Model reference · open weights

indicconformer_stt_hi_hybrid_ctc_rnnt_large

indicconformer_stt_hi_hybrid_ctc_rnnt_large is an open-weight audio or speech model from ai4bharat.

• IndicConformer is a large-scale automatic speech recognition model developed by ai4bharat for transcribing Hindi audio.
• It features a conformer encoder with 120 million parameters and 17 blocks, paired with a hybrid CTC-RNNT decoder.
• The model accepts 16 kHz mono-channel audio input and is released under the MIT license.

Summary of the ai4bharat/indicconformer_stt_hi_hybrid_ctc_rnnt_large model card, 2026-10-01

Hugging Face ↗