This project explores personalized music creation using advanced ML, aiming for music aligned with preferences. It evaluates models like Jukebox, MuseNet, and Music Transformer for musical nuances, cohesion, and emotional depth, with thorough measures and expert feedback for iterative enhancement. It envisions transforming music engagement into an interactive, co-creative process, leveraging ML and user-centric design to redefine music's role in daily life.

Firstly train the model

• Each notebook is named in parts (1-5)
• Final outputs is the trained model with weights and biases configured

For the server side

• open flask-server folder
• copy paste the Final outputs folder into the flask-server folder
• run docker commands docker build -t okbenzene/python-api:latest .
• run the container docekr container run -d -p 4000:3000 okbenzene/python-api:latest

For the client side

• open the react-app folder
• open a new terminal ctrl+`
• run npm i
• type npm run musicgpt

This project is divided into three core sections.

• Jupyter Notebook - Model creation, Model Training, Model Checkpoints.
• Server - Creating Model API and deploying the model using Docker.
• Client - Creating a webpage to interact with the model (Experimental).

• Midi Chunking
• Midi Tokenization
• GPT-2 Model
• HuggingFace for saving Tokenizer, Dataset
• Wandb for logging the model training

• Larger training data
• Imporve tokenization process to diversify on variety of instruments
• GPT-2 Tweaks for better predictions (Hardware intensive)

• Model API made using flask
• Test script included
• Easier debugging
• Song generations (Parsed and saved as file .wav format)

• Code optimizations
• Minor issue with generations (Memory cosumption)
• error handling
• API for parsing piano notations directly

• Component based
• Latest Framework (React)
• Input based generation
• History Component for previously generated compositions

• Memory leak issues when POST route is called (FIXED - Still needs an improvement)
• About me page to be built
• In app Piano interaction for generating composition
• Email validations
• Visualizer improvements

• @Liyakhat Yousuf Mogal

Thanks to @juancopi81 for supporting