First iteration of the Elaine AI voice (counter-)model for Femke Herregraven
The project started as an experiment to test the idea of a counter voice model which includes all the noise, mistakes, errors that typical voice models filter out in the pre-training and training phase. I built a system that trains a model with a bespoke, audio version of stable diffusion and generates vocalisations when the model is inferred.
The generated sounds were composed by Benny Nilsen to the soundtrack of the central video piece of the work, titled: The Murmur of the Dying.
- Google Collab
- AWS SageMaker
- Python
- PyTorch
- Weights&Biases