Transformer-based models excel in speech recognition. Existing efforts to optimize Transformer inference, typically for long-context applications, center on simplifying attention score calculations. However, streaming speech recognition models usually process a limited number of tokens each time, making attention score calculation less of a bottleneck. Instead, the bottleneck lies in the linear projection layers of multi-head attention and feedforward networks, constituting a substantial portion of the model size and contributing significantly to computation, memory, and power usage. To address this bottleneck, we propose folding attention, a technique targeting these linear layers, significantly reducing model size and improving memory and power efficiency. Experiments on on-device Transformer-based streaming speech recognition models show that folding attention reduces model size (and corresponding memory consumption) by up to 24% and power consumption by up to 23%, all without compromising model accuracy or computation overhead.
@article{arxiv.2309.07988,
title = {Folding Attention: Memory and Power Optimization for On-Device Transformer-based Streaming Speech Recognition},
author = {Yang Li and Liangzhen Lai and Yuan Shangguan and Forrest N. Iandola and Zhaoheng Ni and Ernie Chang and Yangyang Shi and Vikas Chandra},
journal= {arXiv preprint arXiv:2309.07988},
year = {2024}
}