Veda-Guru is a Sanskrit language model specifically designed for the Vedas, including the Rigveda, Samaveda, Yajurveda, and Atharvaveda. The project aims to provide a comprehensive tool for understanding and analyzing Vedic texts, with a focus on audio mode training and handling special symbols in Vedic sentences.
The Veda-Guru project involves the following key steps:
- Text Collection and Preprocessing: Downloading and preprocessing texts from the Vedas.
- Model Training: Fine-tuning a pre-trained BERT model on the Vedic texts.
- Evaluation and Fine-tuning: Assessing the model's performance and making necessary adjustments.
- API Development: Creating an API for interacting with the trained model.
- Audio Mode Training: Incorporating techniques for decoding audio from different parts of the throat and handling special symbols in Vedic sentences.
To set up the project, follow these steps:
-
Clone the repository:
git clone https://github.com/kasinadhsarma/Veda-Guru.git cd Veda-Guru -
Install the required dependencies:
pip install -r requirements.txt
The Veda-Guru API provides the following endpoints:
POST /predict: Receives an audio file, processes it, and returns a prediction.
To use the API for audio predictions, send a POST request with a .wav audio file to the /predict endpoint. The API will return the predicted label.
Example:
curl -X POST -F "audio_file=@./audio/vedic_chanting_Vedic Chanting ｜ Rudri Path by 21 Brahmins.wav" http://localhost:5000/predictThe API includes error handling for common issues that users might encounter. Here are some examples:
- Invalid File Format: If the uploaded file is not a
.wavfile, the API will return a400 Bad Requesterror with a message indicating the invalid file format. - Missing File: If no file is uploaded, the API will return a
400 Bad Requesterror with a message indicating that the file is missing. - Internal Server Error: If there is an issue processing the file or making a prediction, the API will return a
500 Internal Server Errorwith a message indicating the problem.
The Veda-Guru model is designed to handle audio files of Vedic chanting and make predictions based on the trained data. However, there are some limitations to be aware of:
- Special Symbols: The model is trained to handle special symbols in Vedic sentences, but its performance may vary depending on the quality and clarity of the audio.
- Bias: Efforts have been made to ensure the model is not biased, but users should be aware that the training data's diversity can impact the model's predictions.
- Audio Quality: The model performs best with high-quality audio recordings. Poor audio quality may affect the accuracy of the predictions.
To train the model, run the following script:
python train_sanskrit_model.pyTo preprocess the Vedic texts, run the following script:
python preprocess_rigveda.pyTo download the Rigveda hymns, run the following script:
python download_rigveda.pyTo train the audio model, run the following script:
python audio_mode_training.pyThe Veda-Guru model is designed to handle special symbols in Vedic sentences. During preprocessing, special symbols are retained and appropriately tokenized to ensure the model can accurately interpret and process them.
Efforts have been made to ensure the Veda-Guru model is not biased. The training data includes a diverse range of examples to minimize bias. Additionally, the model's predictions are regularly evaluated to identify and address any potential biases.
The trained model can be interacted with through the provided API. The POST /predict endpoint allows users to send audio files for prediction. The model is saved in the Keras format (.keras extension) and can be loaded using the following code snippet:
from tensorflow.keras.models import load_model
model = load_model('fine_tuned_model/model.keras')Contributions are welcome! Please open an issue or submit a pull request for any improvements or bug fixes.
This project is licensed under the MIT License.