YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
# Moroccan Darija ASR Demo
## Overview
This project focuses on Automatic Speech Recognition (ASR) for Moroccan Darija.
We fine-tuned a pretrained ASR model on the DODa dataset using the Arabic transcription target darija\_Arab\_new.
## Main Components
- Fine-tuned Moroccan Darija ASR model
- Local inference script
- Local Gradio demo app
- Evaluation with WER and CER
- Comparison between base and fine-tuned model
## Model
Base model:
- boumehdi/wav2vec2-large-xlsr-moroccan-darija
Fine-tuned model:
- boumehdi\_darija\_arab\_full\_run1
## Main Results
- Base model WER: **27.27%**
- Fine-tuned model WER: **6.06%**
- Base model CER: **5.33%**
- Fine-tuned model CER: **1.78%**
## Project Structure
asr-darija-local/
├── app/
│ └── app.py
├── models/
│ └── boumehdi\_darija\_arab\_full\_run1/
├── sample\_audio/
│ └── test.wav
├── notebooks/
├── README.md
├── requirements.txt
├── test\_inference.py
└── .gitignore