Project Overview
This repository contains a collection of tools, experiments, and utilities for building and fine-tuning automatic speech recognition (ASR) systems—particularly for the Bambara language—and for experimenting with reinforcement learning from human feedback (RLHF) techniques for ASR training.
License
This repository is licensed under the MIT License. See LICENSE for details.
-
Clone the repository:
git clone https://github.com/RobotsMali-AI/bambara-asr.git cd bambara-asr -
Install requirements (for RLNF toolkit & core dependencies):
Before you try to experiment with Reinforcement Learning from Nouhoum Feedback
pip install -r rlnf/requirements.txt
pip install -e rlnfNote that the second line will install the package in edit mode, remove -e option to create a permanent install
-
Explore sub‑folders:
Each subfolder in this repo holds code and config for different experiments that we have done. Each one also possesses a dedicated README for further details and instructions. Looking for details on Listen, Attend, Understand?
RLNF is still experimental and contributions are welcome! Please open an issue or submit a pull request for any bug fix or feature request. For questions or support, file an issue in this repo.
README files for individual modules are provided in their respective folders. Start with rlnf/README.md for detailed instructions on the RLHF toolkit.