Speech Profile Recognition by Fourier Spectral, FFNN and ANFIS Techniques

Abstract

This paper presents a combined approach for recognition of speech profiles based on FFT windowing, Feed-Forward Neural Networks (FFNN) and Adaptive Neuro-Fuzzy Interface Systems (ANFIS). By using spectral analysis of the speech of physical entities, there have been carried out feature extraction during application of Hamming, 4 Term B-Harris, Flat Top and Hanning windows. Individual informative sets (data sets) have been specified for the employed mathematical recognition apparatuses. A FFNN model has been synthesized during implementation of Scaled Conjugate Gradient (SCG) training for the purpose of speech profiles recognition with attained accuracy of 93.50 %. There has been selected neuro-fuzzy classifier in accordance with Hybrid learning algorithm and Pi shaped membership function of input variables. During testing of selected ANFIS model there has been established a 100.00 % accuracy in speech profiles recognition.