摘要
前端噪声处理直接关系着语音识别的准确性和稳定性,针对小波去噪算法所分离出的信号不是原始信号的最佳估计,提出一种基于子带谱熵的仿生小波变换(BWT)去噪算法。充分利用子带谱熵端点检测的精确性,区分含噪语音部分和噪声部分,实时更新仿生小波变换中的阈值,精确地区分出噪声信号小波系数,达到语音增强目的。实验结果表明,提出的基于子带谱熵的仿生小波语音增强方法与维纳滤波方法相比,信噪比(SNR)平均提高约8%,所提方法对噪声环境下语音信号有显著的增强效果。
Front end noise processing has a direct impact upon the accuracy and stability of the speech recognition. According to the fact that the signal separated by wavelet denoising algorithm isn't its optimal estimation, a novel Bionic Wavelet Transform (BWT) de-noising algorithm based on subband spectrum entropy was proposed. To achieve the purpose of speech enhancement, the subband spectrum entropy, which has a good accuracy of the endpoint detection, was taken full advantage to distinguish the parts of speech and noise, to real-timely update the threshold of BWT, and to precisely determine the noise signal wavelet coefficients. The experimental results indicate that the Signal-to-Noise Ratio (SNR) of the proposed algorithm is 8% higher than the Wiener filter algorithm. The proposed method has significant enhancement effect on speech signal in noisy environments.
出处
《计算机应用》
CSCD
北大核心
2015年第3期868-871,共4页
journal of Computer Applications
基金
辽宁省教育厅科学计划项目(L2013463)
关键词
语音增强
子带谱熵
仿生小波变换
去噪
阈值
speech enhancement
subband spectrum entropy
Bionic Wavelet Transform (BWT)
denoising
threshold