Neuroscientists attempting to create effective brain-computer interfaces have difficulty turning the wide variety of brain impulses into flawlessly conveyed words and images. Gradually, the science fiction notion of controlling machines or interacting with people using only the mind is becoming a reality. Despite the significant progress achieved with deep learning algorithms, converting brainwaves to words has been a major challenge for scientists. Deep learning algorithms create a neural network for the brain and a neural network-based mapping that translates all brain inputs into actions. The work focuses on the imagined speech recognition of electroencephalography (EEG) brain signals by expanding the brain-computer interface to include persons who struggle with speech and communication. Decoding imagined speech from nonlinear and nonstationary EEG brain signals is challenging. Research in imagined speech has indicated that decoding performance and precision must be enhanced. Developing deep learning technology increases the likelihood that imagined speech may be deciphered from EEG data with improved performance. To retrieve information from EEG signals, we suggested an unsupervised deep learning model Deep Belief Networks (DBN), for determining the subject's vowel from their cognition. Using DBN, the overall classification accuracy was 80 per cent. The investigation used an open-access dataset containing fifty individuals to imagine vowel voice sounds. Multichannel EEG raw data from multiple participants were preprocessed using bandpass filters, Independent Component Analysis and Median Absolute Deviation, and then the features were extracted using discrete wavelet processing. The model was trained and validated to improve the accuracy of imagined speech recognition. Index Terms: Speech imagery Deep Belief Networks (DBN) Electroencephalography (EEG) Vowel recognition