Automatic Speech Recognition
Transformers
Safetensors
Ikposo
whisper
african-languages
waxal
waxalnet
Instructions to use waxal-benchmarking/whisper-small-waxal-kpo with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use waxal-benchmarking/whisper-small-waxal-kpo with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("automatic-speech-recognition", model="waxal-benchmarking/whisper-small-waxal-kpo")# Load model directly from transformers import AutoProcessor, AutoModelForSpeechSeq2Seq processor = AutoProcessor.from_pretrained("waxal-benchmarking/whisper-small-waxal-kpo") model = AutoModelForSpeechSeq2Seq.from_pretrained("waxal-benchmarking/whisper-small-waxal-kpo", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Add comprehensive model card (WAXALNet benchmark)
Browse files
README.md
CHANGED
|
@@ -61,7 +61,7 @@ Evaluated on the filtered WAXAL test set (duration >= 1.5s, speech rate >= 4 WPS
|
|
| 61 |
| **WER** | 77.5% |
|
| 62 |
| **CER** | 26.5% |
|
| 63 |
|
| 64 |
-
Full benchmark results across all 19 languages and 6 models are reported in the WAXAL ASR Benchmark paper (citation below).
|
| 65 |
|
| 66 |
## Training procedure
|
| 67 |
|
|
@@ -132,3 +132,38 @@ Ajara Oyinloye, Abubakari Sadic Mohammed, Hafiz Adjei, Aliga Norah Lele, Marie-L
|
|
| 132 |
This work was supported by **[Lynguallabs](https://lynguallabs.org/)** (compute, researchers & storage),
|
| 133 |
**[Open Token](https://opentoken.global/)** (compute resources), and
|
| 134 |
**[CMU Africa](https://www.africa.engineering.cmu.edu/)** (researchers & native speakers).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
| **WER** | 77.5% |
|
| 62 |
| **CER** | 26.5% |
|
| 63 |
|
| 64 |
+
Full benchmark results across all 19 languages and 6 models are reported in the [WAXAL ASR Benchmark paper](https://arxiv.org/abs/2606.02375) (citation below).
|
| 65 |
|
| 66 |
## Training procedure
|
| 67 |
|
|
|
|
| 132 |
This work was supported by **[Lynguallabs](https://lynguallabs.org/)** (compute, researchers & storage),
|
| 133 |
**[Open Token](https://opentoken.global/)** (compute resources), and
|
| 134 |
**[CMU Africa](https://www.africa.engineering.cmu.edu/)** (researchers & native speakers).
|
| 135 |
+
|
| 136 |
+
|
| 137 |
+
## Citation
|
| 138 |
+
|
| 139 |
+
```bibtex
|
| 140 |
+
@article{waxalnet2026,
|
| 141 |
+
title = {The WAXAL ASR Benchmark: Fine-Tuned Edge Models Across 19 African Languages},
|
| 142 |
+
author = {Olufemi, Victor Tolulope and Babatunde, Oreoluwa and Njema, Ramsey and
|
| 143 |
+
Gbotemi, Bolarinwa and Yen, Wanchi Lucia and Uzodinma, John and
|
| 144 |
+
Ajayi, Sunday and Williams, Oluwademilade and Moshood, Kausar and
|
| 145 |
+
Anyaele, Innocent Elendu and Arefaine, Akebert Tesfahunegn and
|
| 146 |
+
Hunzwi, Candace and Daniel, Wongel Dawit and Namuganga, Emmilly Immaculate and
|
| 147 |
+
Kadima, Cleophas and Bahizire, Athanase Biluge and Ranaivoson, Onitsiky and
|
| 148 |
+
Aaron, Emmanuel and Ladislaus, Nicholaus Dismas and Muhammed, Idris and
|
| 149 |
+
Simenya, Jonathan Enoch and Koome, Martin and Endaylalu, Matewos Tegete and
|
| 150 |
+
Adeyemo, Peter Ifeoluwa and Birindwa, Hondi Prisca and Eze-Mbey, Ukachi Agnes and
|
| 151 |
+
Oduro-Yeboah, Yacoba and Aremu, Toluwani and Adjovi, Pericles and
|
| 152 |
+
Ngueajio, Mikel K and Mitra, Prasenjit},
|
| 153 |
+
year = {2026},
|
| 154 |
+
note = {arXiv preprint arXiv:2606.02375}
|
| 155 |
+
}
|
| 156 |
+
```
|
| 157 |
+
|
| 158 |
+
## Authors
|
| 159 |
+
|
| 160 |
+
Victor Tolulope Olufemi 路 Oreoluwa Babatunde 路 Ramsey Njema 路 Bolarinwa Gbotemi 路 Wanchi Lucia Yen 路 John Uzodinma 路 Sunday Ajayi 路 Oluwademilade Williams 路 Kausar Moshood 路 Innocent Elendu Anyaele 路 Akebert Tesfahunegn Arefaine 路 Candace Hunzwi 路 Wongel Dawit Daniel 路 Emmilly Immaculate Namuganga 路 Cleophas Kadima 路 Athanase Biluge Bahizire 路 Onitsiky Ranaivoson 路 Emmanuel Aaron 路 Nicholaus Dismas Ladislaus 路 Idris Muhammed 路 Jonathan Enoch Simenya 路 Martin Koome 路 Matewos Tegete Endaylalu 路 Peter Ifeoluwa Adeyemo 路 Hondi Prisca Birindwa 路 Ukachi Agnes Eze-Mbey 路 Yacoba Oduro-Yeboah 路 Toluwani Aremu 路 Pericles Adjovi 路 Mikel K Ngueajio 路 Prasenjit Mitra
|
| 161 |
+
|
| 162 |
+
## Acknowledgements
|
| 163 |
+
|
| 164 |
+
We thank the following contributors for their language expertise and native-speaker evaluation support:
|
| 165 |
+
Ajara Oyinloye, Abubakari Sadic Mohammed, Hafiz Adjei, Aliga Norah Lele, Marie-Louise B. Ndamuso, and Odong Diana.
|
| 166 |
+
|
| 167 |
+
This work was supported by **[Lynguallabs](https://lynguallabs.org/)** (compute, researchers & storage),
|
| 168 |
+
**[Open Token](https://opentoken.global/)** (compute resources), and
|
| 169 |
+
**[CMU Africa](https://www.africa.engineering.cmu.edu/)** (researchers & native speakers).
|