Feature extraction and pattern identification of silent speech by using MFCC, DTW and AI algorithms
ARTIGO
Inglês
At present are many the ways of communication, that allow the interaction among people of the same society, a particular case of these communication ways appears the silent speech, but it is still in study and development. Silent speech, is to acquire the signals generated in the vocal apparatus...
At present are many the ways of communication, that allow the interaction among people of the same society, a particular case of these communication ways appears the silent speech, but it is still in study and development. Silent speech, is to acquire the signals generated in the vocal apparatus before a sound occur, in order to establish a channel of information transmission in environments with a considerable amount of noise or among people whose ability to emit sounds is limited due to various pathologies. In this work the results of capturing and analyzing signals of silent speech, with the aim of identifying phonological units of Spanish language are presented. Initially the signals acquisition was performed by NAM microphone, for further processing with feature extraction techniques like MFCC (Mel Frequency Cepstral Coefficients) and DTW (Dynamic Time Warping), which provided the necessary data for training the neural network for the pattern recognition task. The classification algorithm was trained with the data of 9 test subjects, all of the male gender, with 5 samples from each of the three phonological units that want to be recognized ('Uno','Dos','Tres'), as a final result the algorithm is able to classify and identify patterns with a success rate over 85%
Fechado
Feature extraction and pattern identification of silent speech by using MFCC, DTW and AI algorithms
Feature extraction and pattern identification of silent speech by using MFCC, DTW and AI algorithms
Fontes
Journal of engineering and applied sciences Vol. 11 (July, 2016), p. 8500-8510 |