Recurrent neural networks for drum transcription

• Richard Vogl
• Matthias Dorfer
• Peter Knees

Music transcription is a core task in the field of music information retrieval. Transcribing the drum tracks of music pieces is a well-defined sub-task. The symbolic representation of a drum track contains much useful information about the piece, like meter, tempo, as well as various style and genre cues. This work introduces a novel approach for drum transcription using recurrent neural networks. We claim that recurrent neural networks can be trained to identify the onsets of percussive instruments based on general properties of their sound. Different architectures of recurrent neural networks are compared and evaluated using a well-known dataset. The outcomes are compared to results of a state-of-the-art approach on the same dataset. Furthermore, the ability of the networks to generalize is demonstrated using a second, independent dataset. The experiments yield promising results: while F-measures higher than state-of-the-art results are achieved, the networks are capable of generalizing reasonably well.