VintageGold/Music_Generation

Pytorch deep learning generative notes based on 50 "Funk" basslines

Music_Generation

Read in Funk basslines Midi files and generate music based on 10 note sequences by using the 10 note sequences to predict the next 10 note sequences

Conclusion

LSTM and GRU models should be trained for 350 epochs to allow for learning room as around 390 model accuracy falls.

Example Song

https://soundcloud.com/vintagegold123/bass-11-robot-notes-sample

Model Parameters

LSTM: batch_size = 256, epochs = 350, n = 1024, num_layers = 2, dropout = 0.3, learning_rate = 0.8, loss_func = nn.CrossEntropyLoss())

GRU: batch_size = 256, epochs =390, n = 1024, num_layers = 2 ,dropout = 0.3, learning_rate = 0.8, loss_func = nn.CrossEntropyLoss())