Gan-audio
Animal vocalizations will be synthesised using various artificial intelligence techniques to interrogate how computation enables new sonic experiences of 'nature' in the contexts of conservation and ecologically-inspired musical composition.
Area of science
Systems used
Nimbus
Applications used
Audio generative models for PyTorch
Partner Institution: The University of Western Australia |
Project Code: 48b1b73c3b59439aa8bbac54e93c8e40
The Challenge
To generate high-fidelity synthetic animal vocalizations and soundscapes from wildlife sound archives and create a software to manipulate them according to musical principles.
The Solution
A machine learning technique (variational autoencoder) capable of generating high-fidelity audio from publicly available sound libraries.
The Outcome
By speeding up the training process through NVIDIA V100 GPUs.
