This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, reproduction and adaptation in any medium and for any purpose provided that it is properly attributed. For attribution, the original author(s), title, publication source (PeerJ PrePrints) and either DOI or URL of the article must be cited.
Automatic music transcription is a difficult task that has provoked extensive research on transcription systems that are predominantly general purpose, processing any number or type of instruments sounding simultaneously. This paper presents a polyphonic transcription system that is constrained to processing the output of a single instrument with an upper bound on polyphony. For example, a guitar has six strings and is limited to producing six notes simultaneously. The transcription system consists of a novel pitch estimation algorithm that uses a deep belief network and multi-label learning techniques to generate multiple pitch estimates for each audio analysis frame, such that the polyphony does not exceed that of the instrument. The implemented transcription system is evaluated on a compiled dataset of synthesized guitar recordings. Comparing these results to a prior single-instrument polyphonic transcription system that received exceptional results, this paper demonstrates the effectiveness of deep, multi-label learning for the task of polyphonic transcription.
This paper is a preprint for possible future publishing in a conference or a journal.