Harcuracy/

google_waxal_asr_challenge

This dataset is a cleaned, corrected subset of google/WaxalNLP, covering the train and validation splits for three languages:

lin_asr — Lingala sna_asr — Shona lug_asr — Luganda

The test split from the original dataset is intentionally excluded.

The original transcriptions for these three languages contained a number of errors. A corrected transcription file was applied on top of the… See the full description on the dataset page: https://huggingface.co/datasets/Harcuracy/google_waxal_asr_challenge.