Sign in to view source links and access this dataset
Description
A 2026 dataset containing approximately 125 hours of conversational speech across 16,383 audio samples. The VoiceAfrica project, a collaboration between Lanfrica and Meta, collected natural speech from 118 speakers in four countries to create expert-curated transcriptions for 11 under-represented African languages.
Use Cases
Train automatic speech recognition (ASR) models based on conversational audio data.
Benchmark multilingual speech processing systems based on the 11-language corpus.
Study natural speech patterns and accents based on recordings from 118 speakers across four countries.
Develop language preservation tools based on expert-curated transcriptions for under-represented languages.
Strengths
Contains ~125 hours of speech data, providing substantial audio material.
Focuses on 11 under-represented African languages, addressing a resource gap.
Uses natural, conversational speech from 118 speakers, likely offering authentic linguistic variation.
Provides expert-curated transcriptions, which may indicate high-quality ground truth labels.
Limitations
Column-level documentation is absent; field semantics must be inferred after download.
The dataset's specific license is unknown, which may restrict usage.
Row count is unknown, which may limit suitability assessment for large-scale training.
Provenance
Source
Lanfrica–Meta collaboration (code-named VoiceAfrica 1), authored by naijavoices.
Collection Method
Collection of natural, conversational speech from 118 speakers.
Time Range
null
Freshness
Last updated 2026-05-25 19:49:32; freshness should be verified.
Geography
Four unspecified African countries.
License restrictions are unknown; verify before use.