Sound texture synthesis using an enhanced overlap-add approach

Texture synthesis has been an active field of research recently, mainly focusing on 2D texture synthesis. Some examples are wall and grass textures in computer-generated graphics and video. In video game sound design, background sounds are very important in creating different mood and feeling when game players enter different scenes of the video game. In order to save disk space and memory, sound designers tend to loop a short (usually 5~15 second) background sound clip to produce a longer (usually more than 1 minute) background sound clip. The background sound produced using this approach often make the player feel bored. So recent work has applied the idea of visual texture synthesis to audio texture synthesis, hoping to synthesis a longer sound clip from a shorter one. However, very little research has successfully synthesized sound textures without introducing audible artifacts. This paper describes an enhanced overlap-add method for sound texture synthesis. It is especially well-suited to sounds with a quasi-oscillatory or random character such as waves crashing on the shore and traffic noise. It also works well for animal sounds when there are several sources, such as a group of chattering birds. These are important classes of sounds in computer games, films, and advertisements. The results also suggest that better textures can be achieved if acoustic effects are applied. The approach allows for real-time synthesis of sound textures, which is useful in applications such as computer games and sound editors