Implementation of the DDSP model using PyTorch. Forked from Artificial Creative Intelligence and Data Science.
Current additions to forked code include: encoders, decoders and a ResNet autoencoder.
Current development: very Experimental autoencoder for unsupervized pitch detection (ConvLSTM).
Potential next steps: creating a synthetic data pipeline for self-supervised feature extraction; adding more autoencoders for unsupervised/semi-supervised synthesis.