Research on Speech Denoising Method in Welding Environment Based on Enhanced Wavelet Thresholding Function
In welding environments,strong noise not only interferes with the quality of speech communication but also generates electromagnetic interference and multipath sound echoes,severely affecting the clarity and intelligibility of speech.Due to its ability to perform localized time-frequency analysis,wavelet transform has been widely used for denoising tasks involving non-stationary and non-steady-state signals.This paper focuses on the issue of speech denoising in welding noise environments and investigates the effects of different wavelet bases,decomposition levels,and thresholding functions.It is found that traditional soft and hard threshol-ding functions do not yield satisfactory denoising results.To address this problem,we propose an adaptive wavelet thresholding func-tion and conduct denoising experiments under three conditions:sine signals with added Gaussian white noise,English speech with added welding noise,and Chinese speech with added welding noise.The results demonstrate that compared to traditional fixed thresh-olding functions,the proposed adaptive thresholding function significantly improves the output signal-to-noise ratio by 2-4 dB and reduces the mean square error of the signal by 10-35％,preserving more speech details.This validates the advantage of the adaptive thresholding function in suppressing welding noise while preserving the speech components.
万方数据
维普
