Every tap by every participant on the 200 GlobalTap benchmark excerpts, in music time. These are the onsets the consensus pipeline consumes: REPP-aligned, expressed in seconds from the first audio sample of the 30 s excerpt, and restricted to the excerpt window.
globalmood200_taps.csv.gz 200 excerpts, 746 participants, 1,201,342 taps
globalmood200_trials.csv.gz one row per trial (23,493) with the recruitment panel
| Column | Description |
|---|---|
participant |
Pseudonymous label (gt-####), stable across the tap and trial tables |
stem_id |
Excerpt identifier, joins to data/stimuli_index_globalmood200.csv |
tap_time_s |
Tap onset in music time, seconds, rounded to the microsecond |
corpus |
globalmood_200 |
country_panel |
ISO-2 recruitment panel |
source_batch |
Collection batch |
n_taps |
Taps in that trial |
Nothing that could identify a participant is included: no platform or session
identifiers, no timestamps, no demographics, no free text. Benchmark
demographics, with the range of panel summaries, are in
supplementary.html §3.
Only the benchmark taps are released. For the validation set, the repository
provides the Stage-2 peaks in
../stage_inputs/canonical240/ and the
consensus annotations in
../annotations/canonical240/, which are the
inputs and beat grids used to render the click-track rating stimuli.
The tables contain the usable trials behind the counts reported in the paper: a
trial is kept when the platform did not mark it failed and both the participant
and the excerpt are identified. That is 23,493 trials from 746 participants,
114–120 tappers per excerpt (median 117), with per-panel participant counts in
data/country_panels.csv.
The paper’s collection total of 2,185,514 parsed taps counts every onset REPP recovered across both studies, including trials that failed quality checks and onsets outside the 30 s window. Those are not part of the release.
scripts/kde_peaks_from_taps.py recomputes Stage 2 from these tables using the
Stage-2 code path and the fixed settings in
METHOD_PARAMETERS.md, so
the benchmark can be rebuilt from raw taps in a fresh clone without audio:
python scripts/kde_peaks_from_taps.py --verify
This recreates the included
data/stage_inputs/globalmood200/cleaned_kde_peaks.csv file, matching it
excerpt by excerpt within the 5 ms analysis grid. Stage 4 then turns those peaks
into the released consensus annotations
(../annotations/).
scripts/build_tap_release.py builds these tables from the raw session
exports, which are not included here. Taps are parsed with
supplementary_analyses/pipeline_subset_core.parse_analysis_taps, the same
routine the published analyses use.
Licensed CC BY 4.0, as with the other data in this repository.