Speech enhancement method based on amplitude phase mixed feature intersection
卿朝进 1付小伟 1唐书海1
扫码查看
点击上方二维码区域，可以放大扫码查看
作者信息
1. 西华大学电气与电子信息学院,四川成都 610039
折叠
摘要
为充分利用含噪语音信号的相位特征信息及其与幅度信息的相关性,提出一种幅度和相位混合特征交叉的单通道语音增强方法.提取含噪信号的对数功率谱和相位特征,依次交叉排列;计算复数掩模,将复数掩模的实虚部依次交叉以保持对称输入特征;在此基础上,构建深度编解码器网络(amplitude phase deep encoder decoder network,APDEDN)增强语音质量.实验结果表明,相较单一特征方法,提出方法获得了语音质量感知评估评分和短时目标可懂度上的改善.
Abstract
To make full use of the phase feature information from the noisy speech signal and its correlation with amplitude infor-mation,a mixed amplitude and phase features-based single channel speech enhancement method was presented.The logarithmic power spectrum and phase features of the noisy signals were extracted and arranged alternately.The complex mask was calculated,and the real and imaginary parts of the calculated complex mask were alternately arranged to maintain the features of input sym-metry.Based on these processing,an amplitude phase deep encoder decoder network(APDEDN)was constructed to enhance speech quality.Compared with the single feature method,experimental results show that the proposed method improves both the speech quality perception evaluation score and short time objective intelligibility.
关键词
语音增强/特征交叉/特征提取/混合特征/复数掩模/编解码器/深度学习
Key words
speech enhancement/feature intersection/feature extraction/mixed feature/complex ideal ratio mask/encoder-decoder/deep learning
维普
万方数据