IRCAM, 31 rue Saint Merri, 75004, Paris, France
Additive methods developed and used at IRCAM and CNMAT are presented. Analysis includes new techniques to find spectral peaks, estimate their parameters, track partial sinusoids, and separate sinusoidal components from the residual considered as noise. Partials are tracked by identifying trajectories in successive set of spectral peaks using a combinatorial hidden Markov model. Control is made easier by use of spectral envelopes instead of the classical time functions. Characteristic features of sounds are given in term of fundamental frequency, spectral envelopes for amplitude, phase, deviation from harmonicity, noise components, etc. This procedure renders user control much more economical and efficient. To reduce the computational cost of the synthesis algorithm, a method has been developed that is based on the short-time Fourier model of a sound signal allowing for the synthesis and control of hundreds of sinusoidal partials and noise components in real time on a desktop computer without custom hardware. Very high quality sound examples are obtained showing that a limited cost real-time multitimbral instrument can easily be designed based on inverse FFT. This would have all the possibilities of present day synthesizers, plus many others such as the precise modifications of sampled sounds, speech, and singing voice synthesis.