A Method of Multi-Scale Forward Attention Model for Speech Recognition

doi:10.3969/j.issn.0372-2112.2020.07.002

您当前的位置：

首页 >

文章列表页 >

A Method of Multi-Scale Forward Attention Model for Speech Recognition

更新时间：2025-07-08

- A Method of Multi-Scale Forward Attention Model for Speech Recognition
- Acta Electronica Sinica Vol. 48, Issue 7, Pages: 1255-1260(2020)
- 作者机构：
  
  哈尔滨工业大学计算机科学与技术学院,黑龙江,哈尔滨,150001
- 作者简介：
- 基金信息：
  
  National Key Research and Development Program (No.2017YFB1002102)
- DOI：10.3969/j.issn.0372-2112.2020.07.002
  CLC： TN912.34
- Published Online：25 July 2020，
  
  Published：2020
- 稿件说明：
移动端阅览
A Method of Multi-Scale Forward Attention Model for Speech Recognition[J]. Acta Electronica Sinica, 2020, 48(7): 1255-1260.
DOI：

A Method of Multi-Scale Forward Attention Model for Speech Recognition[J]. Acta Electronica Sinica, 2020, 48(7): 1255-1260. DOI： 10.3969/j.issn.0372-2112.2020.07.002.

摘要

注意力模型是当前语音识别中的主流模型，然而其存在一个缺点，即当前时刻的注意力模型可能产生异常得分.为此，本文首先提出前向注意力模型，其采用上一时刻正常注意力得分平滑当前时刻异常得分.接着通过对上一时刻的注意力得分添加约束因子来对前向注意力模型进行优化，达到自适应平滑的目的.最后，在优化模型基础上提出多尺度前向注意力模型，其通过引入多尺度模型来对不同等级的语音基元进行建模，进而将所得到的不同等级目标向量进行融合，以达到解决注意力得分异常值的目的.采用SwitchBoard作为训练集，Hub5'00作为测试集进行实验，相比于基线系统，多尺度前向注意力模型的词错误率（Word Error Rate，WER）相对降低14.28%.

Abstract

Attention-based model is a popular model in speech recognition

however it has a disadvantage that the attention-based model may produce abnormal scores. To solve this problem

this paper first proposes a forward attention model

which adopts normal attention score at the previous moment to smooth the abnormal score at the current moment. Then

the model is optimized to add constraint factors to the attention score at the previous moment to achieve the purpose of adaptive smoothing of the above abnormal scores. Then

a multi-scale forward attention model is proposed on the above model. This model introduces a multi-scale method to model the speech primitives of different levels

and then fuses the target vectors of different levels to solve the outliers of attention score. In the experiment

SwitchBoard is adopted as the training set and Hub5'00 as the test set. Compared with the baseline system

the Word Error Rate (WER) of the proposed system decreased by 14.28% relatively.

关键词

Keywords

references

Views

262

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Low-Light Small Target Detection Method Combining Feature Fusion Enhancement and Detail Features

Unsupervised Monocular Depth Estimation Based on Scale Clue Enhancement

Medical Image Segmentation Based on Multi‑Scale Convolution Modulation

An In-Vehicle Interaction Speech Enhancement and Recognition Method Based on Lightweight Models in Complex Environment

Remote Sensing Image Segmentation of Around Plateau Lakes Based on Multi-Attention Fusion

Related Author

JIANG Ze-tao

CHENG Liu-ming

YANG Jian-chen

QU Yi

CHEN Ying

ZHOU Xin-min

XIONG Zhi-mou

SHI Chang-fa

Related Institution

Guangxi Key Laboratory of Image and Graphics Processing and Intelligent Processing, Guilin University of Electronic Technology

Key Laboratory of Advanced Process Control for Light Industry (Ministry of Education), Jiangnan University

Changsha Social Laboratory of Artificial Intelligence， Hunan University of Technology and Business

School of Intelligent Engineering and Intelligent Manufacturing， Hunan University of Technology and Business

School of Computer Science， Hunan University of Technology and Business

⁰