English

4-bit Quantization of LSTM-based Speech Recognition Models

Computation and Language 2021-08-30 v1 Machine Learning Sound Audio and Speech Processing

Abstract

We investigate the impact of aggressive low-precision representations of weights and activations in two families of large LSTM-based architectures for Automatic Speech Recognition (ASR): hybrid Deep Bidirectional LSTM - Hidden Markov Models (DBLSTM-HMMs) and Recurrent Neural Network - Transducers (RNN-Ts). Using a 4-bit integer representation, a na\"ive quantization approach applied to the LSTM portion of these models results in significant Word Error Rate (WER) degradation. On the other hand, we show that minimal accuracy loss is achievable with an appropriate choice of quantizers and initializations. In particular, we customize quantization schemes depending on the local properties of the network, improving recognition performance while limiting computational time. We demonstrate our solution on the Switchboard (SWB) and CallHome (CH) test sets of the NIST Hub5-2000 evaluation. DBLSTM-HMMs trained with 300 or 2000 hours of SWB data achieves <<0.5% and <<1% average WER degradation, respectively. On the more challenging RNN-T models, our quantization strategy limits degradation in 4-bit inference to 1.3%.

Keywords

Cite

@article{arxiv.2108.12074,
  title  = {4-bit Quantization of LSTM-based Speech Recognition Models},
  author = {Andrea Fasoli and Chia-Yu Chen and Mauricio Serrano and Xiao Sun and Naigang Wang and Swagath Venkataramani and George Saon and Xiaodong Cui and Brian Kingsbury and Wei Zhang and Zoltán Tüske and Kailash Gopalakrishnan},
  journal= {arXiv preprint arXiv:2108.12074},
  year   = {2021}
}

Comments

5 pages, 3 figures, Andrea Fasoli and Chia-Yu Chen equally contributed to this work. Paper accepted to Interspeech 2021

R2 v1 2026-06-24T05:27:29.306Z