Projects >
WAXALNET
Overview
WAXALNet is a large-scale African ASR benchmark consisting of 57 open-source models fine-tuned across 19 African languages, demonstrating a 26.9pp WER reduction compared to foundation models.
The WAXAL ASR Benchmark evaluates zero-shot foundation models against compact, fine-tuned edge models across 19 African languages using the conversational WAXAL corpus. Fine-tuned edge models achieve a macro-averaged WER of 38.0% compared to 64.9% for the best zero-shot baseline — a 26.9pp reduction using models 3–40× smaller. This demonstrates that compact, domain-specialized models are highly efficient for low-resource environments.