Deepgram adds 36 languages to Nova-3 speech recognition for batch and streaming
Deepgram says its Nova-3 speech recognition now supports 36 new languages for both pre-recorded audio transcription and streaming. The additions include one Nordic language, Icelandic.
Deepgram announced the expansion in its changelog on 16 November. The models for the new languages are monolingual: each model transcribes one language, which the user selects in the request.
The new languages are Abkhaz, Adyghe, Akan, Albanian, Amharic, Azerbaijani, Basque, Bodo, Central Kurdish, Dhivehi, Eastern Mari, Galician, Hausa, Icelandic, Kabardian, Kabyle, Kalenjin, Kashmiri, Kinyarwanda, Konkani, Kyrgyz, Ganda, Luo, Malayalam, Northern Kurdish, Sindhi, Sinhala, Swahili, Taita, Tswana, Uyghur, Uzbek, Venda, Welsh, Yoruba and Zulu.
Icelandic is the only Nordic language among the additions. Danish was already supported: on 12 November Deepgram released improved Nova-3 models for 13 languages it already supported, including Danish, Arabic and Greek, and separately for Catalan and Japanese. The full language list is on Deepgram’s Models & Languages Overview page.
The expansion covers the monolingual models. In October Deepgram separately announced a multilingual Nova-3 model that transcribes 55 languages with a single model and recognises when the speaker switches language mid-speech.
To use a new language, set the model to nova-3 in the request and put the language code in the language parameter, for example eu for Basque. Users must specify the language themselves: Deepgram’s Python SDK changelog says Nova-3 assumes English when no language is given. Self-hosted customers must request the updated models for download from their Deepgram account representative.
Read more: Models