Multi-Microphone Speech Dereverberation and Denoising using Inverse Sparse approximation based GSC

Speech signals collected by receivers in speech communication applications contain reverberation and additive sounds in addition to the intended speech signals. This audio signal collected by a nearby microphone is just a collection of delayed and fading copies of anechoic speech, and differentiating direct-path speech from reverberated speech may be difficult, particularly when room reverberation and background noise are present. Motivated by the historical rise of sparse estimation in a number of signal processing applications, we explore its applicability to improving the speech signal in a noisy environment. This research focuses on optimizing the front-end of the Sparse-based Generalized Sidelobe Canceller (GSC) in order to achieve an improved speech signal for the corresponding reverb and noisy signal. The enhancement includes multi-channel speech signal processing using a generalized side-lobe canceller with inverse sparse estimation followed by a single-channel Wiener post filter. Speech samples from the IEEE corpus dataset are used to validate the proposed procedure. The obtained results have been compared to those well-known state of the art approaches such as MCWF?WPE, WPE-MVDR, ISC-LP and BS-WPE under the same room conditions. The PESQ achieved with the proposed algorithm is up to 3.4, whereas other methods have a PESQ value of between 2.4 and 3.1. The LSD for the proposed algorithm is decreased to 1.11 and others have 1.4 to 1.9, which is better than the reviewed methods. According to the results, the proposed methodology significantly outperformed the competing approaches in all instances.