Lwazi Afrikaans ASR Corpus: Telephone Speech and Transcriptions

Lwazi Afrikaans ASR corpus provides matched audio recordings and orthographic transcriptions designed for speech recognition systems. Audio files are telephone-quality, recorded at 8 KHz, 16-bit, and single-channel, with each utterance stored in a separate text file. This dataset was created to support the development of Automatic Speech Recognition (ASR) for the Afrikaans language.

Use Cases

• Training acoustic models for Afrikaans ASR based on telephone-quality audio.
• Benchmarking speech recognition systems using orthographic transcriptions.
• Studying phonetic or linguistic patterns in South African Afrikaans speech.
• Developing or fine-tuning language models for Afrikaans from transcribed utterances.

Strengths

• Includes complete, matched pairs of audio and text transcriptions for each utterance.
• Audio specifications (8 KHz, 16-bit, 1 channel) are explicitly documented for telephone speech.
• Released under a permissive CC-BY-4.0 license, facilitating open research and reuse.

Limitations

• Key metadata such as row count, dataset size, and specific column structure is unavailable.
• The listed last update date of 2026-03 24 appears to be a future date, indicating a potential metadata error.
• Documentation is sparse, lacking details on speaker demographics, recording conditions, or corpus size.

Provenance

Source

Charl van Heerden

Collection Method

Created for the Lwazi speech recognition systems; specific collection method not detailed.

Time Range

null

Freshness

2026-03-24 12:39:11

Geography

South Africa