Characterization of acoustic signals through continuous linear time-frequency representations

One important field in the framework of computer music concerns the modeling of sounds. In order to design digital models mirroring as closely as possible a real sound and permitting in addition intimate transformations by altering the synthesis parameters, we look for a signal model based on additive synthesis whose parameters are estimated by the analysis of real sounds. This model is relevant from both the physical and perceptual points of view, especially when the sound to be analyzed comes from a musical instrument. We present some techniques, mostly unpublished, based on time-frequency representations which make possible the estimation of relevant parameters such as frequency and amplitude modulation laws corresponding to each spectral component of the sound. The techniques described extend the results presented by Delprat et al. (see IEEE Trans. Inform. Theory, vol.38, p.644-65, March 1992). These methods are then transposed to broadband signals, allowing the characterization of transients.