Cymbal percussion: multimodal motion capture, audio and video

Multimodal recordings of cymbal-playing gestures captured at the fourMs Lab, University of Oslo, on 12 April 2011.

A percussionist performed a structured set of cymbal strokes (flams, rolls, staccato strokes, suspended-cymbal attack/nonattack/tremolo strokes and a reference take, at several dynamic levels, with and without damping, plus a free improvisation), recorded simultaneously with two motion capture systems and synchronised audio and video.

The two mocap systems are an optical infrared system controlled by Qualisys Track Manager (raw QTM captures and QAM marker models) and a full-body Xsens MVN inertial suit (native MVN/MVNX plus C3D exports). The data are distributed as six ZIP archives grouped by modality:

• QTM (72 raw Qualisys captures)
• XsensMvn (71 takes + naming key)
• MP4 video (72 clips, 720x480, 30 fps)
• WAV audio (72 files, 44.1 kHz mono)
• XsensC3D (71 takes + naming key)
• QAM (3-marker models).

Take names are identical across the QTM, MP4, and WAV archives (verified, zero mismatches), allowing optical capture, video, and audio of the same stroke to be matched by filename. The Xsens takes map to strokes via a bundled naming key. The original web page contained no descriptive text; the stroke taxonomy is inferred from the filenames, and the performer is not named on the page.

Version 2.1 (2026-07-30) — the optical side is now open for every take. Earlier versions carried the Qualisys optical data only as 72 proprietary .qtm projects, so for all but a couple of takes the concurrent inertial-against-optical pairing this dataset exists for could not be read without commercial software. The cause was not data loss: every .qtm holds its 3D point stream and its raw 2D camera data, but 70 of the 72 takes had never had their trajectories identified, and QTM omits unidentified trajectories on export. Re-exporting with unidentified trajectories included recovers all of it. The new QTM_tsv.zip adds 71 takes with unnamed trajectories — a median of 34 per take, of which a median of 29 are more than 90 per cent complete — plus 2 takes with named markers, including the session reference take suspeso_reference0001 (31 markers, all complete, covering the whole body plus both mallets and the cymbal). trajectory_inventory.tsv inside the archive lists every export with its trajectory count, rate, duration and completeness. These are QTM FILE_VERSION 2.0.0 exports.

Funding. Research Council of Norway (funder DOI 10.13039/501100005416): 322364 (fourMs Lab, University of Oslo).