reproduced from a short sample.
Ship it as an API
Released through karakalpakvoice.uz so institutions and developers can integrate it directly instead of commissioning their own model.
The working demo lives on the product site, not here: open karakalpakvoice.uz to try the voices and read the API docs.