The project "Scyclone," developed by AK-Students Fares Schulz, Christian Scheer, Amirpasha Mobini Tehrani, and Valentin Ackva during the AK-Seminar “Deep Learning for Audio Event Detection", won first place in this year's Neural Audio Plug-in Competition (https://www.theaudioprogrammer.com/neural-audio).
Scyclone is an audio plugin that employs neural timbre transfer technology to introduce a novel approach to audio production. It is based on the RAVE methodology, which is a real-time audio variational auto encoder that enables neural timbre transfer in both single and couple inference mode. This plugin facilitates a new artificial layering technique that can be used to produce richer drum palettes, fuller atmospheres, or simply transfer the timbre of the raw signal to another sound palette. To offer more control over the behaviour and production of the neural networks, the plugin is internally equipped with signal processing modules that allow users to shape, control, and enhance the source and target timbres in a unique way.