Investigation on environmental noise classification and cancellation using optimization techniques for speech quality improvement

Abstract

The purpose of a communication system is to transmit and receive information truly and safely The information through the medium of transmission is likely to be distorted and corrupted due to background noise Background noise is an important factor for devices used in mobile environments such as mobile phones and hand free telephones in cars The quality and intelligibility of the information signal in the presence of background noise can be improved by speech enhancement techniques The performance of hearing aids can also be improved by speech enhancement techniques Current speech enhancement methods are classified into two categories namely time domain methods such as the subspace approaches and frequency domain methods such as the spectral subtraction Minimum Mean Square Error estimator and adaptive filters Adaptive filter has been widely used in communications control and many other systems It is based newlineon the observation of the existing signal and adjusts the parameters automatically and changes the performance So its design does not require any of the prior knowledge of signal or noise characteristics Adaptive systems using interference cancellation techniques are used very efficiently to reduce noise from the signals But the interference newlinecancellation technique is efficient only when there is a scope of obtaining noise correlated to the information signal where it is used as a reference signal But when the signal is transmitted through the channel the noise that gets added in the channel is totally random Hence there is no means of creating a correlated noise at the receiving end The only possible way is to somehow extract the noise from the received signal itself Practical and efficient noise cancellation system is wholly reliant on an accurate analysis of sources of noise The main problem in most of the environmental noise cancellation systems is the source of noise signal which is to be used as a reference signal Once the noise source is known then the noise elimination process will become easier In any of the noise cancellation system noise classification is an important task before noise cancellation In this thesis different noise classification methods are proposed using fuzzy inference system and neural networks such as Fuzzy ARTMAP Neural Network newline Radial Basis Function Network and Extreme Learning Machine In addition noise cancellation methods are also proposed newline newline