Kazakhstan unveils largest automatic speech recognition model for Turkic Languages

Kazakhstan has introduced the largest automatic speech recognition (ASR) model developed specifically for Turkic languages, marking a major milestone in the region’s artificial intelligence ecosystem, Azernews reports, citing Kazakhstan’s Ministry of Artificial Intelligence and Digital Development.

The ASR system was developed by the startup Cybernet AI.

The project is described as the first full-scale artificial intelligence model of its kind in Central Asia. It was created with the linguistic and phonetic characteristics of Turkic languages in mind, addressing long-standing challenges in speech recognition accuracy.

Development of the model was supported by Astana Hub and Kazakhstan’s Ministry of Artificial Intelligence and Digital Development, using Microsoft’s GPU infrastructure.

The ASR model covers Kazakh, Turkish, Uzbek, Kyrgyz, Azerbaijani, and Tatar languages. In addition, it is capable of accurately recognizing mixed Turkic–Russian speech, which is widely used in everyday conversations, business communications, and public services across the region.

Until now, Central Asian countries have largely relied on foreign ASR models primarily optimized for English. These systems often struggled with the phonetic features and accents of Turkic languages, resulting in lower recognition accuracy.

Cybernet AI stated that the new model was developed within the region and for the region, based on real spoken language patterns rather than theoretical datasets.

The AI solution is aimed at banks, telecom operators, contact centers, logistics companies, and government institutions. According to estimates, it can significantly reduce customer service costs while enabling 24/7 voice support.

The system is also reported to be resilient to background noise, accents, and code-switching within a single sentence, maintaining consistent accuracy across sectors ranging from fintech to public administration.