Papers › Beat this! Accurate beat tracking without DBN postprocessing

Beat this! Accurate beat tracking without DBN postprocessing

31 Jul 2024arXiv:2407.21658archive 2025-07-28

We propose a system for tracking beats and downbeats with two objectives: generality across a diverse music range, and high accuracy. We achieve generality by training on multiple datasets -- including solo instrument recordings, pieces with time signature changes, and classical music with high tempo variations -- and by removing the commonly used Dynamic Bayesian Network (DBN) postprocessing, which introduces constraints on the meter and tempo. For high accuracy, among other improvements, we develop a loss function tolerant to small time shifts of annotations, and an architecture alternating convolutions with transformers either over frequency or time. Our system surpasses the current state of the art in F1 score despite using no DBN. However, it can still fail, especially for difficult and underrepresented genres, and performs worse on continuity metrics, so we publish our model, code, and preprocessed datasets, and invite others to beat this.

PaperPDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

CPJKU/beat_this officialmentioned on GitHubpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Beat TrackingDownbeat Tracking

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Beat Tracking ASAP Beat This! F1 76.3 #1 of 1 Archive leaderboard report
Beat Tracking Ballroom Beat This! F1 97.5 #1 of 1 Archive leaderboard report
Beat Tracking Beatles Beat This! F1 94.5 #1 of 1 Archive leaderboard report
Beat Tracking Candombe Beat This! F1 99.7 #1 of 1 Archive leaderboard report
Beat Tracking Filosax Beat This! F1 99.5 #1 of 1 Archive leaderboard report
Beat Tracking GTZAN Beat This! F1 89.1 #1 of 1 Archive leaderboard report
Beat Tracking Groove Beat This! F1 93.7 #1 of 1 Archive leaderboard report
Beat Tracking GuitarSet Beat This! F1 92.0 #1 of 1 Archive leaderboard report
Beat Tracking HJDB Beat This! F1 98.2 #1 of 1 Archive leaderboard report
Beat Tracking Hainsworth Beat This! F1 91.9 #1 of 1 Archive leaderboard report
Beat Tracking Harmonix Beat This! F1 95.8 #1 of 1 Archive leaderboard report
Beat Tracking JAAH Beat This! F1 95.1 #1 of 1 Archive leaderboard report
Beat Tracking SIMAC Beat This! F1 77.9 #1 of 1 Archive leaderboard report
Beat Tracking SMC Beat This! F1 62.7 #1 of 1 Archive leaderboard report
Beat Tracking TapCorrect Beat This! F1 93.0 #1 of 1 Archive leaderboard report
Downbeat Tracking ASAP Beat This! F1 61.2 #1 of 1 Archive leaderboard report
Downbeat Tracking Ballroom Beat This! F1 95.3 #1 of 1 Archive leaderboard report
Downbeat Tracking Beatles Beat This! F1 88.8 #1 of 1 Archive leaderboard report
Downbeat Tracking Candombe Beat This! F1 99.7 #1 of 1 Archive leaderboard report
Downbeat Tracking Filosax Beat This! F1 98.5 #1 of 1 Archive leaderboard report
Downbeat Tracking GTZAN Beat This! F1 78.3 #1 of 1 Archive leaderboard report
Downbeat Tracking Groove Beat This! F1 82.1 #1 of 1 Archive leaderboard report
Downbeat Tracking GuitarSet Beat This! F1 88.1 #1 of 1 Archive leaderboard report
Downbeat Tracking HJDB Beat This! F1 96.6 #1 of 1 Archive leaderboard report
Downbeat Tracking Hainsworth Beat This! F1 80.0 #1 of 1 Archive leaderboard report
Downbeat Tracking Harmonix Beat This! F1 90.7 #1 of 1 Archive leaderboard report
Downbeat Tracking JAAH Beat This! F1 85.0 #1 of 1 Archive leaderboard report
Downbeat Tracking TapCorrect Beat This! F1 86.4 #1 of 1 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

AdamWAttentionBatch NormalizationDense ConnectionsDropoutLinear LayerMulti-Head AttentionPosition-Wise Feed-Forward LayerRMSNormResidual ConnectionSoftmaxTransformer

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections