Adaptive Blowing Interaction Method Based on a Siamese Network

Abstract

Breathing is a natural and directly controllable human activity. Currently, some works have considered breath as a direct input controlling mechanism. The equipment relied upon in these works is generally complicated, expensive, inconvenient to wear, and sometimes insufficiently controllable. The use of breathing interaction is also limited to a certain scene and is not universal. This paper proposes an adaptive interaction method, which is a natural and directly controllable interaction based on blowing air that only uses headset microphones to obtain the sound waveform of the blowing action without requiring expensive equipment, and that can be used conveniently anytime and anywhere. This blowing interaction uses a Siamese network to achieve “self-adaptation” - the first step adapts to noise interference, including environmental noise and the user's own speaking interference, and the second step adapts to different users and equipment, that is, the blowing interaction is used by different people or on different equipment, and the interaction mode can accurately identify the type of blowing. This paper also develops several applications of the blowing interaction method to test the algorithm. During tests, it's proved that this interface not only increases the type of blowing used for interaction but also eliminates interference from speaking in a normal volume effectively and addresses the problem of individual differences.

Keywords

• Blowing interaction
• domain adaptation
• Siamese network
• adaptive