Whisper Small FR - Radiologie
This model is a fine-tuned version of leduckhai/MultiMed-ST/asr/whisper-small-french on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.0488
- Wer: 7.5491
Model Description
The model focuses on two main adaptations:
- Acoustic Adaptation: Capturing the phonetic nuances of French-speaking African regions to improve recognition of local accents.
- Medical Terminology: Stabilizing technical radiology terms (Spine, Shoulder, Thorax, Mammography, CT scans) in a dictation context.
It uses LoRA (Low-Rank Adaptation) via the adapters library, specifically targeting the first 4 layers of the Encoder (for acoustic/accent adaptation) and the full Decoder (for medical jargon and linguistic structure).
Training and Evaluation Data
- Training Dataset: ~4.5 hours of specialized radiology recordings (562 audios).
Intended uses & limitations
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 3e-05
- train_batch_size: 12
- eval_batch_size: 8
- seed: 3407
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- training_steps: 1000
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| No log | 3.0303 | 100 | 0.1033 | 23.0029 |
| No log | 6.0606 | 200 | 0.0697 | 14.2200 |
| No log | 9.0909 | 300 | 0.0556 | 14.6173 |
| No log | 12.1212 | 400 | 0.0482 | 8.4065 |
| 0.0838 | 15.1515 | 500 | 0.0479 | 8.4483 |
| 0.0838 | 18.1818 | 600 | 0.0483 | 8.9502 |
| 0.0838 | 21.2121 | 700 | 0.0484 | 8.6784 |
| 0.0838 | 24.2424 | 800 | 0.0483 | 7.6328 |
| 0.0838 | 27.2727 | 900 | 0.0485 | 8.8666 |
| 0.0001 | 30.3030 | 1000 | 0.0488 | 7.5491 |
Performance on the Test Set
The model was evaluated on the AfroRadVoice-FR test split (75 audio files, independent of training), using identical decoding settings (temperature = 0.0) across all models for a fair comparison.
| Rank | Model | WER (%) | CER (%) | Sentence Accuracy (%) |
|---|---|---|---|---|
| 1 | Whisper-AfroRad-FR | 20.93 | 16.80 | 34.67 |
| 2 | Med-Whisper-AfroRad-FR (this model) | 21.84 | 17.68 | 29.33 |
| 3 | whisper-small-rad-FR | 25.12 | 20.89 | 33.33 |
| 4 | nvidia/canary-1b-v2 | 33.96 | 11.10 | 1.33 |
| 5 | Qwen/Qwen3-ASR-0.6B | 45.40 | 17.55 | 0.00 |
| 6 | bofenghuang/whisper-small-cv11-french | 75.11 | 53.65 | 0.00 |
| 7 | openai/whisper-small (baseline) | 79.12 | 54.47 | 0.00 |
| 8 | openai/whisper-large-v3 | 120.41 | 84.02 | 0.00 |
Framework versions
- Transformers 4.51.3
- Pytorch 2.8.0+cu126
- Datasets 4.4.2
- Tokenizers 0.21.4
Citation
If you use this model in your research, please cite:
@misc{med-whisper-afrorad-fr,
author = {StephaneBah},
title = {Med-Whisper-AfroRad-FR: Medical Radiology ASR for Afro-French Context},
year = {2026},
publisher = {Hugging Face},
howpublished = {\\url{https://huggingface.co/StephaneBah/Med-Whisper-AfroRad-FR}}