电子学报 ›› 2019, Vol. 47 ›› Issue (1): 227-233.DOI: 10.3969/j.issn.0372-2112.2019.01.030

• 科研通信 • 上一篇    下一篇

基于双耳线索编码原理的语音增强方法

陈楠, 鲍长春   

  1. 北京工业大学信息学部, 北京 100124
  • 收稿日期:2017-04-20 修回日期:2018-01-10 出版日期:2019-01-25
    • 作者简介:
    • 陈楠 女,1992年生于北京密云.北京工业大学硕士研究生.主要研究方向为语音增强.E-mail:chennan12@emails.bjut.edu.cn;鲍长春 男,1965年生于内蒙古赤峰.现为北京工业大学教授、博士生导师.主要研究方向为语音与音频信号处理.E-mail:chchbao@bjut.edu.cn
    • 基金资助:
    • 基金项目:国家自然科学基金 (No.61471014)

Speech Enhancement Method Based on Binaural Cues Coding Principle

CHEN Nan, BAO Chang-chun   

  1. Faculty of Information Technology, Beijing University of Technology, Beijing 100124, China
  • Received:2017-04-20 Revised:2018-01-10 Online:2019-01-25 Published:2019-01-25

摘要: 借助双耳线索编码原理,通过构建一个语音和噪声的双耳线索先验码书,本文提出一种单通道语音增强方法.首先,该算法将语音和噪声的双耳线索作为语音和噪声的先验知识,在线下被训练成为先验码书.之后,在线上通过加权码书映射(Weighted CodeBook Mapping,WCBM)算法估计纯净线索参数,最后,利用双耳线索编码原理增强含噪语音.此外,本文采用深度神经网络,即堆栈式自编码器(Stacked Auto-Encoders,SAE)代替WCBM算法估计纯净线索参数,提出了基于深度神经网络的双耳线索语音增强算法.进一步提高了增强算法的性能.客观测试结果表明,本文所提方法优于参考算法.

关键词: 语音增强, 双耳线索编码, 码书驱动, 深度神经网络

Abstract: In this paper,a single channel speech enhancement method is proposed by constructing a priori binaural cue codebook of speech and noise based on binaural cue coding principle.Firstly,as a priori information,the binaural cues of speech and noise are offline trained to form a priori codebook.Then,the weighted codebook mapping (WCBM) algorithm is used to estimate the clean cue.At last,the noisy speech is enhanced with binaural cue coding (BCC) model.Moreover,an estimation method of the clean cue is proposed for further improving performance based on deep neural network,namely stacked auto-encoders (SAE),instead of WCBM algorithm.Objective test results show that the proposed method is superior to the reference methods.

Key words: speech enhancement, binaural cue coding, codebook driven, deep neural network

中图分类号: