Acoustical and behavioral heuristics for fast interactive sound design

Area of Science:

• Human-Computer Interaction
• Acoustic Engineering
• Computational Design

Background:

• Collecting user feedback in design is crucial but often time-consuming and costly.
• Interactive sound design faces inefficiency risks due to vast design spaces and limited human subject testing.
• Optimizing perceptual qualities like pleasantness and detectability requires efficient exploration of design variations.

Purpose of the Study:

• To propose and evaluate heuristics for reducing the design space in interactive optimization processes.
• To investigate the effectiveness of surrogate models in approximating perceptual quantities for design optimization.
• To accelerate the convergence of interactive design, particularly for tasks with large search spaces.

Main Methods:

• Development of heuristics utilizing surrogate models to approximate perceptual quantities.
• Application to the interactive design of pleasant and detectable electric vehicle sounds using a genetic algorithm.
• Comparison of acoustical descriptor-based and behavioral data-based surrogate models.

Main Results:

• Heuristics reduced the design space by up to 64 times, decreasing iterations by up to 2 for equivalent performance.
• Behavioral data-based surrogate models yielded superior overall design improvements.
• Acoustical descriptor-based models require careful selection of descriptors to be effective.

Conclusions:

• The proposed heuristics significantly accelerate interactive design convergence.
• Surrogate models, especially those based on behavioral data, are effective in optimizing complex sound design parameters.
• This approach is highly suitable for applications where exhaustive search is impractical due to time or cost constraints.