Music Synthesis

In this repo, I'm developing models for a two-stage music synthsis pipeline.

1. A generative model that can produce sequences of low-frequency audio features, such as a mel spectrogram, or a sequence of chroma and MFCC features.
2. A conditional generative model that can produce raw audio from the low-frequency features.

The second stage is inspired by papers developing spectrogram-to-speech vocoders such as:

• MelGAN: Generative Adversarial Networks for Conditional Waveform Synthesis
• High Fidelity Speech Synthesis with Adversarial Networks

You can read more about early experiments developing models for the second stage in this blog post.