Advanced CNN-based Model for Imagined Speech Classification Using Electroencephalogram

This paper aims to propose a robust electro- encephalogram (EEG) based Imagined vowel classification system using GoogleNet. For people who are unable to speak due to motor impairments or other neurological issues, imagined speech is a new form of communication. In other words, it can be defined as an individual’s spontaneous thinking without any articulatory movements. Imagined speech can be obtained by measuring brain activity in a non-invasive procedure through EEG. This study classified imagined speech vowels using an Advanced CNN-based model, and the results were compared to Google Net model performance. This study used the variational Mode Decomposition (VMD) technique to denoise the EEG signal. The "Correto DB," an open-access database containing both imagined and pronounced speech as well as the corresponding 6-channel EEG data, was used in this study.