Abstract: Arc sound is well known as one of the potential source signals for controlling of welding process quality, which is vulnerable to noise pollution by equipment and environment in sensing, conversing and transmission.Therefore, it is necessary to reduce arc sound noise by hardware and software.Hardware measures include low-noise microphone, grounding devices, eliminating the DC offset exchange coupling, equipment amplifiers, anti-alias filtering and shielding cables, etc.In terms of software, by using Mallat algorithm based on principle of multi-resolution, the noise of arc sound signal is reduced via wavelet decomposition and reconstruction based on LabVIEW.The results show that the burr of arc sound is effectively filtered out, the mutations of waveform are clearer and the noise ratio is obviously increased.The reducing noise means provide technical basis for extracting characteristics of arc signal and monitoring welding quality via arc signal.