Adaptive Noise Cancelation in Speech Signals

Today, adaptive algorithms represent one of the most frequently used computational tools for the processing of digital speech signals. This work investigates and analyzes the properties of adaptive algorithms in speech communication applications where rigorous conditions apply, such as noise and echo cancelation. Like other theses in this field do, it tries to tackle the ever-lasting problem of computational complexity vs. rate of convergence. It introduces some new adaptive methods that stem from the existing algorithms as well as a novel concept which has been entitled Optimal Step-Size (OSS). In the first part of the thesis we investigate some well-known, widely used adaptive techniques such as the Normalized Least Mean Squares (NLMS) and the Recursive Least Mean Squares (RLS). In spite of the fact that the NLMS and the RLS belong to the “simplest” principles, as far as complexity is concerned, they are widely adopted across the spectrum of telecommunication products. It appears, however, that with the development of voice-operated devices the future systems would require a more powerful noise and echo suppression than could be achieved with the conventional meth ods. In the second part of the thesis we will show that, by modifying these methods better results can be achieved. In particular, we will introduce the Fast Block Least Mean Squares (FBLMS) method and the Simultaneous Perturbation Stochastic Approximation (SPSA) method as examples of methods that are based on a conventional LMS algorithm. However, due to some sophisticated modifications they yield better results. For the purposes of comparison of the developed methods with the existing ones we conduct several experiments in three different applications. These are system identification, background noise suppression and inverse filtering. In the last part of the thesis we describe a novel approach to adaptive filtering that we developed and which we named OSS. The proposed method possesses some features that make it robust even in a non-stationary environment. Experimental evaluations will show that its performance is comparable with the conventional methods and in some cases it will even exhibit an improvement. At the end of the thesis we discuss some issues pertaining to the testing of echo cancelers. In particular we describe our software solution that we developed according to the ITU-T G.168 recommendation. We have designed a stand-alone application in Matlab Graphical User Interface Development Environment (GUIDE) that can be used to test the performance of AECs (Adaptive Echo Cancelers). By carrying out a graphical analysis the users are able to reveal the weak parts of their methods and estimate whether they meet the G.168 conditions or not.