Speak Turkish, see it written.

ufakzeka-kulak is a small Turkish speech recognition model by ufak AI, and it runs entirely in your browser.

Before you start

This is a research model; check the text it writes.

This model

What it does

ufakzeka-kulak is a small Turkish speech recognition model that ufak AI trained from scratch. It writes out Turkish speech from the microphone as Turkish text, with capital letters, full stops, commas, question marks, apostrophes and hyphens inside words.

What it is not

It is not a chatbot; it does not answer questions or translate. It does not tell speakers apart. It writes out each stretch of speech as a whole after you pause, so words do not appear one by one while you speak. The text it writes can be wrong.

How it runs

The page runs the open-source sherpa-onnx runtime (Apache-2.0) in your browser through WebAssembly. Silero VAD (MIT) finds where speech starts and ends. When you pause for about half a second, that stretch goes to the model with a short margin of the audio around it, and the text appears with the time it took to write it out.

Licence texts of the third-party code

Your voice and the text are sent nowhere; there is no server that receives them. The model files come from this site, the fonts load from Google Fonts, and visits are counted with Cloudflare Web Analytics, without cookies.