IndicWhisper Language Releases
IndicWhisper fine-tunes OpenAI Whisper on Vistaar speech data. The project publishes separate language archives rather than one universal IndicWhisper checkpoint.
Language-specific downloads share one research program
All twelve linked archive routes responded successfully during this review. The Hindi example names a Whisper-medium checkpoint. Odia inference needs a special decoder-language setting because the inherited Whisper language control does not support it directly. Archive contents were not downloaded to audit every internal checkpoint.Vistaar and its twelve IndicWhisper language releases
Keep the chosen language and checkpoint in the deployment record
Practical guidance. A result from the Hindi release does not establish quality for the Odia release. Compare the language-specific system on the recording types that your application receives.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | IndicWhisperIndicWhisper BengaliIndicWhisper GujaratiIndicWhisper HindiIndicWhisper KannadaIndicWhisper MalayalamIndicWhisper MarathiIndicWhisper OdiaIndicWhisper PunjabiIndicWhisper SanskritIndicWhisper TamilIndicWhisper TeluguIndicWhisper Urdu |
| Released languages | Bengali, Gujarati, Hindi, Kannada, Malayalam, Marathi, Odia, Punjabi, Sanskrit, Tamil, Telugu and Urdu.Vistaar and its twelve IndicWhisper language releases |
| Terms | MIT applies to the project's fine-tuned models.Vistaar and its twelve IndicWhisper language releases |
| Training collection | More than 10,700 hours across twelve languages. This is the collection total, not a per-model amount.Vistaar and its twelve IndicWhisper language releases |
Related models
- IndicConformer 600M Multilingual
AI4Bharat ยท IndicConformer
600-million-parameter Conformer speech recognizer covering India's 22 scheduled languages.
Launch date not confirmedRestricted accessDownloadable