Human Robot Interaction in a Drum Playing Game

This paper presents a system that enables the drum beats to be automatically detected using auditory cues and will be used ina drum-based human-robot interaction game. In pursuit of creating an interactive drumming game, where a human user is able to play drums with a humanoid robot relying on turn-taking and imitation principles, in this work, we specifically focus on creating an online and real-time system to process the audio streams acquired by a microphone. The idea of discriminating sounds of a toy drum set from other sounds in the environment enables the user to have a realistic experience of playing drums and having fun at the same time. This is achieved by generating patterns in the spectral domain belonging to the drum beats and classifing them based on the result of an instantaneous comparison with the existing auditory drum templates learned by the system. To improve the noise robustness of the recognition system against the background noise in different environments, the drum beat patterns are learned online without requiring any prior information. The contribution of the calibration procedure to the results, as well as the applied noise reduction to the performance is evaluated using criteria such as detection rates, recognition rates and error. We demonstrate that the proposed system achieves a high performance of drum beat recognition (around 97 percent) in the online experiments.