Decoding Imagined Speech using Wavelet Features and Deep Neural Networks

Abstract

This paper proposes a novel approach that uses deep neural networks for classifying imagined speech, significantly increasing the classification accuracy. The proposed approach employs only the EEG channels over specific areas of the brain for classification, and derives distinct feature vectors from each of those channels. This gives us more data to train a classifier, enabling us to use deep learning approaches. Wavelet and temporal domain features are extracted from each channel. The final class label of each test trial is obtained by applying a majority voting on the classification results of the individual channels considered in the trial. This approach is used for classifying all the 11 prompts in the KaraOne dataset of imagined speech. The proposed architecture and the approach of treating the data have resulted in an average classification accuracy of 57.15%, which is an improvement of around 35% over the stateof-the-art results.Index Terms-imagined speech, brain-computer interaction, deep neural network, commone spatial pattern, EEG

Silent EEG-Speech Recognition Using Convolutional and Recurrent Neural Network with 85% Accuracy of 9 Words Classification

“…Taking into account the presented works, the average recognition accuracy for the binary classification of words or syllables approached 72% with a maximum accuracy of 85.57% [7][8][9]11,14,15,21,23,24]. For four words, according to the papers presented, the classification accuracy averaged 59%, while the maximum accuracy reached 99.38% [13,15,23,25].…”

“…In 2019, scientists from India published a method for recognizing seven syllables, /iy/, /piy/, /tiy/, /diy/, /uw/, /m/, and /n/, and four words, pat, pot, knew, and gnaw, on data from 11 EEG channels located in areas of the cerebral cortex involved in speech presentation [14]. Due to the high correlation between different channels, each channel was used as a separate object for the neural network training.…”

Decoding Covert Speech From EEG-A Comprehensive Review

“…Similar to García et al ( 2012 ), EEG channels are manually chosen in Panachakel et al ( 2019 ). Specifically, the following 11 EEG channels are chosen based on the significance of the cortical region they cover in language processing (Marslen-Wilson and Tyler, 2007 ; Alderson-Day et al, 2015 ):…”

“…In another work by Panachakel et al ( 2019 ), a combination of time and wavelet domain features was employed. Corresponding to each trial, EEG signal of 3-s duration was decomposed into 7 levels using db4 wavelet and five statistical features, namely, root mean square, variance, kurtosis, skewness, and fifth order moment were extracted from the last three detail coefficients and from the last approximation coefficient.…”

Decoding Imagined Speech From EEG Using Transfer Learning

“…In most of the works in the literature such as [3]- [6], [8]- [18], [20]- [22], individual EEG channels are considered separately for extracting features. In these works except [22], [26], wavelet domain features, mel-frequency cepstral coefficients (MFCCs) and/or temporal domain features are extracted from each channel and are concatenated to obtain the feature vector for each trial. In [22], [26], the features extracted from individual channels are considered as distinct data vector and the decisions of the classifier for each channel are combined to obtain the final classification result.…”