English

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

Cryptography and Security 2024-06-07 v2 Computation and Language Sound Audio and Speech Processing

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.

Keywords

Cite

@article{arxiv.2406.02004,
  title  = {Efficiently Train ASR Models that Memorize Less and Perform Better with Per-core Clipping},
  author = {Lun Wang and Om Thakkar and Zhong Meng and Nicole Rafidi and Rohit Prabhavalkar and Arun Narayanan},
  journal= {arXiv preprint arXiv:2406.02004},
  year   = {2024}
}

Comments

Accepted to Interspeech'24

R2 v1 2026-06-28T16:52:27.206Z