WAXALNet - a waxal-benchmarking Collection

waxal-benchmarking 's Collections

WAXALNet - Research ablations

WAXALNet - Multilingual (family & all-19)

WAXALNet - Omnilingual LLM-300M

WAXALNet - Omnilingual CTC-300M

WAXALNet — Whisper Tiny

WAXALNet — Whisper Small

WAXALNet — MMS-300M

WAXALNet

updated 1 day ago

WaxalNet: ASR benchmark and fine-tuned models for 19 African languages (WAXAL).

google/WaxalNLP

Viewer • Updated Sep 1 • 3.38M • 19.9k • 303

Note Training and test data (spontaneous speech, 19 languages).
WAXALNet - Omnilingual LLM-300M

Collection

Omnilingual ASR (wav2vec2-Llama decoder, 300M) fine-tuned per WAXAL language. • 19 items • Updated Sep 8
WAXALNet - Omnilingual CTC-300M

Collection

Omnilingual ASR (wav2vec2 CTC head, 300M) fine-tuned per WAXAL language. • 19 items • Updated Sep 8
WAXALNet — Whisper Tiny

Collection

Whisper Tiny fine-tuned on the WAXAL corpus across 19 African languages. Part of the WAXAL ASR Benchmark (EMNLP 2025). • 19 items • Updated May 29
WAXALNet — Whisper Small

Collection

Whisper Small fine-tuned on the WAXAL corpus across 19 African languages. Part of the WAXAL ASR Benchmark (EMNLP 2025). • 19 items • Updated May 29
WAXALNet — MMS-300M

Collection

MMS-300M fine-tuned on the WAXAL corpus across 19 African languages. Part of the WAXAL ASR Benchmark (EMNLP 2025). • 19 items • Updated May 29
WAXALNet - Multilingual (family & all-19)

Collection

Omnilingual LLM-300M fine-tuned on pooled WAXAL languages: one model per language family and one for all 19. • 7 items • Updated 23 days ago
WAXALNet - Research ablations

Collection

Archival models from the WaxalNet ablations (leave-one-language-out, matched-recipe). Not intended for use. • 13 items • Updated 23 days ago
waxal-benchmarking/waxalnet-asr-results

Updated 1 day ago • 75

Note Per-utterance transcripts and analysis behind every number in the paper.