Performance analysis of adaptive variational mode decomposition approach for speech enhancement

Speech enhancement is an important pre-processing task in the area of speech processing research. Many techniques have been applied in this area since four/five decades. With progressive research it occupies a special position in various fields like engineering, medicine, society and security. Adaptive algorithms found effective for such cases and are utilized in this problem. The work is based on decomposition method using variational mode decomposition (VMD) technique, where the decomposed components signify the frequency characteristics of the signal. Since Wiener filtering is used in VMD inherently, it is modified with the least mean squares (LMS) adaptive algorithm for good accuracy and adaptability in this work. Different noises like Babble noise, Street noise, and Exhibition noise are considered and the corresponding signals are decomposed into five intrinsic mode functions (IMFs). Basically, the lower modes are of high frequency and noisy; whereas the higher mode IMFs contain the low and medium frequency components and are considered as the enhanced signal. The results of the proposed algorithm are found excellent as compared to earlier techniques. The resultant wave forms are visually observed and the sound is verified for audible range. Also different measuring parameters are considered for its performance measure. It is measured in terms of signal-to-noise ratio (SNR), segmental signal to noise ratio (SegSNR), perceptual evaluation of speech quality (PESQ) and log spectral distance (LSD). The technique is verified with standard database NOIZEUS for 0, 5, 10, 15 dB respectively and also in real world case.