Reverberation is a critical factor in creating an immersive gameplay for 3D games or an interactive VR experience. What’s called room impulse response (RIR) is used to replicate reverberation. RIR is a simulated path of sound particles on a time domain that includes time delay and level attenuation information in regards to the reflection, absorption and diffraction, etc. occurring in the room. Thus, RIR contains the room geometry information, such as physical characteristics of the room and source-to-listener position information, as well as occlusion.
Direct Sound, Early Reflection and Late Reverberation
RIR in general consists of direct sound, early reflection and late reverberation. Each of these components depicts different aspects of a space. So, each component should be fully utilized to create believable reverberation.
Both direct sound and early reflection characterize the direction information of a sound source in a room. The main difference between the two is that early reflection contains the physical room characteristics. Direct sound is the direct path between the sound source and the listener, so the frequency is determined by the listener’s head size and shape, as well as ear size and shape. Early reflection frequency is influenced by the material and structure of the wall or obstacles laid in the space because the sound reaches both ears after being reflected from the room, as well as occlusion. The propagation path of early reflection is longer than that of direct sound, resulting in more time delay and level attenuation.
Direct sound and early reflection can be better perceived with binaural signal rather than mono signal, as humans recognize the direction or position of a sound source using both ears. Binaural signal is also more effective in sound source spatialization and reproducing interactivity of sound source. However, using the Binaural Room Impulse Response (BRIR) data to create reverberation for VR is not as easy as using the Head Related Transfer Function (HRTF) data for VR. Unlike HRTF, where sound is measured in an anechoic or semi-anechoic space, BRIR already contains the room information. BRIR therefore has much larger possible scenarios, which results in a huge database size. This requires large memory for a renderer and is definitely a burden to process.
Late reverberation is highly influenced by the physical characteristics of the room. Its frequency is rather diffused without direction information of a microphone or listener. It is more like an ambience, pinpointing the richness of sound.
Creating Reverberation Using Image Source Method and Ray-based Method
If not measured, RIR can be created by calculating the sound particle propagation path using geometry-based methods, such as Image Source Method and Ray-based Method.
The Image Source Method simulates the virtual position of reflected sound, while the Ray-based Method tracks the propagation path of sound particles. The result is not vastly different. It’s a matter of choice regarding which method will require a smaller real-time computation when a given room environment is used—Ray-based method requires smaller computation power for a more complex room structure, and vice versa.
Geometry-based methods need extra layers of processing because they cannot naturally create late-reverb-specific frequency responses or time decay of energy. They assume sound as a ball, as if the ball is reflected from a wall. However, sound is a wave, and it can be affected by the material it hits. Whether the wall is made of glass, cement or brick can make the sound totally different, so just simulating the propagation path is not enough to create realistic sound.
These geometry-based methods require more complex computation than choosing a measured RIR from the RIR database. However, these methods have been frequently used in 3D game production, as the video tracing technology such as lighting is already embedded in 3D game development tools. Audio tracing can leverage video tracing technology and apply it similarly.
When it comes to creating reverberation for VR or game audio, the geometry-based solutions currently available in the market depend on desktop-based VR with high performance GPU–the ones with high barrier for consumer adoption. Audio solution companies, including ourselves, are developing a reverb solution with low device dependency and real-time RIR synthesizing support, while requiring low computation.