A system was designed to recognize English vowels from electromyography (EMG) signals produced during vowel speech. Electrodes were positioned to detect articulations of the jaw, lips, and tongue, as described by the International Phonetic Association. Features were generated by lowpass filtering recorded EMG and calculating root mean square. A non-linear support vector machine was used to classify feature vectors into vowels. After comparison with alternative methods, the proposed system was found to have the highest accuracy, at 95.50percent.
展开▼