R-Spin: Efficient Speaker and Noise-invariant Representation Learning with Acoustic Pieces

Anonymous

TL;DR: A data-efficient self-supervised fine-tuning framework with perturbation-invariant online clustering and acoustic pieces for robust speech recognition.

Abstract: This paper introduces Robust Spin (R-Spin), a data-efficient self-supervised fine-tuning framework for speaker and noise-invariant speech representations by learning discrete acoustic units with speaker-invariant clustering (Spin). R-Spin resolves Spin's issues and enhances content representations by learning to predict acoustic pieces. R-Spin offers a 12X reduction in computational resources compared to previous state-of-the-art methods while outperforming them in severely distorted speech scenarios. This paper provides detailed analyses to show how discrete units contribute to speech encoder training and improving robustness in diverse acoustic environments.

Paper Type: long

Research Area: Speech recognition, text-to-speech and spoken language understanding

Contribution Types: Model analysis & interpretability, Approaches low compute settings-efficiency

Languages Studied: English