Implementación de un módulo de reconocimiento de voz para niños mediante el procesamiento de señales aplicado en un caso práctico.

Abstract

A Child voice recognition was developed by signal processing applied to the control of 4wd robot car througth voice commands: right, left, forward, backward and halt. The program consists of two graphical interfaces developed in Matlab called database and voice recognition for children. In the database interface processes were performed corresponding to the acquisition of the voice using a Stereo Headset microphone, once acquired the speech signal processing was performed whose objective was to eliminate unwanted silences of the signal for which the method was used of energy calculation which encompasses normalization, threshold determination and segmentation processes. To this new signal, we used a pre-emphasis filter to accentuate the high frequencies and smooth the spectrum, from this filtered signal we extracted the voice characteristics through the Calculation of Linear Prediction Coefficients (LPC), to obtain the signal was performed an average between the coefficients of each recording. The second graphical interface performs the same processes described above for the speech signal to be recognized, then a comparison of distances between the speech signal to be recognized and each of the pattern signals is performed, the speech command having a minimum distance being recognized. According to the tests carried out with the Voice Recognition Module for children, it was determined that it has a total effectiveness of 84.70%.These results were compared with the existing Module in the SR-07 market used in the thesis "Design and implementation of the System of directional movement of a wheelchair to be controlled by recognition of a voice pattern by means of power electronics and DC motors as actuators ", thus concluding that it fulfilled each of the objectives proposed in this titration work due to that the implemented module provided better results, however it is recommended to test the current methods used in voice recognition in order to further optimize the results.