TriNet: stabilizing self-supervised learning from complete or slow collapse on ASR

TL;DR

TriNet introduces a triple-branch architecture to prevent collapse and stabilize self-supervised learning in ASR, leading to faster training and improved word error rates over existing methods.

Contribution

The paper presents TriNet, a novel architecture that enhances SSL stability and efficiency in ASR by integrating a triple-branch design and pseudo target prediction.

Findings

Stabilizes SSL pre-training in ASR tasks.

Achieves 6.06% relative WERR over Data2vec.

Speeds up the pre-training process.

Abstract

Self-supervised learning (SSL) models confront challenges of abrupt informational collapse or slow dimensional collapse. We propose TriNet, which introduces a novel triple-branch architecture for preventing collapse and stabilizing the pre-training. TriNet learns the SSL latent embedding space and incorporates it to a higher level space for predicting pseudo target vectors generated by a frozen teacher. Our experimental results show that the proposed method notably stabilizes and accelerates pre-training and achieves a relative word error rate reduction (WERR) of 6.06% compared to the state-of-the-art (SOTA) Data2vec for a downstream benchmark ASR task. We will release our code at https://github.com/tencent-ailab/.