Sesotho Speech Corpus with Orthographic Transcriptions

A corpus of orthographically transcribed broadband speech for Sesotho, one of South Africa's eleven official languages. It was created by researcher Febe de Wet and the NCHLT project, with transcriptions provided in XML format. The dataset was last updated in March 2026.

Use Cases

• Train automatic speech recognition models using the orthographic transcriptions paired with the audio recordings.
• Analyze phonetic and phonological patterns in Sesotho speech from the transcribed corpus.
• Benchmark speech synthesis systems by using the transcribed text and corresponding audio as a training or evaluation set.
• Develop language models for Sesotho using the textual content extracted from the XML transcriptions.

Strengths

• Covers one of South Africa's eleven official languages, providing a resource for a low-resource language.
• Includes orthographic transcriptions in XML format, enabling alignment with audio data.
• Released under a permissive CC-BY-4.0 license for open use.

Limitations

• The specific size (row count, audio hours) and temporal coverage of the corpus are unknown.
• Lacks detailed column or feature descriptions, making precise data exploration difficult.
• The 0.0 B size indicator suggests a very limited scope or potential metadata-only record.

Provenance

Source

Febe de Wet, NCHLT project.

Collection Method

null

Time Range

null

Freshness

null

Geography

South Africa.