Methodology for voice commands recognition using stochastic classifiers

The incidence of people with motor disabilities in Colombia is around 6.4%, which is a major social problem, because people with such disabilities lose their autonomy to perform basic actions such as displacement. Therefore, we propose a solution to the problem of mobility in people with motor disabilities, allowing to take control of the engines, with a voice comand recognition system. This paper presents a methodology for recognition of isolated spanish words (silla, atrás, adelante, derecha, izquierda, pare). To this end, we use a methodology based on the wavelet transform preprocessing. The characterization of the filtered signal is performed by Mel Cepstral Coefficients and classification stage using hidden Markov models. The methodology has proven to be robust, because the databases used for training the system have been acquired in noisy environments as well as controlled, presenting performances in classification acuracy of 98%.