A Compact Spectrum-Assisted Human Beatboxing Reinforcement Learning Tool On Smartphone

Music is expressive and hard to be described by words. Learning music istherefore not a straightforward task especially for vocal music such as humanbeatboxing. People usually learn beatboxing in the traditional way of imitatingaudio sample without steps and instructions. Spectrogram contains a lot ofinformation about audio, but it is too complicated to be understood inreal-time. Reinforcement learning is a psychological method, which makes use ofreward and/or punishment as stimulus to train the decision-making process ofhuman. We propose a novel music learning approach based on the reinforcementlearning method, which makes use of compact and easy-to-read spectruminformation as visual clue to assist human beatboxing learning on smartphone.Experimental result shows that the visual information is easy to understand inreal-time, which improves the effectiveness of beatboxing self-learning.