پروژه کارشناسی دانشگاه شهید بهشتی Imagined Speech Classification

عنوان پروژه: بررسي روشهاي مدلسازي مبتني برشبكه عصبي براي دسته بندي تصورات كلمات و واج ها از روي سيگنال مغزي EEG دانشجو: عرفان قبادیان استاد راهنما: دکتر یاسر شکفته استاد داور: دکتر آرمین سلیمی بدر دانشکده مهندسی و علوم کامپیوتر دانشگاه شهید بهشتی Speech is a complex mechanism, which involves multiple brain areas in the process of production, planning, and controlling multiple muscles related to the utterance to create phenomes, words, and finally sentences. Speaking is one of the most important ways humans use to communicate. Some people are not able to speak due to some sickness and disorders. To facilitate these people's communication, Brain-Computer Interfaces try to recreate words from brain activities so that these people can communicate with other people without having to speak. Recognition of words from brain signals can be done using artificial intelligence and machine learning. In this project, an intelligent system is proposed for recognizing 4words and 7phenomes. The system has been trained on Kara One dataset and feature extraction is done using cross-covariance in the time and frequency domain. We showed that using cross-covariance in the frequency domain for feature extraction has better results than not using cross-covariance or using signals in the time domain. In the classification section, we examined multiple CNN architectures and LSTM. The best result accuracy in this project is 43.34 for 11 classes.