Synthesis 3:
Source models

·   Source-filter models (e.g. LPC)

- excitation modified by resonances



fidelity:	moderate-good for appropriate signals
controls:	excitation and resonance separate
cost:	CPU moderate / storage moderate

- good extraction algorithms available
- works best for speech


·   Physical models (e.g. waveguide)

- common structure for musical instruments:



fidelity:	often startling when it works
controls:	reflect physical source, excellent
cost:	CPU moderate / parameter store low

- each model is limited / hard to extend