Adaptive spectral subtraction to improve quality of speech in mobile communication

Abstract

With the development of VLSI technology, mobile communication is supported by smart devices for transmission and reception of information in various forms. In order to improve the quality of speech communication smart devices are provided with pre-processing algorithms. These algorithms should be adaptive in nature to suppress the real-time noise. The performance of smart devices depends on the algorithms. The performance of smart devices is excellent in noise-free surroundings; however, their performances worsen in noisy surroundings. Spectral domain weighting approaches, which estimate spectral density of noise, are considered for speech enhancement. These algorithms process speech in short frames. In this paper, we propose one such novel algorithm for assessment of noise in very small bands based on type of disturbance. Thus, using multiband time-varying filtering coefficients, speech signals are modelled by autoregressive process. Experimental results demonstrate an improvement of 25% to 45% as compared to other conventional multiband approach.