End-to-End Integration of Speech Recognition, Speech Enhancement, and Self-Supervised Learning Representation

TL;DR

This paper introduces IRIS, an end-to-end speech recognition model that combines speech enhancement and self-supervised learning to improve robustness in noisy environments, achieving state-of-the-art results on CHiME-4.

Contribution

The paper presents a novel integrated E2E ASR model that combines speech enhancement and SSLR modules, with an efficient training scheme for robust speech recognition.

Findings

Achieves 2.0% WER on CHiME-4 real development set.

Outperforms previous models on single-channel CHiME-4 benchmark.

Demonstrates the effectiveness of combining SE and SSLR modules.

Abstract

This work presents our end-to-end (E2E) automatic speech recognition (ASR) model targetting at robust speech recognition, called Integraded speech Recognition with enhanced speech Input for Self-supervised learning representation (IRIS). Compared with conventional E2E ASR models, the proposed E2E model integrates two important modules including a speech enhancement (SE) module and a self-supervised learning representation (SSLR) module. The SE module enhances the noisy speech. Then the SSLR module extracts features from enhanced speech to be used for speech recognition (ASR). To train the proposed model, we establish an efficient learning scheme. Evaluation results on the monaural CHiME-4 task show that the IRIS model achieves the best performance reported in the literature for the single-channel CHiME-4 benchmark (2.0% for the real development and 3.9% for the real test) thanks to the powerful pre-trained SSLR module and the fine-tuned SE module.