Towards One-bit ASR: Extremely Low-bit Conformer Quantization Using Co-training and Stochastic Precision

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Li, Zhaoqing, Xu, Haoning, Jin, Zengrui, Meng, Lingwei, Wang, Tianzi, Wang, Huimeng, Chen, Youjun, Cui, Mingyu, Hu, Shujie, Liu, Xunying
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866913861441945600
author Li, Zhaoqing
Xu, Haoning
Jin, Zengrui
Meng, Lingwei
Wang, Tianzi
Wang, Huimeng
Chen, Youjun
Cui, Mingyu
Hu, Shujie
Liu, Xunying
author_facet Li, Zhaoqing
Xu, Haoning
Jin, Zengrui
Meng, Lingwei
Wang, Tianzi
Wang, Huimeng
Chen, Youjun
Cui, Mingyu
Hu, Shujie
Liu, Xunying
contents Model compression has become an emerging need as the sizes of modern speech systems rapidly increase. In this paper, we study model weight quantization, which directly reduces the memory footprint to accommodate computationally resource-constrained applications. We propose novel approaches to perform extremely low-bit (i.e., 2-bit and 1-bit) quantization of Conformer automatic speech recognition systems using multiple precision model co-training, stochastic precision, and tensor-wise learnable scaling factors to alleviate quantization incurred performance loss. The proposed methods can achieve performance-lossless 2-bit and 1-bit quantization of Conformer ASR systems trained with the 300-hr Switchboard and 960-hr LibriSpeech corpus. Maximum overall performance-lossless compression ratios of 16.2 and 16.6 times are achieved without a statistically significant increase in the word error rate (WER) over the full precision baseline systems, respectively.
format Preprint
id arxiv_https___arxiv_org_abs_2505_21245
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Towards One-bit ASR: Extremely Low-bit Conformer Quantization Using Co-training and Stochastic Precision
Li, Zhaoqing
Xu, Haoning
Jin, Zengrui
Meng, Lingwei
Wang, Tianzi
Wang, Huimeng
Chen, Youjun
Cui, Mingyu
Hu, Shujie
Liu, Xunying
Sound
Audio and Speech Processing
Model compression has become an emerging need as the sizes of modern speech systems rapidly increase. In this paper, we study model weight quantization, which directly reduces the memory footprint to accommodate computationally resource-constrained applications. We propose novel approaches to perform extremely low-bit (i.e., 2-bit and 1-bit) quantization of Conformer automatic speech recognition systems using multiple precision model co-training, stochastic precision, and tensor-wise learnable scaling factors to alleviate quantization incurred performance loss. The proposed methods can achieve performance-lossless 2-bit and 1-bit quantization of Conformer ASR systems trained with the 300-hr Switchboard and 960-hr LibriSpeech corpus. Maximum overall performance-lossless compression ratios of 16.2 and 16.6 times are achieved without a statistically significant increase in the word error rate (WER) over the full precision baseline systems, respectively.
title Towards One-bit ASR: Extremely Low-bit Conformer Quantization Using Co-training and Stochastic Precision
topic Sound
Audio and Speech Processing
url https://arxiv.org/abs/2505.21245