基于DFRCT滤波器组自适应语音增强算法

Method for speech enhancement based on discrete fractional cosine transform and sub-band adaptive filter

提出一种改进的语音增强方法,将带噪语音信号进行子带分解,再对子带信号进行离散分数余弦变换(DFRCT)域滤波,利用了DFRCT良好的正交特性,且自适应滤波采用最小均方(LMS)算法。对滤波后的信号进行DFRCT逆变换得到增强后的子带语音信号,合成增强后的语音信号。仿真结果表明,该算法在减少输入信号自相关程度的基础上,提高了收敛速度,减少了计算时间(约10s),增强后的语音信号的分段信噪比(SegSNR)和PESQ值都有所提高,具有良好的语音增强效果。

This paper presents an improved speech enhancement method, this method decomposes the noisy speech signal into sub-band signals, then filters the sub-band signals using the adaptive filter in Discrete Fractional Cosine Transform （DFRCT） which has good orthogonal properties, and adaptive filter uses Least Mean Square （LMS） algorithm. It obtains the enhanced sub-band speech signals by using inverse discrete fractional cosine transform to the filtered signal. It composes all enhanced sub-bands signal. Simulation experiments indicate that the proposed algorithm reduces the computing time about 10 s and possesses good performance both in SegSNR and PESQ tests.