ENST-Drums: an extensive audio-visual database for drum signals processing

Abstract

One of the main bottlenecks in the progress of the Music Information Retrieval (MIR) research field is the limited access to common, large and annotated audio databases that could serve for technology development and/or evaluation.The aim of this paper is to present in detail the ENST-Drums database, emphasizing on both the content and the recording process. This audiovisual database of drum performances by three professional drummers was recorded on 8 audio channels and 2 video channels. The drum sequences are fully annotated and will be, for a large part, freely distributed for research purposes. The large variety in its content should serve research in various domains of audio signal processing involving drums, ranging from single drum event classification to complex multimodal drum track transcription and extraction from polyphonic music.

Tasks

Reproductions