Concatenative Music Synthesis by Note Separation using Onset Velocity Detection and Statistical Models.

摘要

Music synthesis is a process of synthesizing musical instruments using signal processing. This thesis is aimed to deliver a realistic concatenative music synthesis, where the system is statistically modeled with the musical notes from recorded music. This is implemented in three phases. First by iteratively extracting out the musical notes from the individual instrumental tracks to create a concordance of musical notes, then by statistically modeling the note parameters and finally reconstructing the parameters to form a concatenative music pattern. Individual musical note is extracted from the separated tracks by Onset Velocity Detection. Hidden Markov Model based Music Synthesis System is built to statistically model the musical parameters. Waveform is reconstructed by passing the parameters through synthetic filters to synthesize music following artists playing and compositional style.