March 3, 2026Open Access

Differentiable Grouped Feedback Delay Networks for Learning Direction and Position-Dependent Late Reverberation

Key Points

The proposed Differentiable Grouped Feedback Delay Networks effectively synthesize late reverberation characteristics.
At 0.6 m and 0.9 m, the method exhibits a slight DEDC error increase of 1.5 dB and 0.9 dB, respectively.
Training focuses on minimizing spectral deviation using a Multi-Layer Perceptron to predict receiver gains based on listener position.
Highlights potential advancements in audio rendering efficiency, particularly in real-time applications.

Abstract

Late reverberation in coupled spaces shows direction and position-dependent decay. Grouped Feedback Delay Net- works (GFDNs) are artificial reverberators capable of ef- ficient rendering of coupled room acoustics by modelling multi-slope decays. We extend the concept of Differentiable GFDN (DiffGFDN) to synthesise position- and direction- dependent late reverberation. The proposed DiffGFDNs operate in octave bands, with spherical harmonic receiver gains predicted from listener positions via a Multi-Layer Per- ceptron. Training minimises spectral deviation, promotes low sparsity, and matches Directional Energy Decay Curves (DEDCs). At inference, signals are processed in frequency subbands, beamformed, and rendered binaurally or over loud- speakers. The proposed method shows slightly higher DEDC errors (+1.5 dB at 0.6 m, +0.9 dB at 0.9 m) than the best baseline, but is the most computationally efficient, enabling faster 6-DoF rendering than time-space-varying convolution.

Differentiable Grouped Feedback Delay Networks for Learning Direction and Position-Dependent Late Reverberation

Key Points

Abstract

Cite This Study