Indic-Transcribe Flex
Flex is the joint speech-recognition release with three output modes and 27 supported languages. Its coverage includes Chhattisgarhi and Haryanvi as well as Bhili and Bhojpuri.
Restricted access
Script selection is part of the model's decoding task
Task tokens select the language and output mode. The card recommends native-script output for production and points to Core for higher transcription accuracy. The vocabulary contains 7,152 tokens. NeMo and Hugging Face packages represent the same trained release.Indic-Transcribe Flex official model card
Match transcript format to the next application
Practical guidance. Mixed-script output can reduce later text conversion for products that retain English names. Compare all required script modes on the same recordings; success in native script does not establish the quality of Romanized output.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Restricted access |
| Access | Downloadable |
| Model identifiers | indic-transcribe-flex |
| Architecture | 1.2B parameters, based on nvidia/canary-1b-v2; 32 encoder layers and 24 decoder layers.Indic-Transcribe Flex official model card |
| Input | Single-speaker, 16 kHz mono audio.Indic-Transcribe Flex official model card |
| Script controls | Native script, mixed script with Latin English and numerals, or Romanized text.Indic-Transcribe Flex official model card |
| Access and terms | Gated weights under Indic Open Model License v1.0; upstream terms also apply.Indic Open Model License v1.0 |
Related models
- Indic-Transcribe Core
Bodhan AI ยท Indic-Transcribe
1.2-billion-parameter Canary-derived speech model that transcribes 25 languages in their native scripts.
Launch date not confirmedRestricted accessDownloadable