Simulation of microphone inaccuracies and robustness analysis of Beamformers inside a reverberant environment

Abstract: All Speech Processing Equipments like noise cancelling headphones and Hearing aids should be able to filter different kinds of interfering signals and present a clear sound to the listener. The effect of these interfering signals varies depending on the surroundings. Confining the area of interest to a closed room, ranging from a class room to an auditorium, the most significant interference that is dominant in the received signal is the “Reverberation Effect”. Reverberation is the combined effect of multiple echoes from the walls of the room and its intensity depends on the size, shape, dimensions and materials used in the construction of the room. This phenomenon is tolerable for some purposes like music recordings but it predominantly effects the speech. Eliminating this effect without affecting the original speech is a challenge and a topic of research in these days. This paper is an excerpt of the collective work done on reducing microphone inaccuracies using various adaptive beamformers and emphasizes on-the Elko’s Beamformer, the Weiner Beamformer, Max-SNIR Beamformer and the delay-and-sum beamformer. These beamformers are implemented with the fractional delay capability and then tested under reverberant conditions. This paper mainly concentrates on the effect of Reverberation on the outcome of a beamformer. A reverberant room environment is simulated and the effect of reverberation on an Elko’s beamformer output is compared with that of a Weiner beamformer. The size of the room is proportionate to the complexity of the implementation, i.e., as the size increases, so does the execution time of the algorithm. All these kinds of factors have been implemented and tested. Improvement in the Signal-to-Noise Ratio and PESQ scores of these beamformers with reverberant room and without reverberant room are noted. Results show that wiener beamformer has better average SNR and PESQ improvements both in reverberant and non-reverberant conditions with more microphones and is suitable for speech enhancement systems. With 2 microphones, wiener beamformer has about 10 dB average SNR-I and an average PESQ-I of 1 in anechoic case and about 6 dB average SNR-I and average PESQ-I of 0.4 for the reverberant case. However, Elko’s beamformer has the typical advantage of better improvement in the back plane (180o to 270o) due to its inherent properties. It has about 12 dB average SNR-I and an average PESQ-I of 1.5 in anechoic case and about 5 dB average SNR-I and average PESQ-I of 0.2 for the reverberant case. Hence, Elko’s beamformer is particularly suitable for Hearing aid implants.