Authors
Abstract
Speaker recognition and voice recognition are subfields of speech processing by computer. They work on the principle that there are features of speech that can be used to discriminate one speaker from another through three stages, preprocessing, feature extraction and classification.
In preprocessing stage average magnitude and zero crossing rate were used to detect start and endpoint of the speech. In feature extraction stage average pitch and 12-linear prediction coefficient were used to represent the important characteristics of the speech. In classification stage most methods used in patterns recognition perform some kind of comparison of a feature-vector with some reference-vector. No things like that is happening here, a new approach is presented based on a set of uniform cellular automata (CA). Computation in CA has been studied from different perspectives and has been constructed for various specific computation tasks as far as it is shown capable of universal computation. So the main object of this research is to discover the capability of the cellular automata of performing one-dimensional and two- dimensional pattern classification when these patterns are feature vectors of speech.
Genetic algorithm was used also with cellular automata as an evolutionary supervised learning algorithms for implementing this task. +
)
