Deepgram adds 12 languages to Nova-3 speech model, expanding support in Europe and South Asia
Update brings production-ready transcription and Keyterm Prompting to new markets with low-latency streaming and batch support
Speech recognition company Deepgram has expanded its Nova-3 model to support 12 additional monolingual languages, increasing coverage across Southeastern Europe and South Asia. The update is now live in production.
The newly added languages are Bosnian (bs), Croatian (hr), Macedonian (mk), Serbian (sr), Slovenian (sl), and Belarusian (be) in Southeastern Europe, along with Bengali (bn), Marathi (mr), Tamil (ta), Telugu (te), Kannada (kn), and Tagalog (tl) in South Asia.
Deepgram said the expansion delivers Nova-3–level accuracy, production-grade performance, and support for Keyterm Prompting in all 12 languages. The model is designed to handle complex linguistic features, including rich inflexion, variable word order, multiple scripts and regional pronunciation, while maintaining low latency and consistent transcription quality for both real-time and batch workflows.
Keyterm Prompting allows developers to guide transcriptions toward domain-specific language, such as brand names or technical terminology, without retraining the model or creating custom vocabularies.
Related reading
- VIP Play completes technology transformation
- AI spending momentum at Davos points to 2026 inflexion year
- Google.org awards $20m to AI-led science projects tackling health, energy and biodiversity
Deepgram said key terms are applied dynamically at inference time, making the feature well-suited for contact centres, interactive voice response systems, voice agents and analytics applications.
All languages are accessible through the Nova-3 API and available for immediate testing via the Deepgram Playground. The company is offering US$200 in credits to new users, which it said equates to more than 750 hours of transcription or roughly 200 hours of speech-to-text using Nova-3.
The Recap
- Deepgram adds 12 monolingual languages to Nova-3 in production.
- Includes languages from Southeastern Europe and South Asia.
- All languages available through the existing Nova-3 API.
