Imagined Speech Classification using EEG based Brain-Computer Interface

EEG-Based Silent Speech Interface and its Challenges: A Survey

“…They reported that covariance features yielded the highest accuracy of 87.90% on binary classification. Pawar et al [44] also used MaxLCor combined with DWT features to classify imagined speech words [47] and achieved an accuracy of 40.64±2.45% (chance level 20%). Furthermore, Chengaiyan et al [45] identified vowels and consonants by applying brain connectivity features on each frequency band; coherence [67] as functional connectivity and partial directed coherence (PDC) [68], direct transfer function (DTF) [69], and transfer entropy [70] as effective connectivity.…”

“…0.5-70 [42] Audio "ambulance", "clock", "hello", "yes", "light", "help me", "pain", "stop", "thank you", "toilet", "TV", "water". 0.5-40 CSP LDA CV [43] Audio "go", "back", "left", "right", and "stop" 0.5-60 covariance and MaxLCor ELM CV [44] Audio "hello", "help me", "stop", "yes", "thank you" 0.5-128 DWT, MaxLCor SVM CV [45] Visual ten words for every vowel: "a" ("can", …, "tap"), "e" ("bed", …, "vex"), etc. This paper did not compare the accuracies quantitatively between studies due to the different techniques, datasets, or computation environments.…”