A comparative study of BA, APSO, GSA, hybrid PSOGSA and SPSO in dual channel speech enhancement

The present paper focuses on the suppression of background noise in speech signal by utilizing a powerful heuristic optimization algorithm called Bat algorithm (BA) for adaptive filtering in dual channel enhancement systems. Bat algorithm is a recently developed population based meta heuristic approach which is inspired by the hunting behavior of the bats. It is developed by Yang (2010). As a novel feature Bat algorithm is based on the echo location behavior of micro bats. BA uses the frequency tuning technique to increase the diversity of the solutions in the population, while at the same time it uses the automatic zooming and tries to balance the exploration and exploitation. A few studies have been carried out on the use of heuristics for ANC in speech enhancement by using standard particle swarm optimization (SPSO) and some of its variants till 2010. In order to extend these heuristic approaches to speech, accelerated particle swarm optimization (APSO) based enhancement approach has been proposed (Prajna et al. in Int J Speech Technol 17(4):341–351, 2014a; Prajna et al. in IJISA 6(4):1–10, 2014b, doi:10.​5815/​ijisa.​2014.​04.​01; Prajna et al. in IEEE international conference on communications and signal processing (ICCSP), April 2014, pp 1457–1461, 2014c, doi:10.​1109/​ICCSP.​2014.​6950090). To overcome the problem of poor exploitation ability of APSO, another approach is proposed based on gravitational search algorithm (GSA) (Prajna et al. 2014a, b, c). To combine both the abilities of PSO and GSA algorithms, Hybrid PSOGSA algorithm is also proposed to ANC (Prajna et al. 2014a, b, c). To further improve the efficiency of adaptive filtering in ANC, by providing a dynamic balance between exploration and exploitation, BA is proposed to speech enhancement. This paper intends to present the Bat algorithm as an improved approach to ANC in speech enhancement when compared with that of SPSO, APSO, GSA and Hybrid PSOGSA based speech enhancement algorithms. The performance of all the algorithms is evaluated by computing four objective measures SNRI, PESQ, FAI and WSS, in two real world noise conditions Babble and Factory, at three different input SNR levels set at −10, 0 and 5 dB. Simulation results prove that BA is the most successful algorithm of all the algorithms studied in this work, to suppress the background noise more effectively.