AI4BharatIndicConformer
IndicConformer Assamese 120M
The Assamese IndicConformer release has a 120M-parameter encoder with 17 Conformer blocks and model width 512.
Restricted access
Both decoders use the language-specific acoustic model
AI4Bharat's NeMo fork is part of the documented loading procedure. The card demonstrates CTC and RNNT decoding. Its introductory sample uses a shortened repository spelling; the actual published repository includes ctc in the name.IndicConformer Assamese 120M official model card
Use the published repository identifier
Practical guidance. Resolve the model package before comparing decoder output. A copied sample name that does not match the repository can prevent loading without saying anything about recognition quality.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Restricted access |
| Access | Downloadable |
| Model identifiers | indicconformer_stt_as_hybrid_ctc_rnnt_large |
| Language | AssameseIndicConformer Assamese 120M official model card |
| Terms | MIT; gated NeMo checkpoint.IndicConformer Assamese 120M official model card |
| Input | Mono audio; preprocessing example uses 16,000 Hz.IndicConformer Assamese 120M official model card |
Related models
- IndicConformer 600M Multilingual
AI4Bharat · IndicConformer
600-million-parameter Conformer speech recognizer covering India's 22 scheduled languages.
Launch date not confirmedRestricted accessDownloadable