Automatic Speech Recognition
Transformers
TensorBoard
Safetensors
English
lasr_ctc
asr
speech-to-text
medical-asr
african-accents
ghanaian-english
twi
akan
fante
ctc
Instructions to use samwell/medasr-ghana with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use samwell/medasr-ghana with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("automatic-speech-recognition", model="samwell/medasr-ghana")# pip install -U transformers accelerate # Load model directly from transformers import AutoModelForCTC model = AutoModelForCTC.from_pretrained("samwell/medasr-ghana", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Commit History
-
120 epochs, LR=3e-5 | Test WER: 0.3753
Verified signature98a34f8 -
80 epochs, LR=3e-5, no augmentation | Test WER: 0.3859
Verified signatureca9b828 -
Simple approach: 40 epochs, LR=3e-5, no augmentation | Test WER: 0.4021
Verified signature315f005 -
Balanced: light SpecAugment, 12 epochs, LR=2e-5 | Test WER: 0.5928
Verified signature8bde912 -
Fine-tuned with SpecAugment (20 epochs, LR=1e-5) | Test WER: 0.6119
Verified signature5a287c4 -
Fine-tuned on ALL Ghanaian accents (twi+akan+fante) | Test WER: 0.5526
Verified signature2220432 -
Fine-tuned on Ghanaian accents (twi) | Test WER: 0.5919
Verified signature4323dd9 -
Fine-tuned on Ghanaian accents | Test WER: 0.6178
Verified signature55283f6