# Improving noise robustness of automatic speech recognition via parallel data and teacher-student learning

## TL;DR

This paper enhances automatic speech recognition robustness to noise by using parallel data, teacher-student learning, and a logits selection method, achieving significant WER reductions on various noisy and clean datasets.

## Contribution

It introduces a novel combination of teacher-student learning with logits selection and large-scale untranscribed data for noise-robust ASR.

## Key findings

- Up to 10.1% relative WER reduction on clean data
- Up to 28.7% relative WER reduction on noisy data
- Effective use of 8000 hours of untranscribed data

## Abstract

For real-world speech recognition applications, noise robustness is still a challenge. In this work, we adopt the teacher-student (T/S) learning technique using a parallel clean and noisy corpus for improving automatic speech recognition (ASR) performance under multimedia noise. On top of that, we apply a logits selection method which only preserves the k highest values to prevent wrong emphasis of knowledge from the teacher and to reduce bandwidth needed for transferring data. We incorporate up to 8000 hours of untranscribed data for training and present our results on sequence trained models apart from cross entropy trained ones. The best sequence trained student model yields relative word error rate (WER) reductions of approximately 10.1%, 28.7% and 19.6% on our clean, simulated noisy and real test sets respectively comparing to a sequence trained teacher.