Stereo panning information for music information retrieval tasks

Abstract

Recording engineers, mixers, and producers play important yet often overlooked roles in defining the sound of a particular record, artist, or group. The placement of different sound sources in space using stereo panning is an important component of their work. Stereo panning information typically is not utilized in music information retrieval (MIR) tasks such as genre and artist classification. A set of audio features is proposed that can be used to characterize stereo panning information and contrast two different methods of calculation. These features are shown to provide statistically important information for nontrivial audio classification tasks and are compared with the traditional mel-frequency cepstral coefficients for different MIR tasks. They can also be viewed as a first attempt to capture extramusical information related to the production process through music information retrieval techniques.