We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Sound

Title: Self-Supervised Beat Tracking in Musical Signals with Polyphonic Contrastive Learning

Abstract: Annotating musical beats is a very long in tedious process. In order to combat this problem, we present a new self-supervised learning pretext task for beat tracking and downbeat estimation. This task makes use of Spleeter, an audio source separation model, to separate a song's drums from the rest of its signal. The first set of signals are used as positives, and by extension negatives, for contrastive learning pre-training. The drum-less signals, on the other hand, are used as anchors. When pre-training a fully-convolutional and recurrent model using this pretext task, an onset function is learned. In some cases, this function was found to be mapped to periodic elements in a song. We found that pre-trained models outperformed randomly initialized models when a beat tracking training set was extremely small (less than 10 examples). When that was not the case, pre-training led to a learning speed-up that caused the model to overfit to the training set. More generally, this work defines new perspectives in the realm of musical self-supervised learning. It is notably one of the first works to use audio source separation as a fundamental component of self-supervision.
Comments: 60 pages, 21 figures, masters thesis, degree granted
Subjects: Sound (cs.SD); Machine Learning (cs.LG); Audio and Speech Processing (eess.AS)
MSC classes: 68T07 (Primary) 00A65, 68T10 (Secondary)
ACM classes: I.5.4
Cite as: arXiv:2201.01771 [cs.SD]
  (or arXiv:2201.01771v1 [cs.SD] for this version)

Submission history

From: Dorian Desblancs [view email]
[v1] Wed, 5 Jan 2022 13:24:39 GMT (7568kb,D)

Link back to: arXiv, form interface, contact.