Music Generation Using LSTM

Conv1D + LSTM model trained on Bach chorales, with a simple web UI to generate new sequences and export MIDI.

What's inside

• Music_Generation_Using_LSTM.ipynb — training and experimentation notebook
• bach_generation_conv1d_lstm.keras — trained model file
• Dataset/ — chorale CSVs used for training/validation
• ui/ — lightweight HTML/CSS/JS UI + Python inference server

Quick start (UI)

cd "ui"
python -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip

Install dependencies

Apple Silicon (M1/M2/M3):

python -m pip install numpy music21 tensorflow-macos tensorflow-metal

Intel Mac:

python -m pip install numpy music21 tensorflow

Run the server

python server.py

Open http://localhost:8000 live https://music-generation-using-lstm-43wx.onrender.com

UI features

• Manual seed chords (8x4)
• Random seed generator
• CSV seed upload
• Piano‑roll visualization
• MIDI download
• In‑browser audio playback

Model notes

• MIDI range is 36–81, with 0 as rest
• Generates note‑by‑note, then reshaped into 4‑note chords

Training

Open the notebook and run the cells in order:

jupyter notebook Music_Generation_Using_LSTM.ipynb