DAFX: Digital Audio Effects, Second Edition

8.3 Source-filter Transformations

8.3.1 Vocoding or Cross-synthesis

The term vocoder has different meanings. One is “voice-coding” and refers directly to speech synthesis. Another meaning for this term is the phase vocoder, which refers to the short-time Fourier transform, as discussed in Section 7.2. The last meaning is the one of the musical instrument named the Vocoder and this is what this paragraph is about: vocoding or cross-synthesis.

This effect takes two sound inputs and generates a third one which is a combination of the two input sounds. The general idea is to combine two sounds by “spectrally shaping” the first sound by the second one and preserving the pitch of the first sound. A variant and improvement are the removal of the spectral envelope of the initial sound (also called whitening) before filtering with the spectral envelope of the second one. This implies the ability to extract a spectral envelope evolving with time and to apply it to a signal.

Although spectral estimation is well represented by its amplitude versus frequency representation, most often it is the filter representation that can be a help in the application of this spectral envelope: the channel vocoder uses the weighted sum of filtered bandpass signals, the LPC calculates an IIR filter, and even the cepstrum method can be seen as a circular convolution with an FIR filter. As this vocoding effect is very important and can give different results depending on the technique used, we will introduce ...