Reliability and variability in facial electromyography for identification of speech and for human computer control: an experimental study

The need for developing reliable and flexible human computer interface is increased and applications of HCI have been in each and every field. Human factors play important role in these kinds of interfaces. This research investigates the use of facial muscle activity for a reliable interface to identify voiceless speech-based commands without any audio signals. We propose a method of measuring the relative activity of the articulatory muscles of the face for recognition of unvoiced vowels. System performance and reliability were also tested for the case of variations like inter-subject, inter-day, and different languages. In these investigations, English vowels and German vowels were used as recognition variables. The designed methodology used linear and non-linear classification based on statistical clustering techniques and artificial neural network architecture. The results show that there is a variability in facial muscle activation during vowel utterance between different subjects, different days. These results will be helpful in use of facial electromyography for identification of speech and in other application such as human computer control.