ghananlpcommunity/nsanku-asr-benchmark

nsanku ASR Benchmark

Benchmarking ASR models on Ghanaian languages using the ghana-speech-eval dataset.

Three tracks: ASR (open models) (downloadable weights anyone can re-run), ASR (closed models) (proprietary hosted APIs) and LLM. Scored per category (Bible / JW / Finance / Unicef / LDS / WAXAL), averaged as the final WER/CER. Ranking uses CER — every model row links to a recipe (code badge) showing the exact inference code used to run it, so model authors can correct how their model is evaluated via a pull request.

Data source: GhanaNLP/nsanku-asr-benchmark