Mel Frequency Cepstral Coefficients Enhance Imagined Speech Decoding Accuracy from EEG

Abstract

Imagined speech has recently become an important neuro-paradigm in the field of brain-computer interface (BCI) research. Electroencephalogram (EEG) recordings during imagined speech production are difficult to decode accurately, due to factors such as weak neural correlates and spatial specificity, and signal noise during the recording process. In this study, a dataset of imagined speech recordings obtained during production of eleven different units of imagined speech is used to investigate the relative effects of different features on classification accuracy. Three distinct feature-sets are computed from the data: a linear feature-set, a non-linear feature-set, and a feature-set comprised only of mel frequency cepstral coefficients (MFCC). Each featureset is used to train a decision tree classifier and a Support Vector Machine classifier. The results indicate that the use of MFCC features provides greater discrimination of imagined speech EEG recordings in comparison with the other features evaluated, and that phonological differences between imagined words can serve as an aid to classification.

Evaluation of Hyperparameter Optimization in Machine and Deep Learning Methods for Decoding Imagined Speech EEG

“…Fourth-order Daubechies were used to achieve the results presented in [56] and were therefore used here. A feature vector was constructed using the RWE of decomposition levels D2 (16)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30)(31)(32), D3 (8-16 Hz), D4 (4-8 Hz), D5 (2-4 Hz), and A5 (<2 Hz), for each channel. This resulted in a 30-element feature vector for each trial.…”

“…Here, six frequency bands are used to construct the filter bank. These are delta (2-4 Hz), theta (4-8 Hz), mu (8-12 Hz), lower beta (12-18 Hz), upper beta (18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28), and gamma (28)(29)(30)(31)(32)(33)(34)(35)(36)(37)(38)(39)(40). Three HPs were selected for optimization, i.e., (1).…”

“…Most imagined speech EEG studies have used traditional BCI approaches to feature extraction and classification. Common spatial patterns (CSP) [25], autoregressive coefficients [26,27], spectro-temporal features [23], Riemannian manifold features [17] and Mel Frequency Cepstral Coefficients (MFCC) [18,28] are among those used to represent imagined speech EEG. As with feature extraction, imagined speech decoding has relied on several classification approaches typical to BCIs.…”

“…As with feature extraction, imagined speech decoding has relied on several classification approaches typical to BCIs. The Support Vector Machine (SVM) has been the most common approach to date [18,20,23,25,[28][29][30], showing results in the range of 77.5%-100% in binary vowel classification tasks [29] and 92.46% average accuracy in a two-class, meaning-based imagined speech task [20]. Linear Discriminant Analysis (LDA) has also been used for DS-BCI decoding [27,31], with one study reporting accuracies in the range of 66.4-76.0% in two-class imagined phoneme production tasks [31].…”

Decoding Covert Speech From EEG-A Comprehensive Review

“…Most works do not employ any spatial filtering in the preprocessing. The only exceptions are the works by Zhao and Rudzicz ( 2015 ) and Cooney et al ( 2018 ), who used a narrow Laplacian filter. A Laplacian filter uses finite difference to approximate the second derivative.…”