介绍了改进谱减法的原理及算法。提出在信噪比(sNR)较低的情况下，根据语音短时能量和过零率，判断在无声 或有声期间是否偶然的噪声过高，从而设定合适的参数降低噪声。-Introduced an improved spectral subtraction principles and algorithms. Proposed signal to noise ratio (sNR) lower case, according to voice short-term energy and zero-crossing rate, to judge whether silent or sound during the occasional noise is too high, and thus set the appropriate parameters to reduce noise.