abeladamushumet/Amharic-ASR-Summarization

Amharic Speech-to-Text system powered by Wav2Vec 2.0 and Hugging Face Transformers, fine-tuned on the Amharic ReadSpeech dataset to achieve high transcription accuracy for low-resource language processing. # 🎙️ Amharic Speech-to-Text Transcription with Wav2Vec 2.0 ## Project Overview This repository hosts the code and resources for an **Amharic Automatic Speech Recognition (ASR)** model built using the advanced **Wav2Vec 2.0** architecture. The model is fine-tuned to accurately transcribe **Amharic speech** into text, leveraging **Hugging Face Transformers** and modern deep learning techniques. This project contributes to the development of **language technology for under-resourced African languages**, offering a robust, open-source solution for Amharic speech-to-text conversion. --- ## Model Performance The fine-tuned **Wav2Vec 2.0** model (available on Hugging Face (agkphysics/wav2vec2-large-x…)) achieves strong performance on the **Amharic ReadSpeech dataset**, with the following metrics: - **Word Error Rate (WER):** 13% - **Character Error Rate (CER):** 5.6% These results demonstrate high transcription accuracy and the effectiveness of Wav2Vec 2.0 for low-resource Amharic ASR tasks. --- ## Repository Structure ``` . ├── lang/ # Language-specific configs (lexicon, phoneme mappings) ├── train/ # Training datasets for model development ├── test/ # Test datasets for evaluation ├── gen_vocab.py # Script to generate vocabulary from training data ├── train.py # Main training and evaluation script ├── vocab.json # Generated vocabulary file ├── requirements.txt # Python dependencies └── README.md # Project documentation (this file) ```` --- ## Installation Set up the environment and install dependencies: 1. **Clone the repository** ```bash git clone github.com cd Amharic-ASR-Summarization ```` 2. **Create and activate a virtual environment** ```bash python3 -m venv venv venv\Scripts\activate ``` 3. **Install dependencies** ```bash pip install -r requirements.txt ``` --- ## Data Pr …