MANC ANet: Multitude Active Noise Cancellation Using Improved Optimized CNN-ALSTM Network

Active Noise Cancellation (ANC) is a crucial component of audio equipment that reduces undesirable background noise. The performance of linear systems is adversely affected by nonlinear distortions. Over a decade, numerous academics created various algorithms to enhance the quality of speech signals and reduce noise. In this research, Multitude Active Noise Cancellation method based Improved White Shark Optimized CNN-ALSTM Network (MANC ANet) is proposed. A Dual Tree Complex Wavelet Transform enhances the quality of audio and multiple noise signals by extracting statistical, spectral and cepstral features. Genetic refinement algorithms select the most desirable features based on community consensus. A hybrid convolutional neural network that uses attention-based long short-term memory (CNN-ALSTM) is applied to classify interference noise signals and desired signals. Also, the Improved White Shark Optimization method is used to increase network accuracy by optimizing hyperparameters. Performance metrics such as accuracy, specificity, sensitivity, NMSE, STOI and PESQ measure the effectiveness of proposed methods. Overall, the proposed approach outperforms GFANC-Bayes, MCANC, GFANC, MCDM-NA and Unsupervised GFANC by 11.15%, 9.44%, 6.48%, 5.15% and 2.3%, respectively.