ngodoannghia/ASR

Speech to Text Project

This project aims to train a wav2vec model for speech-to-text tasks using both Vietnamese and English data. Here is the structure and setup of the project.

Project Structure

ASR/ ├── configs/ │ ├── config.yaml │ ├── data/ │ │ └── default.yaml │ ├── training/ │ │ └── default.yaml │ ├── wandb/ │ └── default.yaml ├── data/ │ ├── raw/ │ ├── processed/ │ ├── exploration/ ├── notebooks/ │ ├── data_exploration.ipynb │ ├── training_analysis.ipynb ├── scripts/ │ ├── data_preprocessing.py │ ├── train.py │ ├── utils.py ├── models/ │ ├── checkpoints/ │ ├── wav2vec_model.py ├── logs/ │ ├── wandb/ ├── environment.yml └── README.md

Setup

1. Create and activate the conda environment:

   conda env create -f environment.yml
   conda activate speech_to_text

2. Preprocess data:

   python scripts/data_preprocessing.py

3. Train the model:

   python scripts/train.py

4. Explore data and training results:

   • Use notebooks/data_exploration.ipynb for data exploration.
   • Use notebooks/training_analysis.ipynb for analyzing training results.

Configurations

Configurations are managed using Hydra. Modify the configs/config.yaml or create new configurations in the configs/ directory.

Logging

We use wandb for logging. Ensure you have set up your wandb account and updated the configs/wandb/default.yaml with your project name and entity.