AI4BharatIndicWav2Vec
IndicWav2Vec Pretrained Large
This is the large pretrained IndicWav2Vec checkpoint, published separately from the language-specific recognizers.
Pretraining precedes language-specific recognition
The model learns speech representations across 40 Indian languages.AI4Bharat authors: Towards Building ASR Systems for the Next Billion Users
Add a task head and downstream training
Practical guidance. Use this checkpoint as an acoustic representation starting point. It is not a ready-made transcript endpoint; recognition requires the downstream model and decoding setup.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | indicwav2vec-pretrained-large |
| Terms | MIT.IndicWav2Vec original pretrained and recognition releases |
| Access | The original Fairseq checkpoint route remains accessible.IndicWav2Vec original pretrained and recognition releases |
Related models
- IndicWav2Vec Language Releases
AI4Bharat · IndicWav2Vec
Nine language-specific Fairseq speech recognizers with separate dictionaries, lexicons and decoding models.
Launch date not confirmedDownloadable - IndicWav2Vec Pretrained Base
AI4Bharat · IndicWav2Vec
Original base-sized IndicWav2Vec speech-pretraining checkpoint for downstream adaptation.
Launch date not confirmedDownloadable