This project aims to fine-tune OpenAI's Whisper model to better understand and transcribe Singaporean English. We utilize the National Speech Corpus provided by IMDA for our dataset.
The preprocessing stage involves downloading files from a Dropbox link specified in the code. This link will vary depending on the specific section of the dataset being used. To initiate the preprocessing, run the main file under the preprocessing-pipeline
directory. The details of this process can be viewed in the DataProcessingPipeline.ipynb
notebook in the repository.
After preprocessing, we move on to the training pipeline. The training process is conducted on a fraction of the dataset for demonstration purposes. The details of this process can be viewed in the Training_pipeline.ipynb
notebook in the repository.
The evaluations of the model and detailed training metrics can be found on our Hugging Face repository: Hugging Face Repository - whisperfinetune_modelcheckpoints. This includes performance metrics, model checkpoints, and other relevant evaluation data.
- Ensure you have Python installed on your system.
- The code is designed to run with Google Drive mounted on macOS. If you're using a different operating system, modifications might be necessary.
Before running the project, you'll need to install the necessary dependencies.
This project requires certain Python packages to be installed. These dependencies are listed in the requirements.txt
file. To install them, follow these steps:
-
Clone the Repository: If you haven't already, clone the repository to your local machine:
git clone https://github.com/MomPansy/WhisperFineTune.git
-
Navigate to the Repository Directory: Change into the repository directory:
cd WhisperFineTune
-
Create a Virtual Environment (Optional but Recommended): It's a good practice to use a virtual environment for your Python projects. Create one using:
python -m venv venv
Activate it with:
- On Windows:
.\venv\Scripts\activate
- On macOS and Linux:
source venv/bin/activate
- On Windows:
-
Install the Dependencies: Install all required packages with the following command:
pip install -r requirements.txt
Now, your environment should be set up with all the dependencies needed for the project.
-
Clone the repository to your local machine.
-
Navigate to the repository's preprocessing-pipeline directory in the command line.
-
Run the main script with your access token:
python main.py 'YOUR_ACCESS_TOKEN'
Replace
'YOUR_ACCESS_TOKEN'
with your actual access token for the Dropbox API.
The current implementation is tailored for macOS with Google Drive integration. If you are using a different operating system or cloud storage solution, you will need to adjust the file paths and storage configurations accordingly.
For a practical demonstration of the fine-tuned Whisper model in action, you can view and interact with our transcription app. This web-based application allows users to experience the capabilities of the model with Singaporean English.
The transcription app is hosted on Hugging Face Spaces. You can access it at the following URL:
WhisperFineTune Transcription App
Feel free to test the app with your own audio samples to see how effectively the model transcribes Singaporean English.