DAFX: Digital Audio Effects, Second Edition

10.3 Techniques

In this section we present and discuss a set of implementations of the different spectral models, building one implementation on top of another and making sure that each implementation is self-contained, thus being useful for some applications. Each implementation is an analysis/synthesis system that has as input a monophonic sound, , plus a set of parameters, and outputs a synthesized sound, y(n). If the parameters are set correctly and the input sound used is adequate for the model, the output sound should be quite close, from a perceptual point of view, to the input sound. We start with an implementation of the STFT, which should yield a mathematical input-ouput identity for any sound. On top of that we build an analysis/synthesis system based on detecting the spectral peaks, and we extend it to implement a simple sinusoidal model. The following system implements a sinusoidal model with a harmonic constraint, thus only working for pseudo-harmonic sounds. We then include the residual analysis, thus implementing a harmonic plus residual model. Finally we assume that the residual is stochastic and implement a harmonic plus stochastic residual model.

10.3.1 Short-time Fourier Transform

Figure 10.3 shows the general block diagram of an analysis/sysnthesis system based on the STFT.

Figure 10.3 Block diagram of an analysis/synthesis system based on the STFT.

In order ...