Computer Generated Chopin Music

This project uses deep learning to generate classical piano music in the style of Chopin through a novel text-based approach. This project converts MIDI files into text representations that can be easily processed by RNNs.

Best output found here: output/2024-pytorch-rnn/temp-0.8/tmp.mid

Project History & Evolution

2017 Version

The project originally used Andrej Karpathy's char-rnn implementation for generating music. This version produced interesting results (preserved in output/2017-karpathy-LSTM/) but relied on external dependencies and older deep learning frameworks.

2024 Update

After 7 years, the project was modernized with a custom PyTorch-based implementation. The new version features:

• Simplified architecture using PyTorch's built-in LSTM modules
• More efficient training pipeline
• Better code organization and documentation
• Improved generation quality through modern deep learning practices

Technical Overview

Architecture

The current implementation uses a character-level RNN built with PyTorch, consisting of:

• Embedding layer for character encoding
• Multi-layer LSTM network
• Dropout regularization for preventing overfitting
• Linear output layer for character prediction
• Mini-batch training with CUDA GPU support

Data Processing Pipeline

1. MIDI Collection: Source MIDI files are stored in data/midi/
2. Text Conversion: MIDI files are converted to ASCII format using conversion tools in src/conversion/
3. Data Preprocessing: ASCII files are stored in data/ascii/ and processed into training data
4. Training: The PyTorch model is trained on the text representation
5. Generation: New music is sampled from the model at different temperature settings

Model Parameters

• Training options:
  • Hidden layer size: 256 (configurable)
  • Number of LSTM layers: 2 (configurable)
  • Dropout rate: 0.2
  • Batch size: 128
  • Sequence length: 100
• Temperature settings tested: 0.4, 0.8, 1.0
  • Higher temperatures produce more creative but potentially error-prone compositions
  • Lower temperatures generate more conservative, structured pieces

Generated Outputs

The project maintains both historical and current outputs:

• output/2018-karpathy-LSTM/: Original char-rnn generated pieces
• output/2024-pytorch-rnn/: New PyTorch model generations
  • temp-0.4/: Conservative generation
  • temp-0.8/: Balanced creativity/structure
  • temp-1.0/: Most experimental generation

The MIDI files can be played using standard music software like GarageBand or this web player.

Technical Roadmap

1. Extended Training: Increase training duration and dataset size
2. Architecture Tuning:
   • Experiment with larger hidden layer sizes
   • Test different dropout rates
   • Try additional LSTM layers
3. Data Augmentation: Add more Chopin compositions to training set
4. Evaluation Metrics: Implement quantitative measures of musical quality

Implementation Details

The core implementation is in src/model/, featuring:

• train.py: Model definition and training logic
• generate.py: Text generation utilities
• Efficient batch processing
• CUDA optimization for GPU acceleration
• Configurable sequence length for capturing musical patterns
• Checkpoint saving/loading for iterative training

For detailed usage instructions and parameters, see the model documentation in src/model/README.md.