# 4-bit Quantization of LSTM-based Speech Recognition Models

arXiv: https://arxiv.org/abs/2108.12074 · HTML: https://mlsanity.com/paper/2108.12074/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- Builds on 19 of its 32 references that are in the map.
- First posted on arXiv 27 Aug 2021.

## Builds on (most cited first, 19 of 19)

- [Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation](https://mlsanity.com/paper/1308.3432/) (2013)
- [Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour](https://mlsanity.com/paper/1706.02677/) (2017)
- [SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition](https://mlsanity.com/paper/1904.08779/) (2019)
- [Quantizing deep convolutional networks for efficient inference: A whitepaper](https://mlsanity.com/paper/1806.08342/) (2018)
- [Quantized Neural Networks: Training Neural Networks with Low Precision Weights and Activations](https://mlsanity.com/paper/1609.07061/) (2016)
- [Streaming End-to-end Speech Recognition For Mobile Devices](https://mlsanity.com/paper/1811.06621/) (2018)
- [Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation](https://mlsanity.com/paper/2004.09602/) (2020)
- [Improving RNN Transducer Modeling for End-to-End Speech Recognition](https://mlsanity.com/paper/1909.12415/) (2019)
- [Advancing RNN Transducer Technology for Speech Recognition](https://mlsanity.com/paper/2103.09935/) (2021)
- [English Conversational Telephone Speech Recognition by Humans and Machines](https://mlsanity.com/paper/1703.02136/) (2017)
- [Single headed attention based sequence-to-sequence model for state-of-the-art results on Switchboard](https://mlsanity.com/paper/2001.07263/) (2020)
- [Optimizing Speech Recognition For The Edge](https://mlsanity.com/paper/1909.12408/) (2019)
- [Alternating Multi-bit Quantization for Recurrent Neural Networks](https://mlsanity.com/paper/1802.00150/) (2018)
- [Large-Batch Training for LSTM and Beyond](https://mlsanity.com/paper/1901.08256/) (2019)
- [Bridging the Accuracy Gap for 2-bit Quantized Neural Networks (QNN)](https://mlsanity.com/paper/1807.06964/) (2018)
- [Effective Quantization Methods for Recurrent Neural Networks](https://mlsanity.com/paper/1611.10176/) (2016)
- [Learning Recurrent Binary/Ternary Weights](https://mlsanity.com/paper/1809.11086/) (2018)
- [Embedding-Based Speaker Adaptive Training of Deep Neural Networks](https://mlsanity.com/paper/1710.06937/) (2017)
- [A Highly Efficient Distributed Deep Learning System For Automatic Speech Recognition](https://mlsanity.com/paper/1907.05701/) (2019)

## Built on by (most cited first, 8 of 8)

- [Recent Advances in End-to-End Automatic Speech Recognition](https://mlsanity.com/paper/2111.01690/) (2021)
- [4-bit Conformer with Native Quantization Aware Training for Speech Recognition](https://mlsanity.com/paper/2203.15952/) (2022)
- [Mixed Precision Low-bit Quantization of Neural Network Language Models for Speech Recognition](https://mlsanity.com/paper/2112.11438/) (2021)
- [Sub-8-bit quantization for on-device speech recognition: a regularization-free approach](https://mlsanity.com/paper/2210.09188/) (2022)
- [Outlier Reduction with Gated Attention for Improved Post-training Quantization in Large Sequence-to-sequence Speech Foundation Models](https://mlsanity.com/paper/2406.11022/) (2024)
- [Accelerating Inference and Language Model Fusion of Recurrent Neural Network Transducers via End-to-End 4-bit Quantization](https://mlsanity.com/paper/2206.07882/) (2022)
- [Quantizing Recursive Reasoning Models](https://mlsanity.com/paper/2607.16237/) (2026)
- [BILLNET: A Binarized Conv3D-LSTM Network with Logic-gated residual architecture for hardware-efficient video inference](https://mlsanity.com/paper/2501.14495/) (2025)

## Abstract (arXiv)

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%.