Beatmap generator for Osu Game using machine learning approach

Abstract

Rhythm game as one of the most-played game genres has its own attractiveness. Each song in the game gives its player new excitement to try another song or another difficulty level. However, behind every song being played is a lot of work. A beatmap should be created in order for a song to be played in the game. This paper presents an alternate way to create a beatmap that is considered playable for Osu Game utilizing beat and melody detection using machine learning approach and SVM as its learning method. The steps consists of notes detection and notes placement. Notes detection basically consists of features extraction from an audio file using DSP Java Library and learning process using Weka and LibSVM. However, detect the presence of notes only does not solve anything. The notes should be placed in the game using PRAAT and Note Placement Algorithm. From this process, a beatmap can be created from a song in about 3 minutes and the accuracy of the note detection is 86%.