Tasks & Data

Dataset Participants

The dataset comprises long-term neck-surface accelerometer (ACC) features collected during daily life from 582 total individuals, including patients diagnosed with phonotraumatic (PVH) and nonphonotraumatic vocal hyperfunction (NPVH), along with age-, sex-, and occupation-matched vocally healthy controls:

• 213 patients with PVH and 169 matched controls

• 116 patients with NPVH and 84 matched controls

Main Features & Data Format

ACC-derived time-series features with 50 ms frame-resolution (.mat):

• Fundamental frequency (fo)

• Amplitude-based sound pressure level (SPL) estimated from ACC signal

• Cepstral peak prominence (CPP)

• Difference between the magnitudes of the first and second harmonics (H1–H2)

• Spectral tilt 

• Low-to-high spectral power ratio (L/H ratio)

• Glottal airflow–based measures derived from ACC (IBIF)

  • For a small number of participants, IBIF features are not available due to data-collection limitations. Teams should confirm the availability of IBIF variables in each .mat file and treat missing values accordingly

Behavioral masks (.mat):

• Voice detection activity binary labels 

• Singing detection activity binary labels

Demographics (.csv):

• Participant sex, anonymized subject ID, and anonymized monitoring date

Diagnostic labels (.csv):

• Provided only in the training dataset (removed from test data)

Challenge Tasks

Models should be built for two binary classification tasks using the labeled training dataset

• Task 1 — PVH detection: Classify PVH (label = 1) vs. non-PVH (label = 0, where non-PVH includes NPVH + healthy controls)

• Task 2 — NPVH detection: Classify NPVH (label = 1) vs. non-NPVH (label = 0, where non-NPVH includes PVH + healthy controls)

Modeling data options: 

• Full time-series (frame or hourly aggregations)

• Use subject-level cross-validation to prevent leakage across monitored days

Further methodological specifications are provided on the Rules & Ethics page

Submission & Evaluation

Submission: 

1. Classification results on the test set (one CSV per task) including three columns:

• Subject ID, predicted probability, and predicted label 

  • Example: NV064, 0.87, 1

                                NV089, 0.24, 0

2. Brief technical report (PDF, 1–2 pages) including:

• Data preprocessing techniques

• Feature selection or representation

• Model configuration

• Validation approach (should be subject-level)

3. Source code used to generate the submitted results

Evaluation: 

• Primary ranking metric: Area Under the ROC Curve (AUC) computed on the held-out unlabeled test set.

• Secondary review: Applied only if needed to assess reproducibility and methodological clarity (based on report and code).

• Only final testing results will be considered