ASR Error Correction

Team Members:

1. Manni Arora
2. Pooja Dattatri
3. Shreyas Ramesh

Abstract

We present a comparative study for solving the problem of ASR error correction. We treat ASR error correction as a post processing task. We have experimented with grammar correction models and GPT-3 to solve this task. We have observed that grammar correction models can help correct contextual errors and if fine-tuned on large volume of data can potentially be very useful for solving the problem of ASR error correction.