Silero Publishes Stable, Fast, High-Quality and Affordable TTS for 29 More Languages of Russia

Silero, a company known for its speech technologies, has published two new text-to-speech models supporting 29 more languages of Russia and neighboring regions. The first model covers mainly Turkic languages (14 languages, 33 voices), and the second covers Caucasian languages (15 languages, 31 voices), including many Abkhaz-Adyghe and Nakh-Dagestanian languages. The models use a concatenation of official Cyrillic alphabets, support SSML, and are optimized similarly to previous models. They cannot control stress placement, which is noted as a limitation. The languages were selected based on data availability and the need to cover major language groups. The company also provides code examples and notebooks for easy usage, and notes that the models support cross-lingual generalization. The release aims to fill gaps in coverage of languages of the Caucasus, Dagestan, and Chechnya.