基于小波包的阈值语音去噪算法研究

传统的小波阈值去噪算法会不可避免地造成有用语音信号的损失，达不到很高的信噪比。基于Donoho提出的传统阈值去噪方法，提出了一种新的阈值函数，该闽值函数结合了软阈值函数和灿律阈值函数的优点，不仅克服了硬阈值函数不连续的缺点，而且解决了软阈值函数存在的恒定偏差。实验仿真表明，新阈值函数的语音去噪方法无论在听觉效果上，还是信噪比指标上均明显优于传统的硬、软阈值方法。

Study on Threshold De-noising Technology for Speech Signal Based on Wavelet Packet

Traditional wavelet threshold de-noising algorithm will result the loss of the useful speech signals unavoidable. Eventually the speech signal can＇t reach good signal to noise ratio. Meanwhile, based on traditional wavelet threshold de-noising method which put forward by Donoho, a new threshold function is posed. This threshold function has the advantages of both soft threshold function and μ-rule threshold function. It can not only overcome the shortcoming of hard threshold function＇s discontinuity, but also solve the constant error of hard threshold function. Simulation results show that the method is better than traditional soft and hard threshold method not only in the audio effects but also in the signal to noise ratio.