Persian words recognition using facial electromyogram signals

Losing of voice and larynx is a major problem for people with speech disorders. It creates serious and negative consequences on the quality of individual and group life of these people, especially in working environments. The development of an intelligent system based on electromyogram signals with the ability to recognize speech (without using sound) can be a window of hope for people who lost their larynx and voice due to cancer. Although progress and studies in this field are growing in our country and in different languages, but these studies have not been done for the Persian language. In this article, for the first time, recognition of Persian words was done using electromyogram of facial muscles. For this purpose, sEMG signals were collected from eight facial muscles and six volunteers while speaking twelve Persian words. Then, MFL, VAR, DAMV, LTKE, IQR and Cardinality features were extracted from each channel and each window from the signal, and the 432 features from each signal were reduced to 33 features using the PCA principal component analysis method. Finally, in order to recognize twelve Persian words, the features were given to SVM, KNN and RF classifiers. The average classification accuracy was 83.16%, 81.91% and 78.97%, respectively. Our evaluation in this article gives the hope that by using EMG signals it is possible to recognize the limited words of Persian language.