HARMONY—
A deep learning model capable of generating original piano melodies by learning musical patterns from MIDI datasets.
Harmony learns musical structures from thousands of MIDI sequences and predicts the next note or chord using stacked LSTM networks. Instead of memorizing songs, the model captures rhythm, harmony and melodic progression to generate entirely new compositions.
The training pipeline converts symbolic music into numerical sequences, performs preprocessing and trains recurrent neural networks capable of generating coherent note progressions. Generated outputs are converted back into playable MIDI files for immediate listening.
The project introduced concepts such as sequence modeling, recurrent neural networks, dataset preprocessing and creative AI, while demonstrating how deep learning can be applied beyond traditional prediction tasks.