imagined-speech-classification

Imagined Speech based on KaraOne Database

Introduction

Karaone is a database of imagined speech. It contains 11 classes (4 words and 7 phonemic prompts), each of which is imagined by 14 different subject. Read more about the database here.

Feature Extraction

The features extracted in different ways like Cross Covariance Matrix of channels in both time and frequency domain.

Classification

The classification is done using different methods like CNN, CNN+LSTM, EEGNet.

Results

The results are shown in the table below:

Model	Accuracy
CNN+Cross Covariance in Time domain+0.25s window	9.61
CNN+Time Signal with 0.25s window	15.66
CNN+Cross Covariance in Frequency domain+0.25s window	43.34
CNN+LSTM+Cross Covariance in Frequency domain+0.25s window	33.24
EEGNet+0.25 window	16.73

Contributors

• Erfan Ghobadian
• Sarah Saremi