Classification of Imagined and Heard Speech Using Amplitude Spectrum and Relative Phase of EEG
Ryota Sakai, Atsuhiko Kai, Seiichi Nakagawa
Abstract
As a brain-computer interface, speech recognition technology based on EEG (Electroencephalography) during speech imagining is desired, and research on this subject is underway. We extracted statistical features, amplitude spectrum, and relative phases from various analysis widths for EEG at the time of hearing and imagining, and performed an articulatory-based binary classification task. The combined use of amplitude spectrum and relative phase outperformed the chance rate. Our experimental results showed the effectiveness of phase information in EEG-based speech recognition.
Topics & Concepts
ElectroencephalographySpeech recognitionComputer scienceAmplitudeBinary numberBinary classificationBrain–computer interfaceTask (project management)Frequency spectrumPhase (matter)Pattern recognition (psychology)Relative phaseArtificial intelligenceSpectral densityMathematicsPsychologySupport vector machineEngineeringPhysicsTelecommunicationsSystems engineeringQuantum mechanicsPsychiatryArithmeticEEG and Brain-Computer InterfacesBlind Source Separation TechniquesNeural Networks and Applications