LRE22: 2022 NIST Language Recognition Test Data for 14 African Languages
by Greenberg, Craig / UBC Abacus Harvested Dataverse·Updated 2mo ago
Available on 1 platform
Sign in to view source links and access this dataset
Description
The 2022 NIST Language Recognition Evaluation Test and Development Sets contain approximately 222 hours of conversational telephone and broadcast narrowband speech. It was developed by the Linguistic Data Consortium and NIST to advance language recognition technology, with a focus on 14 African languages including low-resource ones. The data includes test segments, answer keys, and metadata extracted from three source collections: SASAL, MAGLIC, and LRAL.
Use Cases
Training language identification models based on conversational telephone speech (CTS) and broadcast narrowband speech (BNBS).
Benchmarking language recognition system performance on short-duration audio segments as emphasized in the evaluation.
Developing speech technology for low-resource African languages like Oromo, Tigrinya, Ndebele, and others mentioned in the description.
Analyzing acoustic features of Maghrebi Arabic varieties (Tunisian, Algerian, Libyan) and North African French from the MAGLIC collection.
Strengths
Includes approximately 222 hours of speech data across 14 languages.
Data is manually audited by native speakers for language and quality verification.
Focuses on African languages, including several low-resource languages, addressing a specific research need.
Segments are provided as standardized 8 kHz, 8-bit a-law SPHERE files with associated metadata.
Limitations
Column-level documentation is absent; field semantics must be inferred after download.
Row count is unknown, which may limit suitability assessment.
Last updated 2026-04-25 08:10:18; freshness should be verified.
Provenance
Source
Linguistic Data Consortium (LDC) and National Institute of Standards and Technology (NIST).
Collection Method
Segments were drawn from three LDC datasets: Speech Archive of South African Languages (SASAL), Maghrebi Linguistic Information Corpus (MAGLIC), and Low Resource African Languages (LRAL) collection, following specific collection protocols.
Time Range
Data is associated with the 2022 evaluation; specific recording dates are not provided.
Freshness
2026-04-25
Geography
Primarily South Africa (for SASAL languages), the Maghreb region (for MAGLIC languages), and Ethiopia/Eritrea (for LRAL languages).
License information is unknown and should be verified before use.