FeaRLESS: Feature Refinement Loss for Ensembling Self-Supervised Learning Features in Robust End-to-end Speech Recognition
This work addresses the challenge of enhancing automatic speech recognition robustness by effectively integrating diverse SSLRs, representing an incremental improvement over using solitary SSLRs.
The study tackled the problem of combining multiple self-supervised learning representations (SSLRs) for robust end-to-end speech recognition by proposing a feature refinement loss to decorrelate and efficiently fuse these features, resulting in improved performance on WSJ and Fearless Steps Challenge corpora compared to systems without the loss.
Self-supervised learning representations (SSLR) have resulted in robust features for downstream tasks in many fields. Recently, several SSLRs have shown promising results on automatic speech recognition (ASR) benchmark corpora. However, previous studies have only shown performance for solitary SSLRs as an input feature for ASR models. In this study, we propose to investigate the effectiveness of diverse SSLR combinations using various fusion methods within end-to-end (E2E) ASR models. In addition, we will show there are correlations between these extracted SSLRs. As such, we further propose a feature refinement loss for decorrelation to efficiently combine the set of input features. For evaluation, we show that the proposed 'FeaRLESS learning features' perform better than systems without the proposed feature refinement loss for both the WSJ and Fearless Steps Challenge (FSC) corpora.