Contrastive Semi-supervised Learning for ASR
This addresses the challenge of improving ASR performance for low-resource and out-of-domain speech data, representing an incremental advancement in semi-supervised learning techniques.
The paper tackles the problem of pseudo-labeling in automatic speech recognition (ASR) suffering from teacher model degradation in low-resource and domain transfer scenarios, proposing Contrastive Semi-supervised Learning (CSL) which reduces word error rate (WER) by up to 19% compared to standard methods in ultra low-resource conditions.
Pseudo-labeling is the most adopted method for pre-training automatic speech recognition (ASR) models. However, its performance suffers from the supervised teacher model's degrading quality in low-resource setups and under domain transfer. Inspired by the successes of contrastive representation learning for computer vision and speech applications, and more recently for supervised learning of visual objects, we propose Contrastive Semi-supervised Learning (CSL). CSL eschews directly predicting teacher-generated pseudo-labels in favor of utilizing them to select positive and negative examples. In the challenging task of transcribing public social media videos, using CSL reduces the WER by 8% compared to the standard Cross-Entropy pseudo-labeling (CE-PL) when 10hr of supervised data is used to annotate 75,000hr of videos. The WER reduction jumps to 19% under the ultra low-resource condition of using 1hr labels for teacher supervision. CSL generalizes much better in out-of-domain conditions, showing up to 17% WER reduction compared to the best CE-PL pre-trained model.