Abstract

Short Paper

The usage of sampling synthesis has become the most common technique for generating sound in computer music pieces and specially when computer generated sounds are incoporated into multimedia productions. Most artists do not conceive any other alternative. This presentation will focus on new developments in the area of digital sound synthesis that take sampling beyond the current limitations, offering new creative possibilities to artists. Sampling is based on reproducing and transforming, in the time domain, preexisting sounds. It has the virtue of maintaining the qualities of real sounds, but the number of possible transformations is limited and most of them are not musically intuitive. Another traditional approach to synthesis is to generate sounds from abstract mathematical equations, such as frequency modulation, which is a powerful and flexible way to synthesize new sounds. This last approach has lost ground in recent years in favor of sampling because of the difficulty in getting rich and realistic sounds. As another alternative, starting to be used by the computer music community, we propose a set of analysis/synthesis techniques based on spectral models. These techniques bring the possibility of obtaining perceptually based parametrizations of most sounds, which can then be transformed in flexible and intuitive ways before resynthesis is done. This approach maintains the richness and realistic qualities of sampling and brings the flexibility of FM. Spectral models can be thought of as a description of the sound characteristics that the listener perceives. There are several signal processing strategies developed in the last few years with which we can obtain these perceptual parameters and synthesize new sounds from the analysis data or its transformations. Fourier analysis would represent the first step towards this perceptual modeling of sounds. Sounds are decomposed in its frequency components, of which we can study their time evolution. A step further is to decompose sounds into sinusoids (partials) and noise (residual component), that is, analizing sounds with this model and generating new ones from the analysis (Serra, 1995; Serra, 1994; Serra and Smith, 1990). The analysis detects the partials by looking at spectra and represents them as sinusoids. These partials are then subtracted from the original sound and the residual is represented as filtered noise. This results in a synthesis process that combines additive and subtractive synthesis techniques. With spectral models we can reproduce existing sounds and we can go beyond them by modifying their perceptual atributes. One class of interesting transformations creates new sounds by mixing or interpolating the perceptual attributes of two or more sounds, resulting into what could be called “sound morphs”. The analysis detects the partials by looking at spectra and represents them as sinusoids. These partials are then subtracted from the original sound and the residual is represented as filtered noise. This results in a synthesis process that combines additive and subtractive synthesis techniques. With spectral models we can reproduce existing sounds and we can go beyond them by modifying their perceptual atributes. One class of interesting transformations creates new sounds by mixing or interpolating the perceptual atributes of two or more sounds, resulting into what could be called “sound morphs”.

• Xavier Serra, Spain en.wikipedia.org/wiki/Xavier_Serra