The sound signals generated by the machine tool punching contain a lot of useful information. In order to filter out the noise generated by the factory environment from the extracted sound signals，the improved threshold wavelet denoising method was used to process the sound signal. SNR and MSE were used as denoising indicators, a piecewise continuous threshold function was proposed to improve the traditional wavelet denoising method. The improved function solves the problem of discontinuity and constant deviation of soft and hard threshold function in traditional threshold function by introducing the constant a.By using the collected sound signal of machine tool punching superimposed with Gaussian white noise as the noise signal of machine tool punching,the improved threshold wavelet denoising method was verified.The optimal wavelet function type and decomposition layer number were determined through experiments, and the improved threshold wavelet denoising algorithm was compared with the traditional algorithm. The results show that the improved algorithm has a good denoising effect and can effectively retain the lower signal energy.