Large Language Model-based metric for Automatic Speech Recognition
Master’s Thesis-I (Nationwide project - Bhashini, NLTM) at the Computational Speech and Language Technologies Lab, IIT Bombay, Guide: Prof. Preethi Jyothi
Summary: This project aimed to develop LASER, a finegrained LLM-based metric for Automatic Speech Recognition.
Paper published at the Empirical Methods in Natural language Processing (EMNLP) 2025 (Main conference)
- LASER handles linguistic nuances of Indian Languages, transfers across languages and correlates well with human scores
- Achieved 88.69% accuracy on finetuning Llama 3-8B to get a standardized and efficient pipeline to execute our metric