New parameters for automatic speech recognition based on the mammalian cochlea model using resonance analysis

Undoubtedly the compact representation by a set of Mel Frequency Cepstrum Coefficients (MFCC) has been used satisfactorily for ASR [9]. The cochlea is an organ, in humans or mammalians that converts the frequency perceived by the ear in punctual stimulation to excite the nerve auditory that receives a set of stimulus that comes from speech sound pressure. A new approach is proposed that considers this phenomenon to construct the bank filter in our parametric representation. Then we substitute the distribution of the bank filter in the Mel scale function for a different distribution that depends of the inner ear response to the stimulus that it receives. The place theory is used which achieves a 99.8% performance. Finally, this paper compares the performance of different acoustic representations in Continuous Automatic Speech Recognition system (CASRs) based on words. The cochlea operation is explained that permits obtaining a model and we will show that one alternative solution to the model based on fluid mechanical proposed by Lesser and Berkley [19], can be obtained if resonance analysis weather Fourier series is used as a solution.