Adaptive Independent Component Analysis for Blind Source Separation in Multi-Speaker Acoustic Environments with Enhanced Noise Resilience

BSS stands for blind source separation is an important method to unmix the resulted speech signal when multiple speakers are involved in the audio signals. This research deals with enhancing the use of Independent Component Analysis (ICA) with particular emphasis on issues of preprocessing and analysis of simultaneous speech occurrence in complicated environments. The NOISEX-92 database was used which is a multi-speaker recording with noise and then ICA algorithms are applied using the programming language Python with Scikit-learn tools. Using an ICA approach, the mixed signals were reversed into their original and independent parts, which should correspond to different speakers. The performance of the separated speech signals was analyzed based on measures such as signal to noise ratio (SNR), and perceptual evaluation of speech quality (PESQ) to effectively demonstrate the success of the method for separating speech sources. The implications of the findings are that ICA should improve source separation in real-world environments and future research into speech enhancement, teleconferencing barren sophisticated automatic speech recognition systems.