Sound propagation in massive scenes: Our algorithm reduces precompute and runtime costs allowing hundreds of sources to be rendered on a single CPU core.

We present a method for coordinating enhanced VoIP communication between many clients in real-time. For each client a unique environmental response is generated based on the client's location in the scene. By efficiently compressing precomputed environmental responses, our system can deliver an appropriate early response for each sound source as well as a late decay for each final output stream.

• Describe acoustic signature to quickly sample large scenes
• Reduce set of precomputed responses by combining similar samples
• Efficiently compute propagation responses on scenes with thousands of receivers
• Compress thousands of responses using efficient data structures

Basic process:

• -Form similarity metric
  -Geometric properties, related to acoustic rendering equation
  -Based on first order direction, distance, material, and diffusion
  
• -Use metric to segment scene into acoustic regions
  -Run simulation on regions
  -Collect responses

 

• -Store non-zero unique responses
  -Hashed list based on scene location
  -Insert time: avg O(log n)
  -Lookup time: avg O(1)
  -Storage size: avg k O(n), k related to audio energy