Dear list,
I would like to share an open-source tool that may be useful to anyone
preparing auditory stimuli for autonomic or psychophysiological experiments.
The problem. Studies on sound and autonomic arousal - sleep, anxiety,
cardiac vagal tone, biofeedback - routinely report a custom subset of level,
temporal, spectral and psychoacoustic descriptors. Because each lab chooses
its own window lengths, A-weighting timebase, onset detector and library
versions, the resulting tables are not comparable across studies even when
they nominally report the same parameter.
DEBUSSY implements an eleven-item minimum acoustic reporting guideline in
one call, with the analysis choices fixed and documented rather than left to
the caller:
pip install debussy-audio
from debussy import analyze_audio
r = analyze_audio("stimulus.wav")
r.laeq_dbfs_a, r.roughness_asper, r.sharpness_acum
r.to_json()
It composes librosa for time-frequency descriptors with MOSQITO for the
ISO/DIN psychoacoustic models (Zwicker loudness, DIN 45692 sharpness,
Daniel-Weber roughness), and adds a small number of internally implemented
metrics. Every parameter carries an evidence tier, which separates "is this
stimulus admissible?" from "what does this stimulus do?".
Validation. A 60-track benchmark spanning DEAM low-arousal, DEAM mid-to-high
plus a genre-stratified FMA-medium subset, and a set of breath-paced
clinical stimuli. The manifest, the raw per-track parameter matrix and the
scripts that regenerate the figures and statistics are all published, so the
numbers can be checked without the audio. The between-music-category
contrast is explicitly underpowered and reported as such; the benchmark's
purpose is to show every parameter returns a sensible, bounded distribution
on very different material.
Source:
https://github.com/hyeonjoong/debussy Docs:
https://hyeonjoong.github.io/debussy Demo (no install):
https://huggingface.co/spaces/jjjooong/debussy MIT licensed, Python 3.10-3.12, tested on Linux and macOS
The reporting guideline comes from a narrative review currently under review
at Neuroscience and Biobehavioral Reviews; a software paper is in
preparation for JOSS.
I would genuinely value criticism of the parameter choices and the fixed
defaults, particularly from anyone who has fought the same comparability
problem. Issues and pull requests are welcome.
Disclosure: I am research director at Bell Therapeutics, which develops
sound-based sleep interventions. One of the three validation categories is
our clinical stimulus set. The tool itself is independent and MIT licensed.
Hyeon-Joong Kim
NeuroTech Research Institute, Bell Therapeutics, Seoul
ORCID 0000-0002-2898-0464