Content-based music retrieval using spectral envelope feature and K-means algorithm

In multimedia database, music retrieval is an important research topic. Current music searching is based on text indexing. However, this kind of method has some drawbacks. It is difficult to remember the text keywords such as song name, author name, singer name or the lyric of songs. Content-based music searching overcomes these drawbacks. In state of art approaches, fearure vetors of music signal are built on their physical characteristics as volume, energy, and spectrum. Recently, some researches use another approach, which based on the signal processing techniques incorporating with human auditory analysis. This approach minimizes the redundant information as well as accurately represents the music signal. This paper presents a method of song searching using Mel ceptral spectral envelope and K-means algorithm.