The development can accurately recognize languages such as Kazakh, Turkish, Uzbek, Kyrgyz, Azerbaijani, and Tatar, as well as mixed speech using Turkic and Russian words. This significantly improves the quality of voice technologies used in business and government institutions.
Previously, Central Asian countries used international ASR models designed for the English language, which reduced recognition accuracy due to the specifics of Turkic phonetics and pronunciation. The new model, developed from scratch for live communication, demonstrates resilience to background noise and pronunciation peculiarities, and can also handle switches between languages in a single sentence.
The project was implemented using Microsoft’s GPU infrastructure with the support of Astana Hub and the Ministry of Digital Development, Innovations, and Aerospace Industry of the Republic of Kazakhstan.
