Note-by-note synthesis is not enough
Roger B. Dannenberg · The Journal of the Acoustical Society of America · 2001
Traditional analysis and synthesis of musical tones are based on the idea that music can be deconstructed into note units that can be individually synthesized and recombined to create music. This ‘‘divide and conquer’’ strategy is natural for scientists and engineers, but it does not always work well in practice. In particular, brass instrument tones in a musical context are quite different from brass tones produced in isolation. Synthesis methods that are capable of reproducing individual tones fail to sound realistic when these tones are sequenced to produce a musical phrase. The combined spectral interpolation synthesis (CSIS) method uses a fairly simple spectrally based model in which the dynamically changing spectrum is a function of fundamental frequency and rms amplitude. The spectrum function is determined by measuring actual acoustic performances. The synthesis model must be controlled by appropriately evolving frequency and amplitude controls. A rule-based system computes these controls from a machine-readable version of a musical score, which includes slurs and other articulation markings. The rules take musical context and phrasing into consideration. The combination of musical context and a good synthesis model results in greatly improved synthesis.