摘要
目前消除环境噪声提高语音信号质量是语音识别技术研究的重要方向之一。小波阈值去噪法是一种常见的得到广泛应用的语音增强方法,然而传统的小波阈值去噪法存在许多不足,如小波系数经过硬阈值函数处理后在正负阈值处不连续,重构后的信号会产生振荡;经过软阈值函数去噪后的语音过于平滑而丢失语音的突变部分影响语音信号的质量。文中将小波去噪应用于语音识别系统的前端处理中,提出一种新的阈值函数,对小波阈值去噪方法中传统阈值函数的缺点进行改进,将带噪语音信号去噪后再进行识别。实验结果表明,新提出的方法使带背景噪声的语音识别系统的识别性能得到显著的提高。
At present,removal of environmental noise to improve the quality of voice signal has become an important direction for voice recognition technology. Wavelet threshold denoising method is a method widely applied to speech enhancement. However,the traditional wavelet threshold denoising method has many deficiencies,like wavelet coefficients processed by the hard threshold function is discontinu- ous in the positive and negative threshold values, resulting in the reconstructed signal oscillation; denoised voice processed by soft thresh- old function is too smooth and lose voice mutation part,which would affect the quality of the voice signal. It applied wavelet denoising to the front-end processing of speech recognition system,and proposed a new threshold function to improve the shortcomings of the tradi- tional threshold function of the wavelet threshold denoising method. It denoised speech signal with noise, and then identified. The experi- mental results show that the new proposed method makes speech recognition performance of the system with background noise significant- ly improved.
出处
《计算机技术与发展》
2013年第5期231-234,共4页
Computer Technology and Development
基金
国家自然科学基金资助项目(51008143)
关键词
语音识别
噪声
小波变换
阈值函数
speech recognition
noise
wavelet transform
threshold function
