Efficiently Train ASR Models that Memorize Less and Perform Better with Per-core Clipping

TL;DR

This paper introduces per-core clipping (PCC) and adaptive per-core clipping (APCC) techniques that improve ASR model training by reducing memorization, enhancing convergence, and lowering word error rates.

Contribution

The paper systematically investigates per-core gradient clipping in ASR training and proposes APCC to simplify hyperparameter tuning while improving model robustness and performance.

Findings

PCC effectively reduces unintended memorization in ASR models.

PCC improves convergence rates and reduces word error rates.

APCC streamlines training without additional hyperparameter tuning.

Abstract

Gradient clipping plays a vital role in training large-scale automatic speech recognition (ASR) models. It is typically applied to minibatch gradients to prevent gradient explosion, and to the individual sample gradients to mitigate unintended memorization. This work systematically investigates the impact of a specific granularity of gradient clipping, namely per-core clip-ping (PCC), across training a wide range of ASR models. We empirically demonstrate that PCC can effectively mitigate unintended memorization in ASR models. Surprisingly, we find that PCC positively influences ASR performance metrics, leading to improved convergence rates and reduced word error rates. To avoid tuning the additional hyperparameter introduced by PCC, we further propose a novel variant, adaptive per-core clipping (APCC), for streamlined optimization. Our findings highlight the multifaceted benefits of PCC as a strategy for robust, privacy-forward ASR model training.