User-independent Accelerometer Gesture Recognition for Participatory Mobile Music

Abstract

The general adoption of smartphones, and the rapid development and implementation of new web standards supporting their capabilities, have created a promising platform for participatory music. In this paper we analyze the use of accelerometer gesture recognition in this context, which brings the issue of generalizing to multiple users. We describe Handwaving, a system based on neural networks for real-time gesture recognition and sonification on mobile browsers. We evaluate the system using a multi-user dataset. Our results show that training with data from multiple users improves classification accuracy, supporting the use of the proposed algorithm for user-independent gesture recognition. Finally, we describe our experiences in participatory music using the system.