Human Voice Dataset

Dataset Overview

• Purpose: This dataset is designed to streamline research on voice‑controlled musical interfaces. It facilitates benchmarking of vocal feature detection algorithms (e.g., pitch detection, onset detection) and provides training material for machine‑learning models.
• Current Version: Recordings from one singer are provided; additional singers will be added in the coming weeks.

Vocal Features

• Notes: Explored in semitone intervals from the lowest to the highest range, e.g., c3.wav, c#3.wav, d3.wav, ...
• Vowels: Form a limited‑value dimension (a, e, i, ...), e.g., _-a-[note].wav, _-e-[note].wav, ...
• Consonants: Must be combined with a vowel to be intelligible, e.g., t-a-[note].wav, t-u-[note].wav, ...
• Dynamics: Volume changes, pitch bends, vibrato (currently unavailable)

Dataset Structure

• File Naming Pattern: [consonant]-[vowel]-[note]-[dynamic].wav
• Directory Layout:
  • data/voices/
    • martin/
      • notes/
        • sources/
        • exports/
      • vowels/
        • sources/
        • exports/
      • consonants/
        • sources/
        • exports/

Singer Information

• Property Files:
  • singer.properties: Contains age, gender, nationality, etc.
  • recorder.properties: Contains recording equipment, recording conditions, etc.

Dataset Expansion

• Adding Samples:
  1. Clone the repository: git clone https://github.com/vocobox/human-voice-dataset.git
  2. Copy the singer folder and commit changes: git add ., git commit -m "[new singer] barbara", git push origin master

Other Useful Sound Datasets

• Piano Note Dataset: MAPS Database
• Singing Voice Dataset: Singing Voice Dataset
• Speech Corpora: CMU Speech, VoxForge, TED‑LIUM Corpus
• Sound and Instruments: IRMAS Dataset