Speech Signal Denoising Based on <scp>GWO</scp> ‐ <scp>VMD</scp> Combined With <scp>TQWT</scp>

ABSTRACT In speech signal processing, noise interference during transmission or recording remains a persistent challenge. To address this, we propose a hybrid denoising method combining Grey Wolf Optimizer‐Variational Mode Decomposition (GWO‐VMD) with Tunable‐Q Wavelet Transform (TQWT). First, the GWO algorithm is utilized to optimize the VMD decomposition parameter combination, specifically the number of modes and the penalty factor. Based on the correlation coefficient screening principle, the Intrinsic Mode Function (IMF) components with relatively low correlation coefficients are then removed. Next, the remaining IMFs are recombined, and multi‐scale analysis is conducted through TQWT. Finally, the sub‐band components whose relative energy exceeds a certain threshold are reconstructed to obtain the denoised signal. For the original three‐segment noisy speech signals, the signal‐to‐noise ratio (SNR) has been remarkably enhanced by using the GWO‐VMD‐TQWT denoising method. It has increased from −1.2227, 8.4398, −4.6318 to 11.4980, 19.2606, 6.1067 dB, respectively. Additionally, the root mean square error of this algorithm after denoising is the smallest among the compared methods. Evidently, the joint denoising method in this study outperforms single VMD, TQWT, and classical denoising techniques like EMD, EWT, and CEEMDAN. It provides an effective solution for low SNR speech signal denoising.