E. Bryan George
Signal Processing Ctr. of Technol., Lockheed Sanders, Inc., PTP2-A001, P.O. Box 868, Nashua, NH 03061-0868
Mark J. T. Smith
Georgia Inst. of Technol., Atlanta, GA 30332-0250
Sinusoidal models have been successfully applied to a variety of speech processing problems, including speech analysis/synthesis. Due to their generality they are also becoming important for the analysis, synthesis, and modification of music signals. The analysis-by-synthesis/overlap-add (ABS/OLA) system is a sinusoidal modeling system which is well adapted to represent and synthesize musical instrument tones. The ABS/OLA system uses an analysis-by-synthesis strategy to determine the parameters of well-resolved sinusoids much more accurately than is possible by direct Fourier analysis, and uses a computationally efficient overlap-add synthesis model that has been generalized to produce artifact-free synthetic tones in the presence of time- and frequency-scale modification. This paper summarizes the ABS/OLA system, details its application to the analysis, synthesis, and modification of musical tones, and presents recent results in its development, including generalization to quasiharmonic tones and improvements in the computational efficiency of analysis.