This gated dataset contains 400 randomly selected prepared speech segments: 100 each in Igbo, Hausa, Nigerian Pidgin, and Yoruba. Every transcript preserves its existing inline audio-event tags. Audio is FLAC, 48 kHz, mono, and 24-bit.

Access requests require manual approval by the dataset owner. The dataset card intentionally contains no sample transcripts or audio.

Columns

audio: decoded automatically from file_name
text: exact prepared transcript, including inline tags
annotation_tags and normalized_tags: existing tag annotations
language, source Mac, timing, speaker, recording, chunk, and integrity metadata

Downloads last month: 13

Total file size:

660 MB