Improving Speech Recognition for Indic Languages using Language Model

TL;DR

This paper enhances Indic language speech recognition by fine-tuning wav2vec 2.0 models and applying language models, significantly reducing error rates and enabling domain-specific transcription without retraining.

Contribution

It introduces a method to improve Indic language ASR using fine-tuned wav2vec 2.0 and diverse language models, achieving substantial error reduction and domain adaptability.

Findings

CER reduced by over 28% with LM

WER reduced by about 36% with LM

Large LM not always better than diverse LM

Abstract

We study the effect of applying a language model (LM) on the output of Automatic Speech Recognition (ASR) systems for Indic languages. We fine-tune wav2vec $2.0$ models for $18$ Indic languages and adjust the results with language models trained on text derived from a variety of sources. Our findings demonstrate that the average Character Error Rate (CER) decreases by over $28$ \% and the average Word Error Rate (WER) decreases by about $36$ \% after decoding with LM. We show that a large LM may not provide a substantial improvement as compared to a diverse one. We also demonstrate that high quality transcriptions can be obtained on domain-specific data without retraining the ASR model and show results on biomedical domain.