OpenSLR Speech Corpus
OpenSLR offers a high quality audio corpus for speech recognition research. The SLR32 game covers 4 South African languages with multi-speaker recordings and associated transcripts.
Transcribed `.wav` files + TSV tables; multi-speakers; 4 languages (af, st, tn, xh)
Creative Commons Attribution-ShareAlike 4.0
Description
OpenSLR Speech Corpus is a multilingual audio dataset designed for speech recognition and synthesis tasks. It contains high quality recordings in four African languages (Afrikaans, Sesotho, Setswana, Xhosa), provided with accurate transcripts in TSV format.
What is this dataset for?
- Train speech recognition models for low-resource languages.
- Develop multilingual TTS (speech synthesis) systems.
- Test and validate multilingual audio processing pipelines.
Can it be enriched or improved?
Yes. The corpus can be completed by phonetic annotations, finer temporal segmentation, or new languages. Text normalization models or more accurate audio/text alignments can also be applied.
🔎 In summary
🧠 Recommended for
- ASR researchers
- TTS developers
- Computer linguists
🔧 Compatible tools
- ESPnet
- Whisper
- Kaldi
- Hugging Face Transformers
💡 Tip
Segment the audios according to punctuation for finer alignment when training.
Frequently Asked Questions
What languages are covered by this corpus?
It covers four South African languages: Afrikaans, Sesotho, Setswana, and Xhosa.
Is this dataset suitable for beginners in audio processing?
Yes, its structured format and the availability of transcripts make it accessible to beginners, especially for simple ASR projects or TTS demonstrations.
Can this dataset be used for business models?
Yes, the CC BY-SA 4.0 license allows commercial use as long as the sharing conditions under the same terms are respected (mention of the author + sharing under an equivalent license).
