Accelerator-Aware Training for Transducer-Based Speech Recognition

Abstract

Machine learning model weights and activations are represented in full-precision during training. This leads to performance degradation in runtime when deployed on neural network accelerator (NNA) chips, which leverage highly parallelized fixed-point arithmetic to improve runtime memory and latency. In this work, we replicate the NNA operators during the training phase, accounting for the degradation due to low-precision inference on the NNA in back-propagation. Our proposed method efficiently emulates NNA operations, thus foregoing the need to transfer quantization error-prone data to the Central Processing Unit (CPU), ultimately reducing the user perceived latency (UPL). We apply our approach to Recurrent Neural Network-Transducer (RNN-T), an attractive architecture for on-device streaming speech recognition tasks. We train and evaluate models on 270K hours of English data and show a 5-7% improvement in engine latency while saving up to 10% relative degradation in WER.

arXiv commentAccepted to SLT 2022

References in corpus (9)

• LSTM: A Search Space Odyssey
• Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Coding
• Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation
• Sequence Transduction with Recurrent Neural Networks
• To prune, or not to prune: exploring the efficacy of pruning for model compression
• Transformer-based Acoustic Modeling for Hybrid Speech Recognition
• Accelerator-Aware Pruning for Convolutional Neural Networks
• Optimizing Speech Recognition For The Edge
• On the quantization of recurrent neural networks