AI4BharatIndicSeamless

IndicSeamless

IndicSeamless adapts SeamlessM4T v2 for speech-to-text translation using the BhasaAnuvaad dataset.

Restricted access

Translation and transcription are different uses

The model card describes speech-to-text translation and supplies batch examples. It also recommends Indic-Transcribe for newer transcription work. The noncommercial license remains a separate deployment constraint.IndicSeamless official model card

Define the output language before comparing systems

Practical guidance. A same-language transcript and a translated transcript solve different tasks. Use target-language reference text when evaluating this model's translation output.

Specifications & access

IndicSeamless specifications
DetailValue and conditions
StatusRestricted access
AccessDownloadable
Model identifiersindic-seamless
LicenseCC BY-NC 4.0; gated weights.IndicSeamless official model card
Training filtersAlignment threshold 0.8 and mining threshold 0.6 in the published recipe.IndicSeamless official model card