Time-varying parametric modeling of ECoG for syllable decoding Vasileios G. Kanas 1 , Iosif Mporas 1,2 , Griffin W Milsap 3 , Kyriakos N Sgarbas 1 , Nathan E Crone 4 and Anastasios Bezerianos 5,6 1 Dept. of Electrical and Computer Engineering, University of Patras, Patras, Greece 2 Computer Informatics Engineering Dept., TEI of Western Greece, Patras, Greece 3 Dept. of Biomedical Engineering, Johns Hopkins University, Baltimore, MD 21205 USA 4 Dept. of Neurology, Johns Hopkins University, Baltimore, MD 21205 USA 5 Dept of Medical Physics, School of Medicine, University of Patras, Patras, Greece 6 Singapore Institute for Neurotechnology, National University of Singapore, Singapore Email: vaskanas@upatras.gr (V.G. Kanas) Abstract. As a step toward developing neuroprostheses, the purpose of this study is to explore syllable decoding in a subject with implanted electrocortico- graphic (ECoG) recordings. For this study, we use ECoG signals recorded while a subject volunteered to perform a task in which the patient has been visually cued to speak isolated consonant-vowel syllables varying in their articulatory features. We propose a recursive estimation method to calculate the parametric model coefficients in each time instant and band power features from individual ECoG sites are extracted to decode the articulated syllables. Our findings may contribute to the development of brain machine interface (BMI) systems for syl- lable-level speech rehabilitation in handicapped individuals. Keywords: electrocorticography, time-varying autoregressive model, speech rehabilitation, brain machine interface 1 Introduction Speech decoding directly or indirectly from neural activity has been extensively studied for several years. The motivation of such approaches is, on the one hand, to examine the underlying functionality of the human brain during speech articulation and, on the other hand, to restore speech capabilities in severely handicapped individ- uals, such those suffering from disorders of consciousness [1]. Currently, electroence- phalography (EEG), intracranial electrocorticography (ECoG), and intracortical mi- croelectrode recordings have been utilized as neurophysiological recording techniques speech restoration. Due to its non-invasive nature, EEG has been widely used in human studies to record brain activity for indirect control of spelling devices. In particular, approaches based on letter or word selection have been proposed utilizing specific brain wave- forms, such as slow cortical potentials [2-4], the P300 event-related potential (ERP)