Perceive and predict: self-supervised speech representation based loss functions for speech enhancement

TL;DR

This paper introduces a novel loss function based on the distance between self-supervised speech feature encodings, which correlates with speech quality and improves speech enhancement performance.

Contribution

It demonstrates that using the distance between self-supervised speech features as a loss function enhances speech enhancement models beyond traditional spectrogram-based losses.

Findings

Feature encoding distance correlates with speech quality and intelligibility.

Proposed loss improves PESQ and STOI scores over traditional methods.

Self-supervised representations are effective in speech enhancement tasks.

Abstract

Recent work in the domain of speech enhancement has explored the use of self-supervised speech representations to aid in the training of neural speech enhancement models. However, much of this work focuses on using the deepest or final outputs of self supervised speech representation models, rather than the earlier feature encodings. The use of self supervised representations in such a way is often not fully motivated. In this work it is shown that the distance between the feature encodings of clean and noisy speech correlate strongly with psychoacoustically motivated measures of speech quality and intelligibility, as well as with human Mean Opinion Score (MOS) ratings. Experiments using this distance as a loss function are performed and improved performance over the use of STFT spectrogram distance based loss as well as other common loss functions from speech enhancement literature is demonstrated using objective measures such as perceptual evaluation of speech quality (PESQ) and short-time objective intelligibility (STOI).