Parametric Sound Texture Generator

Diploma Thesis (8.452 KB)

Sound texture modeling is a widely used concept in computer music. Although, a concrete definition of sound texture is elusive, with this thesis I try to determine the scope of the different fields of acoustic texture. After the report on the current state of different sound texture generation methods I will outline common problems of the sound texture examples.

From the presented literature two existing algorithms, namely audio texture and natural grains, which are based on a similar granular analysis/resynthesis approach will be further investigated. Both algorithms can be used for creating a parametric sound texture generator that allows creating sounds out of a simple parameter control structure. Starting from a short input texture, different, new sound textures of variable length are produced. These new textures consist of consecutively connected segments that are similar to the input texture.

A special characteristic of the selected algorithms is their property of segmenting the original signal into perceptually meaningful units. Hence, sound textures which are rather associated as soundscapes that consist of different streams are particularly examined. The analysis and the resynthesis of the algorithms is effectively improved by means of parametric modifications so that soundscapes with distinct events are broken up only at points that make sense to our auditory perception.

The implementation goal of this thesis is a real-time Pure Data interface for demonstration purposes allowing the user to manually adjust parameters until the produced sound texture sounds plausible and realistic with respect to a particular use of the texture. Ideally, the sound textures are not recognized as a resynthesized version, new sounds are considered as being natural, segments are concatenated in a natural flow and no disturbing repetitive patterns or artefacts can be heard.