Room Response Equalization of Non-Minimum Phase Systems Using Kautz Filter and Sparse Autoencoder: A Hybrid Approach

Abstract

Room Response Equalization (RRE) systems play a vital role in enhancing the hearing experience in different real-time application areas such as cinema theatres, home theatres, hearing aid implementation, car hi-fi systems, etc. Thenceforth, RRE using digital signal processing techniques has been a matter of great interest to researchers for a long time. RRE is intended to provide a listener with a profound audio experience alike the original audio signal. Conventional filtering approaches have been used by several researchers to perform room equalization. However, this article presents a novel technique by cascading a Kautz filter and a sparse autoencoder in performing room equalization. Kautz filters belong to a category of fixed pole IIR filters that use least-square (LS) minimization to generate orthonormal tap-output impulse responses. This LS approximated signal is further trained by a sparse autoencoder, which provides filter coefficients, iteratively optimized, to minimize a cost function. The cost function is considered here in terms of the reconstruction error function. The performance of this hybrid network of equalization is evaluated in terms of mean square error and spectral deviation measures. Computational results show that this hybrid approach yields better results both qualitatively and quantitatively in comparison with the other filtering techniques.