Nzyoka19/Kenyan-Swahili-Speech

Read Speech in Kenyan Swahili (6h)

A single-speaker read speech dataset in Kenyan Swahili, containing approximately 6 hours of prompted recordings from an anonymous male speaker. The dataset was produced as part of CLEAR Global's Gamayun Language Data Kits initiative, which develops open-source language resources for under-resourced languages used in humanitarian contexts.

Dataset description

• —Language: Swahili (Kenyan variety)
• —Speaker: 1 anonymous male Kenyan speaker
• —Recording type: Prompted read speech (speaker read sentences aloud from a script)
• —Total duration: ~6 hours (21,852 seconds)
• —Utterances: 4,700
• —Audio format: WAV, pre-segmented
• —Transcriptions: Swahili translations of sentences from the Tatoeba repository

The sentence set is shared with CLEAR Global's Gamayun Swahili–English parallel text kit. English source sentences were selected from Tatoeba using a frequency-based algorithm documented in the corepus-gen repository; Swahili translations were produced by the CLEAR Global translator community.

Loading the dataset

from datasets import load_dataset

ds = load_dataset("CLEAR-Global/Kenyan-Swahili-Speech")

File structure

├── clips/
│   ├── 1/          # recording session subdirectory
│   │   └── *.wav
│   ├── 7/
│   └── ...
├── metadata.csv    # file_name, transcription, duration
└── LICENSE

The metadata.csv file contains:

Restrictions and forbidden uses

This dataset is released under CC-BY-NC-4.0. In addition to the license terms, the following uses are explicitly forbidden:

• —Attempting to determine the identity of the speaker
• —Voice cloning or TTS synthesis that matches the characteristics of the speaker
• —Re-hosting or re-sharing this dataset without CLEAR Global's explicit permission
• —Commercial use of any kind
• —Usage without attribution to CLEAR Global

Citation

If you use this dataset, please cite *Gamayun – Language Technology for Humanitarian Response*:

@inproceedings{oktem2020gamayun,
  title     = {Gamayun -- Language Technology for Humanitarian Response},
  author    = {Öktem, Alp and Albayk Jaam, Muhannad and DeLuca, Eric and Tang, Grace},
  booktitle = {2020 IEEE Global Humanitarian Technology Conference (GHTC)},
  year      = {2020},
  address   = {Virtual},
  month     = {October 29 -- November 1}
}

Acknowledgements

This dataset was produced by CLEAR Global (formerly Translators without Borders). The Gamayun initiative is part of CLEAR Global's effort to develop open-source language technology for under-resourced languages used in humanitarian contexts.