Silent Speech Recognition: Algorithm Research

Silent communication BCI (brain computer interface) is used to extract silent speechsignals and analyze their characteristics accurately which can decode language cortexindirectly. Related technical research will bring hope for special populations such asmuscular dystrophy or stroke patients who can’t speak like healthy people. Starting fromthe research paradigm of signal recognition of human brain, NIRS-BCI (near infraredspectroscope-brain computer interface) is applied to collect signals at Forehead-point onauricle while silent thinking Chinese words based on rule of neurovascular coupling,mechanism of brain era reflection and principle of near infrared scattering imaging.Nonlinear characteristics of signals are analyzed from the angles of chaos andinstantaneous energy once more, chaotic algorithm and spike potential detectionalgorithm are used to decode silent thinking words.The conclusions of our research are shown below:The first, silent speech signals are chaotic. In experiments, attractor, Lyapunovexponent and fractal dimension algorithms show chaotic characteristics of signals. Theirmaximum Lyapunov exponents are all greater than zero. Every phoneme has its chaoticattractor and their attractors are all irregular curves. Silent thinking words have fractalcharacteristics, their box counting dimensions are between1.5and2, about1.66. Thewomen’s are greater than the man’s.The second, nonlinear energy operator is used to extract diverse and detailed actionfeatures of different words. The main characteristics of spike potential of the same wordsare similar, different words have various spike potential features. The recognition rate ofthe same single word can reach60%. Spike potential figures of double words arecomplex and their recognition rates are low, but their phonemes features consistent withtheir single word. So, signals extracted at Forehead-point on auricle based on NIRS-Based BCI andanalyzed by effectively nonlinear algorithms, different characteristics can be recognizedwhile silent thinking Chinese words.