Evolutionary Techniques for Speech Enhancement

Genetic Algorithms have become increasingly appreciated as an easy-to-use general method for a wide range of optimization problems. Their principle consists of maintaining and manipulating a population of solutions and implementing a ‘survival of the fittest’ strategy in their search for better solutions. In this chapter, GAs are combined with a signal subspace decomposition technique to enhance speech that is severely degraded by noise. To evaluate the effectiveness of this hybrid approach, a set of continuous speech recognition experiments is carried out by using the NTIMIT telephone speech database.