Understanding the Perception of Granular Processing

July 15th, 2003

SMPC 2003 National Conference @ University of Nevada – Las Vegas.

Co-authors: Gary Kendall and Scott Lipscomb.

EXCERPT -Granular processing of sampled audio signals (Roads 1985; Traux 1987) is a technique that has experienced increased interest in recent years. This processing technique has origins in the work of Dennis Gabor and his concept of the acoustical quanta (Gabor 1947), which was developed in response to perceptual research findings. He asserted that, “it is our most elementary experience that sound has a time pattern as well as a frequency pattern (Gabor 1947, p. 591).” Although several software applications are now widely available for realizing granular processing effects (Roads and Alexander 1997; Behles, Starke and Roebel 1998; van der Schoot 1999; Rolfe and Keller 2000), these programs have done little to clarify the perceptual connection between interface controls and audio output, a problem that has persisted since the first computer implementation was reported twenty-five years ago (Roads 1978). A better understanding of how the audio output is perceived is a necessary precursor to the development of a simplified interface that would require the computer “to interpret how to approximate a desired result” (Roads 2001a, p. 27).

In order to further this goal of understanding the perception of granular processing better, the author has conducted a series of three experiments based on models found within studies of musical timbre (Grey 1977; Wessel 1979; Kendall and Carterette 1991; Iverson and Krumhansl 1993). These studies have employed a method of exploring the topic using a similarity-scaling task. Subjects would listen to a series of pairs of sounds, rating the similarity of each pair member to the other while moving through all possible couplings of the stimuli. These similarity ratings were then averaged together and used as the basis for formulating a multi-dimensional scaling solution (Shepard 1962a, 1962b; Kruskal 1964a, 1964b). MDS uses these similarity ratings to produce a graphic representation of the relationships that exist within that data. Stimuli that are viewed as similar will be placed in close proximity to one another, while those viewed as dissimilar will be have a greater distances between them. The differences are represented in a single graphic plot of points representing the stimuli used. This seemed like an appropriate method to employ in our study, however we must emphasize that we are not attempting to draw connections between the results of our study and those found in timbre research. Our comparison is simply with their experimental methods.

Leave a Reply