arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2302.13451cs.SDcs.CLcs.LGeess.AS
收藏

用于流式自监督语音表征学习的低延迟注意力模块

A low latency attention module for streaming self-supervised speech representation learning

Dolby Laboratories(杜比实验室)

Jianbo Ma, Siqi Pan, Deepak Chandran, Andrea Fanelli, Richard Cartwright

AI总结:

本文提出一种低延迟注意力模块,包含流式注意力和低延迟流式注意力,用于流式自监督语音表征学习,在保持低延迟的同时显著降低词错误率。

摘要:

Transformer是深度学习中的基本构建模块,而注意力机制是Transformer的核心组件。自监督语音表征学习(SSRL)是Transformer架构的一个典型应用场景。由于Transformer具有非因果(acausal)特性,其在SSRL中的应用主要集中在非因果任务上。然而,许多媒体处理问题(如语音处理)需要实时解决方案。在本文中,我们提出了一种注意力模块的实现,该实现使得SSRL架构的训练具有低计算和内存需求,同时支持低且固定的延迟实时推理。本文提出的注意力模块包含两个组件:流式注意力(SA)和低延迟流式注意力(LLSA)。SA是我们提出的高效流式SSRL实现方案,而LLSA解决了其他流式注意力架构(如掩蔽非因果注意力(MAA))的延迟累积问题,即使堆叠多层也能保证延迟等于一层。我们通过训练一个以自动语音识别(ASR)为下游任务的流式SSRL模型,对普通注意力(此处称为非因果注意力(AA))、SA和LLSA进行了比较分析。在librispeech-clean-100上训练并在librispeech-test-clean上测试时,我们的低延迟注意力模块的词错误率(WER)为5.84%,相较于MAA(WER=13.82%)有显著改进。我们的实现还将推理延迟从1.92秒降低到0.16秒。所提出的低延迟模块保留了许多传统非因果Transformer的优势,同时具备适用于实时流式应用的延迟特性。

英文摘要:

The transformer is a fundamental building block in deep learning, and the attention mechanism is the transformer's core component. Self-supervised speech representation learning (SSRL) represents a popular use-case for the transformer architecture. Due to transformers' acausal behavior, the use of transformers for SSRL has been predominantly focused on acausal applications. However, several media processing problems, such as speech processing, require real-time solutions. In this paper, we present an implementation of the attention module that enables training of SSRL architectures with low compute and memory requirements, while allowing real-time inference with low and fixed latency. The attention module proposed in this paper includes two components, streaming attention (SA) and low-latency streaming attention (LLSA). The SA represents our proposal for an efficient streaming SSRL implementation, while the LLSA solves the latency build-up problem of other streaming attention architectures, such as the masked acausal attention (MAA), guaranteeing a latency equal to one layer even when multiple layers are stacked. We present a comparative analysis between the vanilla attention, which we will refer here as acausal attention (AA), the SA, and the LLSA, by training a streaming SSRL with automatic speech recognition as downstream task. When training on librispeech-clean-100 and testing on librispeech-test-clean, our low-latency attention module has a word error rate (WER) of 5.84%, which represents a significant improvement over the MAA (WER = 13.82%). Our implementation also reduces the inference latency from 1.92 to 0.16 seconds. The proposed low-latency module preserves many of the benefits of conventional acausal transformers, but also enables latency characteristics that make it applicable to real-time streaming applications.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑