NVIDIA NeMo stt_en_fastconformer_ctc_large finetuned on open-source Telugu data.
Data sources:
- https://ai4bharat.iitm.ac.in/shrutilipi/
- https://www.openslr.org/66/
- https://huggingface.co/datasets/google/fleurs/viewer/te_in - train
- https://www.iitm.ac.in/donlab/indictts/database
(plus white noise, background noise, pitch-shifted augmentations)
Performance: 0.21 CER on Google FLEURS Test data
You can find the model and the inference interface at https://huggingface.co/spaces/bharathraj-v/fastconformer_ctc_telugu