Towards a robust drum stroke recognition system for human robot interaction

This paper presents a system that enables the drum strokes to be automatically detected and recognized using auditory cues. In pursuit of creating an interactive drumming game, in which a human user is able to play drums with a humanoid robot (Robotic Drum Mate) 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 a more robust/fruitful human-robot interaction at the same time. This is achieved by learning auditory drum stroke templates and improving the noise robustness of the recognition system against the background noise without requiring any prior information. The system is evaluated using criteria such as detection and recognition rates in the online experiments in real world. We verify the applicability of our system in a drum playing game with human test subjects and assess the effects of the training session prior to the game and the feedback mechanism utilized on the robot about the drumming performance of the user.