In this paper, we present a novel cover detection approach leveraging lyrics information extracted from audio. We use as Singing Voice Recognition system to extract automatically a lyrics transcripts from the audio and Cover detection is then framed as a noisy text-matching task between obtained transcripts.

As songs may not contain lyrics we complement this lyrics-based cover detection system with a more classic tonal-based method.

We show extensive empirical evidence that lyrics information can significantly help cover detection and that both lyrics and tonal information are complementary for this task.

This paper has been accepted for publication in the proceedings of the International Society for Music Information Retrieval Conference (ISMIR 2021).