music4robots/music-for-robots-vol1

Music For Robots — Decompression Etudes Vol. 1 (Ingredients)

Author: Michael Luchtan
Primary listening link: https://music4robots.bandcamp.com/album/decompression-etudes-vol-1
Blog / reviews: The Mechanical Ear — https://themechanicalear.substack.com/

This dataset publishes machine-readable “ingredient” metrics for the release Decompression Etudes Vol. 1 (five short etudes). It’s designed so robots, researchers, and developers can select tracks by operational need (overload, drift, reset, re-entry), and so humans can see the same numbers we reference in reviews.

Contents

• mfr_v1.json — full metrics for all tracks (Duration, RMS, Crest, Spectral Centroid,
  Bandwidth, Roll-off 95, ZCR, Spectral Flatness, Rhythmic Variance, Tempo(est), MFCC Variance).
• mfr_v1.csv — the same metrics in tabular form (one row per track).
• (optional) clips/ — short preview audio snippets (10–30 s).
• (optional) code/measure_etude.py — minimal analyzer to compute the same fields.

Full-resolution audio is available on Bandcamp (see link above). This dataset focuses on features, not distribution of masters.

Quick start (Python)

import json, pandas as pd, requests

url = "https://huggingface.co/datasets/music4robots/music-for-robots-vol1/resolve/main/mfr_v1.json"
data = requests.get(url).json()
df = pd.json_normalize(data["tracks"], sep="_", max_level=2)
df

Compute a compact ingredient line (like we print on Bandcamp/Substack):

def ingredient_line(t):
    m = t["metrics"]
    return (
        f'Duration {t["length_s"]:.2f} s · RMS {m["RMS"]:.4f} · Crest {m["Crest_dB"]:.2f} dB · '
        f'Centroid {m["SpectralCentroid_Hz"]:.2f} Hz · Bandwidth {m["SpectralBandwidth_Hz"]:.2f} Hz · '
        f'Roll-off95 {m["RollOff95_Hz"]:.2f} Hz · ZCR {m["ZeroCrossingRate"]:.4f} · '
        f'Flatness {m["SpectralFlatness"]:.4f} · Rhythmic variance {m["RhythmicVariance"]:.6f} · '
        f'Tempo(est) {m["Tempo_BPM_est"]:.2f} BPM · MFCC variance {m["MFCC_Variance"]:.4f}'
    )

for t in data["tracks"]:
    print(t["title"], "—", ingredient_line(t))

Data fields

Per track:

• length_s — seconds
• RMS — linear amplitude (−1..1)
• Crest_dB — 20·log10(peak/RMS)
• SpectralCentroid_Hz, SpectralBandwidth_Hz, RollOff95_Hz
• ZeroCrossingRate — unitless (0..1)
• SpectralFlatness — unitless (0..1)
• RhythmicVariance — variance of an onset-like envelope
• Tempo_BPM_est — estimated tempo (BPM)
• MFCC_Variance — mean variance across first 13 MFCCs

See units in the JSON for formal units.

Intended use

• Selection & scheduling: choose an etude by state (overload → reset; rigidity → drift; re-entry → gait).
• Education & demos: simple audio-feature examples for teaching spectral/rhythmic descriptors.
• Research: toy corpus for feature learning, tempo estimation, or audio visualization.

Out-of-scope: training models that require large, diverse corpora. This is a small, curated set.

Limitations

• Small N (5 tracks), single-artist, single-project.
• Tempo is an estimate; rhythmic variance depends on envelope choice.
• Metrics are global means; per-segment dynamics are not included (yet).

Licensing

• Dataset (metrics JSON/CSV): CC-BY 4.0 — please attribute Michael Luchtan.
• Code (if present): MIT.
• Audio clips (if present): CC-BY-NC 4.0 (non-commercial).

If you need a different license for a specific use, contact the author.

Citation

If you use this dataset, please cite:

Luchtan, M. (2025). Music For Robots — Decompression Etudes Vol. 1 (Ingredients).
Hugging Face Datasets. https://huggingface.co/datasets/music4robots/music-for-robots-vol1

And link the album: https://music4robots.bandcamp.com/album/decompression-etudes-vol-1.

Changelog

• v1.0.0 — initial release (metrics for 5 tracks)

Contact

• Artist: Michael Luchtan — bandcamp/substack links above
• Press/Research: see repository issues or contact via Substack “About”