Department of Software Technology
Vienna University of Technology

The SOMeJB Music Digital Library - Experiments: SOMeJB 2: Collection 335

Below we provide some experimental results obtained from our experiments with a sub-collection consisting of 335 pieces of music, focusing on Electronic an pop music.

1. Audio Data

The 335 pieces of music are manually filed into 14 genres within 4 main categories, as depicted in Figure 1, listing the number of titles per category. Contrary to other settings, this sub-collection does not try to constitute a balanced corpus across a large variety of genres. It rather exhibits a strong focus on two categories of music (Electronic and Pop), which are sub-divided in finer granularity. This setting also does not try to limit itself to clean and nicely separable genres, resulting in significant challenges during the manual organization phase, with some titles not being straight-forward to classify. The collection also incorporates a number of songs in different versions, i.e. cover songs, remixes, or new interpretations of songs.

structuring of Collection335 into 4 main genre categories and subsequent second-layer subdivision
Figure 1: Structuring of Collection 335 into 4 main and 14 sub-categories.

A detailed listing of the various titles is provided below:

(Up to the top of the page)

2. Feature Vectors

The MP3-files have been preprocessed (mono-conversion, downsampling, etc. - see SOMeJB architecture description for more details), and subsequently segmented into 6-second segments. Specifically, the first two segments, as well as the last two segments of each piece of music have been removed to eliminate lead-in and fade-out effects. Every third 6-seconds segment has been retained for further analysis. For this set of experiments, the GENRE features were also extracted using the Marsyas System. Specifically, the following subset of features was extracted: (Up to the top of the page)

3. Self-Organizing Map

Below, we provide links to GHSOMs trained with the 335-Collection dataset. All maps are interactively explorable. Currently we do not provide links to the audio samples themselves for this data set.

(Up to the top of the page)

Up to the SOMeJB Homepage