This repository will host the code for our paper:
RECAST: Retrieval-Augmented Contextual ASR via Decoder-State Keyword Spotting
Ashish Mittal, Sunita Sarawagi, Preethi Jyothi
IBM Research & IIT Bombay
RECAST introduces a lightweight retrieval-augmented framework for contextual ASR.
It leverages decoder-state embeddings of a pretrained ASR model to retrieve and bias relevant keywords from large dictionaries, achieving strong improvements in WER and entity recall across diverse domains and languages.
🚧 Code will be added soon. Stay tuned!
If you find this work useful, please cite:
@inproceedings{mittal2025recast,
title={RECAST: Retrieval-Augmented Contextual ASR via Decoder-State Keyword Spotting},
author={Mittal, Ashish and Sarawagi, Sunita and Jyothi, Preethi},
booktitle={Proceedings of ACL},
year={2025}
}