Core Concepts
The proposed method automatically learns the parameters of a feedback delay network (FDN) to closely match the perceptual qualities of a measured room impulse response, by optimizing the FDN's delay lines, feedback matrix, and gains via backpropagation.
Abstract
The content discusses a novel method for automatically tuning the parameters of a feedback delay network (FDN) to model the acoustic response of a physical environment.
Key highlights:
- Existing methods for automatic FDN parameter tuning either focus on off-the-shelf reverb plug-ins, limit the set of target parameters, or augment the FDN topology with auxiliary frequency-dependent components.
- The proposed approach involves implementing a differentiable FDN with trainable delay lines, allowing simultaneous optimization of all FDN parameters (delays, feedback matrix, gains) via backpropagation.
- The optimization process seeks to minimize a time-domain loss function incorporating differentiable terms accounting for energy decay and echo density.
- Experimental validation shows the proposed method yields FDNs capable of closely matching the desired acoustical characteristics, outperforming existing gradient-free and gradient-based techniques.
- The learned FDN parameters can be seamlessly plugged into off-the-shelf FDN software without further processing or mapping.
Stats
The room reverberation time (T20, T30, T60) is used to evaluate the performance of the proposed method.