dsfsi/anv-za-sot-1h-sample-dataset

Sesotho Sample Dataset - Next Voices-ZA (South Africa) - Multilingual Speech Dataset - Sesotho

This dataset includes scripted and unscripted speech across various domains such as agriculture, health, finance, sports, transport, culture, society and general topics. It is primarily designed for automatic speech recognition (ASR).

Folder structure

The dataset is organised hierarchically as follows:

Folder Structure

ANV-ZA-SOT-1h/
├── sot/                  # Folder for Sesotho
│   ├── recorder_uuid/    # Contains all audio files
│   │   ├── recording-1731053452.wav
│   │   ├── ...
│   ├── transcripts.csv    # Contains transcripts of all audio recordings
│   ├── meta.csv          # Contains additional metadata
├── README.md             # Description of the dataset

Data overview

Audio

• Format: 16-bit PCM WAV
• Sample rate: 48kHz

Transcriptions

• Provided in transcript.csv with fields:   - file_name: Name of the audio file.   - transcript: Text transcription of the audio.   - duration: Duration of the recording in seconds.   - type: Scripted or unscripted.

Metadata

• Provided in meta.csv with fields such as:   - recorder_uuid: Unique speaker identifier.   - age_range, gender

Citation Information

Bibtex Reference (last updated 10/02/2025)

@dataset{marivate_2024_14336304,
  author       = {Marivate, Vukosi and
                  Olaleye, Kayode and
                  Mundia, Sitwala and
                  van Wyk, Nia Zion and
                  Bakainaga, Andinda and
                  Morrissey, Graham and
                  Dunbar, Dale and
                  Smit, Francois and
                  Mogale, Hope Tsholofelo},
  title        = {ANV-SOT-Sample-1: Sesotho Sample Dataset - Next
                   Voices-ZA (South Africa)
                  },
  month        = dec,
  year         = 2024,
  publisher    = {Zenodo},
  version      = {0.0.2},
  doi          = {10.5281/zenodo.14336304},
  url          = {https://doi.org/10.5281/zenodo.14336304},
}