Language Expansion

Deepgram's Nova-3 speech-to-text model now supports six additional languages and language variants:

New Languages:

• Chinese (Cantonese, Traditional): zh-HK
• Thai: th, th-TH

Improved Models:

• Bengali (bn)
• Marathi (mr)
• Tamil (ta)
• Telugu (te)

Implementation

To use the new language support, set model="nova-3" in your API request and specify the relevant language code. For example, to transcribe Cantonese audio, use language="zh-HK".

For a complete list of supported languages and detailed implementation guidance, refer to the Models and Language Overview documentation.

Impact

This expansion brings Nova-3 coverage to critical markets in Asia, enabling developers to build speech-to-text applications for Cantonese and Thai speakers, while also improving accuracy for South Asian languages.