Generating Artificial Reverberation via Genetic Algorithms for Real-Time Applications.

Area of Science:

• Digital Signal Processing
• Acoustics
• Artificial Intelligence

Background:

• Convolution reverb is crucial for realistic audio effects.
• Generating diverse and controllable Room Impulse Responses (RIRs) is challenging.
• Existing methods may not offer sufficient control over acoustic parameters.

Purpose of the Study:

• To develop a Virtual Studio Technology (VST) 2 audio effect plugin for convolution reverb.
• To utilize a Genetic Algorithm (GA) for generating synthetic Room Impulse Responses (RIRs).
• To enable user control over RIRs using ISO 3382-1 standard parameters.

Main Methods:

• A Genetic Algorithm (GA) evolves synthetic RIRs from Gaussian noise.
• GA employs truncation selection, weighted average crossover, and Gaussian multiplication mutation.
• Binaural Room Impulse Responses (BRIRs) are generated for stereo output.
• ISO 3382-1 parameters (reverberation time, clarity) guide RIR fitness.

Main Results:

• Objective evaluation confirmed that contradictory parameter combinations yield low-fitness RIRs.
• Subjective evaluation showed GA-generated RIRs are perceptually distinguishable from real ones.
• Perceptual differences decreased with longer GA execution times and speech-only signals.

Conclusions:

• The proposed VST plugin effectively generates controllable synthetic RIRs and BRIRs.
• The GA approach allows for the creation of novel virtual acoustic spaces.
• Further optimization of GA parameters can enhance the perceptual realism of synthetic RIRs.