Machine learning for game audio


Using Flucoma as a main tool Francesco will try to build real examples on how sound designers can use neural networks on their advantage. Flucoma is a toolset that provides max msp, pure data and supercollider objects to investigate manipulation on large banks of sounds. Three degrees of manipulations are explored: (1) expressive browsing and descriptor-based taxonomy, (2) remixing, component replacement, and hybridisation by concatenation, and (3) pattern recognition at component level, with interpolating and variation-making potential. These novel manipulations will yield new sounds, new musical ideas, and new approaches to large corpora.


• Sound recognition thorugh spectral analysis
• Classifying sounds using a neural network
• How to train a neural network in order to generate bespoken UI sounds
• Exploring text-to-sound techniques
• Synthetising foley using GAN (generative adversarial network)