WAXAL-NET: Finetuned Edge ASR Across 19 African Languages
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866918535256604672 |
|---|---|
| author | Olufemi, Victor Tolulope Babatunde, Oreoluwa Njema, Ramsey Gbotemi, Bolarinwa Yen, Wanchi Lucia Uzodinma, John Ajayi, Sunday Williams, Oluwademilade Moshood, Kausar Anyaele, Innocent Elendu Arefaine, Akebert Hunzwi, Candace Daniel, Wongel Dawit Namuganga, Emmilly Kadima, Cleophas Bahizire, Athanase Ranaivoson, Onitsiky Aaron, Emmanuel Ladislaus, Nicholaus Muhammed, Idris Simenya, Jonathan Enoch Koome, Martin Endaylalu, Matewos Tegete Adeyemo, Peter Ifeoluwa Birindwa, Hondi Prisca Eze-Mbey, Ukachi Agnes Oduro-Yeboah, Yacoba Adjovi, Pericles Ngueajio, Mikel K. Aremu, Toluwani Mitra, Prasenjit |
| author_facet | Olufemi, Victor Tolulope Babatunde, Oreoluwa Njema, Ramsey Gbotemi, Bolarinwa Yen, Wanchi Lucia Uzodinma, John Ajayi, Sunday Williams, Oluwademilade Moshood, Kausar Anyaele, Innocent Elendu Arefaine, Akebert Hunzwi, Candace Daniel, Wongel Dawit Namuganga, Emmilly Kadima, Cleophas Bahizire, Athanase Ranaivoson, Onitsiky Aaron, Emmanuel Ladislaus, Nicholaus Muhammed, Idris Simenya, Jonathan Enoch Koome, Martin Endaylalu, Matewos Tegete Adeyemo, Peter Ifeoluwa Birindwa, Hondi Prisca Eze-Mbey, Ukachi Agnes Oduro-Yeboah, Yacoba Adjovi, Pericles Ngueajio, Mikel K. Aremu, Toluwani Mitra, Prasenjit |
| contents | We evaluate whether compact domain-specialized ASR models can outperform massively multilingual foundation models for conversational African speech across 19 languages in the 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 $26.9$ percentage-point reduction using models $3-40\times$ smaller. Results confirm that domain specialization dominates scale for spontaneous African speech. Cross-domain evaluation shows that fine-tuned models recover usable performance on out-of-distribution (OOD) speech, while zero-shot models regain an advantage when the test domain matches their pretraining distribution. A distributed native-speaker audit across all surveyed languages produces a linguistically-grounded error taxonomy, showing that CTC and autoregressive architectures behave differently across language families. We further show that WER alone misrepresents performance for syllabary-script languages where CER/WER ratios reveal substantially higher character-level accuracy than headline WER suggests. Finally, to contribute to future African ASR research, we release all model weights, fine-tuning and evaluation scripts, and a cleaned WAXAL subset covering all $19$ languages. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2606_02375 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | WAXAL-NET: Finetuned Edge ASR Across 19 African Languages Olufemi, Victor Tolulope Babatunde, Oreoluwa Njema, Ramsey Gbotemi, Bolarinwa Yen, Wanchi Lucia Uzodinma, John Ajayi, Sunday Williams, Oluwademilade Moshood, Kausar Anyaele, Innocent Elendu Arefaine, Akebert Hunzwi, Candace Daniel, Wongel Dawit Namuganga, Emmilly Kadima, Cleophas Bahizire, Athanase Ranaivoson, Onitsiky Aaron, Emmanuel Ladislaus, Nicholaus Muhammed, Idris Simenya, Jonathan Enoch Koome, Martin Endaylalu, Matewos Tegete Adeyemo, Peter Ifeoluwa Birindwa, Hondi Prisca Eze-Mbey, Ukachi Agnes Oduro-Yeboah, Yacoba Adjovi, Pericles Ngueajio, Mikel K. Aremu, Toluwani Mitra, Prasenjit Computation and Language Computers and Society Human-Computer Interaction We evaluate whether compact domain-specialized ASR models can outperform massively multilingual foundation models for conversational African speech across 19 languages in the 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 $26.9$ percentage-point reduction using models $3-40\times$ smaller. Results confirm that domain specialization dominates scale for spontaneous African speech. Cross-domain evaluation shows that fine-tuned models recover usable performance on out-of-distribution (OOD) speech, while zero-shot models regain an advantage when the test domain matches their pretraining distribution. A distributed native-speaker audit across all surveyed languages produces a linguistically-grounded error taxonomy, showing that CTC and autoregressive architectures behave differently across language families. We further show that WER alone misrepresents performance for syllabary-script languages where CER/WER ratios reveal substantially higher character-level accuracy than headline WER suggests. Finally, to contribute to future African ASR research, we release all model weights, fine-tuning and evaluation scripts, and a cleaned WAXAL subset covering all $19$ languages. |
| title | WAXAL-NET: Finetuned Edge ASR Across 19 African Languages |
| topic | Computation and Language Computers and Society Human-Computer Interaction |
| url | https://arxiv.org/abs/2606.02375 |