Neural Network Model for Use in Performing Pitch Correction in a Voice-Driven Musical Instrument
John Carelli · 2020
A neural network model is presented for use in musical note recognition and pitch correction using only recently sung notes captured in real time. The goal is to improve, by correcting inaccuracies in singing execution, the performance of a voice driven musical instrument that translates sung pitch into notes played by a separate virtual instrument. This is accomplished without knowledge of musical key and, in order to enable real-time response, with a minimal number of recently sung notes. Model development, training, and testing in the voice-driven instrument are described. Overall, the study provides a unique and potentially useful investigation of a human/computer interaction with application in live musical performance.