Back to Gap Map Model components / Language-specific datasets

Vuk'uzenzele isiXhosa Speech Dataset (ViXSD)

Lelapa AI
gated / Overall score: 1.7

The Vuk'uzenzele isiXhosa Speech Dataset (ViXSD) is about 10 hours of read isiXhosa speech in which eight native speakers, four men and four women, narrate articles from the Vuk'uzenzele corpus. Each of its 395 recordings comes with a transcript and speaker metadata on age, accent and linguistic background. Lelapa AI released it as the proof of concept for its community-centered Esethu Framework and license.

Openness

3 high confidence
3.0
license
Esethu-License (card; API metadata says "other")
access
auto (automatic Hugging Face gate)
dataset_card
present (contents, speakers, fields, splits and a datasheet)

It opens after an automatic click-through and is governed by the Esethu License, a community-centric data license designed to share benefits with the speakers' community, rather than a standard open license. African entities may use it commercially for free and other commercial users pay a fee, so commercial use is possible but not on equal terms for everyone.

Adoption

1 high confidence
1.0

Hugging Face downloads of the single repository, behind an automatic gate. Downloads count file fetches, not users.

Capability

2 medium confidence
2.0

ViXSD gives isiXhosa read speech from eight native speakers, documented with a paper, speaker metadata and a datasheet, though nothing beyond its own paper's recognition experiments uses it. Judged against what isiXhosa has, an official language with over a million speakers and a small written web, its ten hours sit one step below the Khmer ASR Cultural Dataset.

Verified 2026-09-24