A Study of Playlist Generation Technique for Moderate Computational Resource Environment

In this paper, we report our work on exploring techniques to automatically generate the playlists for personal media player. The proposed approach intends to be able to perform on a standard personal computer platform. Given the first song, the system automatically generates the playlist from personal music collection based on the continuity of the melody and rhythm. Our technique relies only on the feature generated directly from the audio signal to avoid the problem that could arise from the lack of song tags and metadata. Two techniques, namely kd-tree and dynamic time warping, were evaluated and compared in terms of user satisfaction on the generated playlist and the execution time.