The project focuses on Speech-to-text Recognition of Urdu audio recipes followed by identification of food entities such as ingredient name, quantity, unit from the generated audio transcriptions. This project was developed as a part of the course Natural Language Processing.
All the source code is maintained in the code folder.
The Speech2text-models.ipynb file generates the audio transcriptions from the two publicly available fine-tuned Urdu models on Hugging Face and saves the output in Speech2Textpredictions.csv.
The Urdu_RecipeNER_Model.ipynb file trains and tests the NER model on Urdu Recipes.
The Combined.ipynb runs the Urdu_RecipeNER_Model on the transcriptions saved in Speech2Textpredictions.csv.