Speech recognition system based multi-level selective filter

In this paper, an updated source-filter based method is used for speech recognition of English speakers. The key goal of this paper is to discuss the influence of using dynamic and selective Filterbank on the efficiency of the system in terms of identifying the word being said. Dynamic structures of the bandpass filter bank are clearly able to adjust their principles in conjunction with the speech database used. In this paper, the proposed methods adopted the idea of re-select or adjusting the filter parameters values at each iteration, in accordance with the word being processed. In other words, the values of the filter components (center frequencies and bandwidths) will adjust until it gets the best set of the filter components. The genetic algorithm (GA) will be used to maintain the components of the filter and each new population of the GA represents a new set of channels for the filter bank. Results show that the efficiency of the recognition system would be improved with each new set of filter components generated by the genetic algorithm. Speech recognition, Source-filter, Filterbank, genetic algorithm, adjusted filter, Dynamic system