Compression of room impulse responses for compact storage and fast low-latency convolution

Area of Science:

• Signal Processing
• Audio Engineering
• Computer Graphics

Background:

• Room impulse responses (RIRs) are crucial for immersive applications like virtual reality (VR) and augmented reality (AR).
• Real-time audio processing in these applications demands efficient RIR convolution, often hindered by large data sizes and latency constraints.

Purpose of the Study:

• To investigate RIR compression techniques suitable for fast time-domain convolution.
• To evaluate novel low-rank approximation methods against state-of-the-art compression algorithms.
• To enable efficient RIR convolution without prior decompression.

Main Methods:

• Three RIR approximation methods were explored for compression.
• Objective quality measures (channel-based and signal-based) were used for evaluation.
• A novel low-rank-based algorithm for fast time-domain convolution was proposed and simulated.

Main Results:

• Low-rank approximation demonstrated comparable or superior performance to Opus compression across most objective quality measures.
• The proposed method allows convolution directly on compressed RIRs, eliminating decompression overhead.
• Simulations used RIRs of varying lengths from three distinct room environments.

Conclusions:

• Low-rank approximation presents a highly effective strategy for RIR compression.
• This approach offers significant advantages for real-time audio processing in VR/AR by combining compression with efficient convolution.
• The method is a compelling alternative to existing compression standards like Opus for audio-related immersive technologies.