zhu3000/squillo-vocal-quality

Squillo & VocalSet Vocal Quality Dataset

This dataset contains acappella singing voice recordings evaluating vocal quality, passaggio transition stability, and emotional dynamic range across two distinct cohorts:

1. SquilloClean Dataset: 24 .mp3 audio files across 6 anonymized vocalists (Singers A through F).
2. VocalSet Standardized Benchmark Subset: 60 .wav audio files across 20 professional vocalists (Singers F1–F9, M1–M11).

Dataset Structure

1. SquilloClean Cohort (Root Directory)

Four recordings per singer (Singers A–F):

• {Singer} scale.mp3: Vibrato vs straight tone scale performance.
• {Singer} passagio.mp3: Passaggio register break shift exercise.
• {Singer} song.mp3: Free choice song performance.
• {Singer} caro mio ben.mp3: Classical aria performance ("Caro Mio Ben").

2. VocalSet Benchmark Cohort (vocalset/ Directory)

Three standardized recordings per singer across 20 professional vocalists (F1–F9, M1–M11):

• vocalset/{Singer} scale.wav: Vibrato scale performance.
• vocalset/{Singer} passagio.wav: Passaggio transition exercise.
• vocalset/{Singer} caro mio ben.wav: Standardized classical aria performance ("Caro Mio Ben").

---

VocalSet Attribution & Citation

Note on VocalSet Excerpts: The files located in the vocalset/ directory are excerpted and standardized subsets from the original open-source VocalSet singing voice dataset (Wilkins et al., 2018), selected specifically for our 3-factor vocal quality study (scale, passagio, and caro mio ben).

For the complete, unedited original VocalSet dataset (comprising 10.1+ hours of audio across 17 extended vocal techniques), please refer to the original Hugging Face repository and Zenodo citation:

• Original Hugging Face Dataset: Bill13579/vocalset-mirror
• Official Zenodo Repository & Paper Citation: Zenodo DOI: 10.5281/zenodo.1442513

@dataset{vocalset2018,
  author       = {Wilkins, Julia and Seetharaman, Prem and Wahl, Alison and Pardo, Bryan},
  title        = {VocalSet: A Singing Voice Dataset},
  year         = 2018,
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.1442513},
  url          = {https://doi.org/10.5281/zenodo.1442513}
}

---

Python Usage

You can download the unified dataset programmatically using huggingface_hub:

from huggingface_hub import snapshot_download

# Download all dataset files to local cache
dataset_dir = snapshot_download(repo_id="zhu3000/squillo-vocal-quality", repo_type="dataset")
print(f"Dataset downloaded to: {dataset_dir}")