Datasets:

vadette
/

macro_prosody_sample_set

license: cc0-1.0
language:
  - hi
tags:
  - audio
  - speech
  - prosody
  - linguistics
  - acoustic-features
  - indo-aryan
pretty_name: Hindi Macro-Prosody Telemetry

Hindi Macro-Prosody Telemetry

Language: Hindi (hi) Language Family: Indo-Aryan (Indo-European) Source Corpus: Mozilla Common Voice CV24 Speech Type: Scripted License: CC0-1.0 (Public Domain)

Description

Macro-prosodic acoustic telemetry extracted from Hindi speech in the Mozilla Common Voice CV24 corpus. Largest corpus in this sample release with full demographic metadata (age, gender, accent). No audio included. CC0 licensed.

Dataset Contents

Field	Value
Rows	18,447
Suppressed (k=5)	34
Format	Parquet (Snappy compressed)
Acoustic metrics	20+ per clip
Audio included	No

Acoustic Features

Pitch (mean, std, range), intensity (mean, max, range), harmonics-to-noise ratio, cepstral peak prominence (CPPS), spectral tilt, zero-crossing rate, formants (F1/F2/F3), normalized pairwise variability index (nPVI), articulation rate, speaking rate, pause rate, jitter, shimmer, MFC delta, TEO mean/std, speech ratio, SNR, C50.

Anonymization

Sequential clip IDs replace original Mozilla filenames
Transcripts removed; approximate syllable count provided
All continuous acoustic variables truncated to 2 decimal places (float32)
Duration bucketed to nearest 100ms
K-anonymity k=5 applied on {gender, age_bucket, duration_bucket}
Anonymization standard: orator_forge_k5_v1

Provenance

Derived from the [Mozilla Common Voice](Mozilla Common Voice CV24) CC0 corpus. Acoustic features extracted via Parselmouth/Praat at 16kHz. Quality-gated via Brouhaha (SNR/C50/VAD scoring). No original audio is included in this dataset.

Citation

If you use this dataset, please credit:

Mozilla Common Voice Project (CC0) — acoustic telemetry by Orator Forge

Part of the Alexandria Voice Corpus

This is a sample pack from the Alexandria Voice Corpus, a multilingual macro-prosodic telemetry database covering 60+ languages.