The study of speech imagery, where individuals generate internal mental representations of speech production without physical movements, has been widely investigated due to its potential application in more efficient assistive technologies. The aim of this study is to classify features among five distinct classes during speech imagery, comparing brain signals in the scalp and around the ear for applications in Brain-Machine Interfaces related to communication. Speech imagery signals were collected from four healthy volunteers with no speech or cognitive impairments. Electrodes were placed on the scalp and around the ear. Five words were utilized for speech imagery tasks and collect electroencephalogram (EEG) and ear-EEG data, and conducted the frequency domain analysis using the Fast Fourier Transform. The mean power in distinct frequency bands was calculated for feature extraction, to after apply the Linear Discriminant Analysis and Principal Component Analysis. Our findings suggest that speech imagery features are a challenge for obtaining an accurate contact-less brain-computer interfaces for communication purpose. It is emphasized the need to further explore speech imagery, as well as to identify and characterize regions of better contribution for discrimination.