AFM Signal Model for Digit Recognition

In this paper, a multicomponent amplitude and frequency modulated (AFM) signal model is proposed for isolated word (digit) recognition. The AFM signal model is designed at phoneme level, and features are extracted from each phoneme of the spoken word. The Fourier-Bessel expansion and discrete energy separation algorithm (FB-DESA) are used for the analysis of the signal model. The AM and FM parameters of the model are the features considered for digit recognition. The selected features are applied to the classifiers for word recognition. We achieved the best classification accuracy using the random forest decision tree classifier for speaker-independent data, and the best accuracy using the artificial neural network classifier for speaker-dependent data.