Ishita95-AIML/Kinyarwanda-Automatic-Speech-Recognition-Track-A-An-ASR-
Domaine:
natural language processingType de record:
dataset
Créateur:
Ish
Hôte:
# Kinyarwanda Automatic Speech Recognition Track A
An ASR hackathon on 500-hour Kinyarwanda speech dataset covering Health, Government, Finance, Education, and Agriculture domains.
Here's a README file for your project, Kinyarwanda Automatic Speech Recognition Track A, outlining its purpose, setup, and usage.
## Overview
This project is part of an **ASR Hackathon** focused on building and improving **Automatic Speech Recognition (ASR) models** for Kinyarwanda. The dataset consists of **500 hours** of Kinyarwanda speech across various domains, including **Health, Government, Finance, Education, and Agriculture**. The goal is to develop models that enhance speech-to-text accuracy for Kinyarwanda speakers.
## Features
- Large-scale **500-hour** dataset for training ASR models.
- Covers multiple **domains** to ensure broad applicability.
- Supports **fine-tuning and benchmarking** ASR models.
- Open-source approach to improving **Kinyarwanda speech recognition**.
## Installation
To set up the environment, clone the repository and install dependencies:
```sh
git clone github.com
cd Kinyarwanda-ASR-TrackA
pip install -r requirements.txt
Dataset
The dataset contains spoken Kinyarwanda sentences collected from various sources. It is structured as follows:
- Health – Medical conversations, health education.
- Government – Speeches, government notices, public communication.
- Finance – Banking discussions, financial reports.
- Education – Classroom lectures, educational materials.
- Agriculture – Farming practices, agricultural advice.
Model Training
To train an ASR model on this dataset, use the following command:
python train.py --data_path data/kinyarwanda/ --epochs 50 --batch_size 32
Evaluation
To evaluate the trained ASR model:
python evaluate.py --model_path models/asr_model.pth --test_data data/kinyarwanda/test/
Technologies Used
- Python
- DeepSpeech / Wav2Vec 2.0
- Pandas & NumPy
- Torch / TensorFlow
- S …