This repo contains:
- Tooling to split monophonic signal into a set of fixed-width waveforms and to reconstruct the signal
- Basic waveform autoencoder
Take a look at demo-last-crossing-point.ipynb and demo-stretch.ipynb for a short demo
Some examples:
- Voice, reconstructed from the latent vectors (stretch)
- Voice, reconstructed from the latent vectors (last crossing point)
- Voice, reconstructed from the original set of waveforms with fixed pitch (stretch)
- AKWF waveforms library, encoded with the human voice autoencoder (last crossing point)
- Voice with fixed pitch and some gaussian smoothing applied in the latent space (stretch)
Two methods for wavelength encoding are provided:
- encoding wavelength into a separate channel
- encoding by last zero-crossing point
