Working speech-to-speech prototype

Speak Somali. Hear English — or choose any listed Kulan language.

Choose any language in the current Kulan catalogue for the source and listener. Kulan listens through the browser microphone, translates each completed phrase, then plays the translated language through cloud or device speech where spoken output is supported.

79 Kulan catalogue languages available in both selectors

Changing language automatically stops the current microphone session.

Active pair: SomaliEnglish

Somali → English is ready. Press Start live translation.

Ready

Speaker

Somali

so-SO
Spoken Somali phrases will appear here.

Listener hears

English

en-GB
The English translation will appear here and play automatically when enabled.

1. Listen

The browser recognises Somali through the microphone.

2. Translate

Kulan sends each completed phrase to the existing translation service.

3. Speak

English is played through cloud speech or the device voice fallback.

Next engineering layer

From prototype to Kulan Language Mesh.

The production target is server-side streaming speech recognition, lower latency, automatic source-language detection, terminology profiles, and a separate translated audio stream for every participant so one Arabic speaker can be heard in French, Somali, English and Japanese simultaneously.