krbl-lmm/amharic-dialect-classifier

An Amharic dialect classifier that differentiates between five dialects of the language. # Amharic Dialect Classifier ## Overview This project identifies the dialect of spoken Amharic from an audio recording. The system extracts MFCC-based acoustic features from speech and uses a Random Forest classifier to predict one of five dialects: - Addis Ababa - Gojjam - Gonder - Shewa - Wello A Streamlit web application allows users to upload audio files and view predictions with confidence scores. ## Features * Audio-only dialect classification * MFCC-based feature extraction * Random Forest classifier * Confidence score prediction * Interactive Streamlit demo * Reproducible training pipeline ## Dataset This project uses the Leyu Amharic Dialect datasets hosted on Hugging Face. | Dialect | Dataset | |----------|----------| | Addis Ababa | huggingface.co | | Gojjam | Leyu Amharic Gojjam Dialect | | Gonder | huggingface.co | | Shewa | Leyu Amharic Shewa Dialect | | Wello | Leyu Amharic Wello Dialect | The datasets contain speech recordings and metadata including dialect, speaker ID, gender, and transcripts. Only the audio recordings and dialect labels were used during training. ### Attribution The Leyu Amharic Dialect datasets are provided by the Leyu Amharic project and are licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. If you use this project or the underlying datasets in research or derivative work, please provide appropriate attribution to the original dataset creators. License: creativecommons.org ## Project Structure ```text amharic-dialect-classifier/ ├── data/ # ignored ├── images/ │ └── demo.png │ ├── models/ │ └── dialect_model.pkl │ ├── src/ │ ├── load_data.py │ ├── feature_extraction.py │ ├── …