Skip to main content
This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub
This node converts a compressed audio representation (latent samples) back into an audio waveform using a Variational Autoencoder (VAE). It processes the data in smaller, overlapping sections (tiles) to manage memory usage, making it suitable for handling longer audio sequences.

Inputs

ParameterData TypeRequiredRangeDescription
samplesLATENTYesN/AThe compressed latent representation of the audio to be decoded.
vaeVAEYesN/AThe Variational Autoencoder model used to perform the decoding.
tile_sizeINTNo32 to 8192The size of each processing tile. The audio is decoded in sections of this length to conserve memory (default: 512).
overlapINTNo0 to 1024The number of samples that adjacent tiles overlap. This helps to reduce artifacts at the boundaries between tiles (default: 64).

Outputs

Output NameData TypeDescription
outputAUDIOThe decoded audio waveform.