摘要
针对谱减法用于语音增强残留噪声的问题,提出了一种新的阈值确定方法对小波系数进行分层阈值处理,并用短时能量和短时过零率对阈值处理过的各小波系数进行端点检测,利用检测结果确定语音信号的起止点并把起止点之间的部分取出形成新的小波系数向量,然后把经上述处理过的各层小波系数重构,以达到去掉噪声的目的。实验结果表明,与谱减法相比,文中方法可以明显提高语音信号信噪比和清晰度。
To the question that residual noise exists in the result of spectral subtraction in speech enhancement,introduces a new method of threshold selection to process wavelet coefficient by gradient threshold processing,detect endpoints of all wavelet coefficient that has been processed by gradient threshold by short-time energy and short-time zero cross rate to determine the start point and end point of the speech signal,take the part between the two points to form a new wavelet coefficient vector and reconstruct all wavelet coefficient to filter the noise.Experimental results indicate that signal to noise ratio and articulation can be improved significantly by using this method comparing with spectral subtraction.
出处
《煤炭技术》
CAS
北大核心
2010年第8期54-56,共3页
Coal Technology
基金
河南省教育厅基础研究项目(2008B120011)
关键词
分层阈值
阈值选择
语音增强
gradient threshold
threshold selection
speech enhancement
