We research on inaudible speech recognition with surface EMGs at around a speaker's lip and neck. When a person speaks naturally (without emphasizing his lip form), he controls power of his muscles with continuous change. That makes it difficult to find a boundary between two phonemes. So, it is important how we abstract distinctive features of sEMG at boundaries. In this paper, we propose a method to describe a trend of muscle activity, which depends on ``Center-of-Balance Transition Method'' proposed on our preceding works.
A Method to Detect Changes of Muscle Activity Using Difference Transition Based on Center-of-Balance Transition Method for Inaudible Speech Recognition