Rodrigo Mauricio Diaz Fernandez
Hybrid Neural Methods for Sound Synthesis
This project will use the Differential DSP approach to Deep Learning to build models for acoustic musical instruments. Such physically inspired neural networks promise to deliver more efficient machine learning solutions that need less training and are less resource-hungry in deployment than current approaches. The research will start by applying the approach to plucked and struck strings, and moving on to membranes (as in drums) and other percussive instruments. The resulting models will expose to the instrumentalist some intuitive parameters that lead to compelling and playable synthetic musical instruments. These will be evaluated both quantitatively and qualitatively.
C4DM theme affiliation:
Sound Synthesis, Audio Engineering