Paper year
2025
Paper dossier
Multi-Class-Token Transformer for Multitask Self-supervised Music Information Retrieval
Detail viewSimilarity handoff
Review source metadata, abstract, authors, topics, and local similarity context before moving into explanation and ranking views.
Citations
0
Authors
0
Topic labels
0
Source readout
Source and corpus status
Venue
Unknown venue
Source slug
unknown
Corpus placement
Controlled edge slice
Similarity rows
Not available yet
Ranking readout
Where this paper lands in the current run
Ranking details could not be loaded (API 503).
Abstract
Contrastive learning and equivariant learning are effective methods for self-supervised learning (SSL) for audio content analysis. Yet, their application to music information retrieval (MIR) faces a dilemma: the former is more effective on tagging (e.g., instrument recognition) but less effective on structured prediction (e.g., tonality estimation); The latter can match supervised methods on the specific task it is designed for, but it does not generalize well to other tasks. In this article, we adopt a best-of-both-worlds approach by training a deep neural network on both kinds of pretext tasks at once. The proposed new architecture is a Vision Transformer with 1-D spectrogram patches (ViT-1D), equipped with two class tokens, which are specialized to different self-supervised pretext tasks but optimized through the same model: hence the qualification of self-supervised multi-class-token multitask (MT2). The former class token optimizes cross-power spectral density (CPSD) for equivariant learning over the circle of fifths, while the latter optimizes normalized temperature-scaled cross-entropy (NT-Xent) for contrastive learning. MT2 combines the strengths of both pretext tasks and outperforms consistently both single-class-token ViT-1D models trained with either contrastive or equivariant learning. Averaging the two class tokens further improves performance on several tasks, highlighting the complementary nature of the representations learned by each class token. Furthermore, using the same single-linear-layer probing method on the features of last layer, MT2 outperforms MERT on all tasks except for beat tracking; achieving this with 18x fewer parameters thanks to its multitasking capabilities. Our SSL benchmark demonstrates the versatility of our multi-class-token multitask learning approach for MIR applications.
Authors
No authors available.
Neighborhood labels
Topics
0 labels
Topic labels are imported metadata and can be noisy; use them as coarse navigation hints, not authoritative classifications.
Neighbor surface
Similar papers
Similar papers use a separately configured neighbor embedding; it may differ from the embedding version used by the current ranked run.
No embedding-backed neighbors available for this paper/version yet.
Next handoff
Best next moves from here
01
Check recommendation families
Use Recommended to see whether this paper behaves like an emerging or undercited signal in the current ranked feed, or how it appears on the bridge preview / diagnostics view.
02
Inspect nearby topics
Use Trends to understand whether its attached labels are heating up or cooling down inside the curated corpus.
03
Cross-check evaluation baselines
Use Evaluation to compare the dossier readout against citation and recency baselines for the same resolved family run.