Comparative Analysis of the RLS Algorithm with Kronecker Product Decomposition for Acoustic Impulse Response Identification

With the proliferation of multimedia content, audio processing has become an important research area in recent years. One significant problem is the interference of various noise signals with speech signals, which hinders effective communication in many applications. Audio information is an integral part of audiovisual files, and its quality greatly affects the viewer's experience. Background noise or interference in the audio can be distracting, and it may even mask important dialogue. Non-professional video recordings may inadvertently capture ambient music along with speech signals, and the music may be recorded at a volume that masks the speech. Adaptive algorithms can be used to enhance speech signals in such recordings. These algorithms are designed to adaptively estimate and cancel the music signal from the mixed audio signal while retaining the speech signal, needing a version of the masking musical piece. Usually, a studio-quality copy of the recorded music is available. The problem can be modeled as a system identification problem. This paper aims to evaluate the performance of the recently proposed Recursive Least-Squares algorithm with Kronecker Product Decomposition (RLS-NKP) in estimating the acoustic impulse response (IR) of six different rooms. The algorithm is compared to other commonly used algorithms such as Normalized Least-Mean-Square (NLMS) and traditional RLS. The goal of this study is to determine the effectiveness of the RLS-NKP algorithm in enhancing speech signal recovery by reducing background noise (e.g., music). The results of the study provide important insights into the application of adaptive algorithms in audio processing for improved speech signal quality.