CoLLD: Contrastive Layer-to-layer Distillation for Compressing Multilingual Pre-trained Speech Encoders
This addresses the high cost and infeasibility of deploying large speech models for new tasks and on-device applications, representing an incremental improvement over existing compression methods.
The paper tackles the problem of compressing large pre-trained speech encoders for efficient deployment by proposing CoLLD, a knowledge distillation method using masked prediction and contrastive learning, which outperforms prior methods and reduces performance gaps on multilingual speech-to-text translation and recognition benchmarks.
Large-scale self-supervised pre-trained speech encoders outperform conventional approaches in speech recognition and translation tasks. Due to the high cost of developing these large models, building new encoders for new tasks and deploying them to on-device applications are infeasible. Prior studies propose model compression methods to address this issue, but those works focus on smaller models and less realistic tasks. Thus, we propose Contrastive Layer-to-layer Distillation (CoLLD), a novel knowledge distillation method to compress pre-trained speech encoders by leveraging masked prediction and contrastive learning to train student models to copy the behavior of a large teacher model. CoLLD outperforms prior methods and closes the gap between small and large models on multilingual speech-to-text translation and recognition benchmarks.