Differentiable Artificial Reverberation
نویسندگان
چکیده
Artificial reverberation (AR) models play a central role in various audio applications. Therefore, estimating the AR model parameters (ARPs) of reference is crucial task. Although few recent deep-learning-based approaches have shown promising performance, their non-end-to-end training scheme prevents them from fully exploiting potential deep neural networks. This motivates introduction differentiable artificial (DAR) models, allowing loss gradients to be back-propagated end-to-end. However, implementing with difference equations “as is” learning framework severely bottlenecks speed when executed parallel processor like GPU due infinite impulse response (IIR) components. We tackle this problem by replacing IIR filters finite (FIR) approximations frequency-sampling method. Using technique, we implement three DAR models—differentiable Filtered Velvet Noise (FVN), Advanced (AFVN), and Delay Network (DN). For each model, train its ARP estimation networks for analysis-synthesis (RIR-to-ARP) blind (reverberant-speech-to-ARP) task an end-to-end manner counterpart. Experiment results show that proposed method achieves consistent performance improvement over both objective metrics subjective listening test results.
منابع مشابه
Circulant and elliptic feedback delay networks for artificial reverberation
The feedback delay network (FDN) has been proposed for digital reverberation. Also proposed with similar advantages is the digital waveguide network (DWN). This paper notes that the commonly used FDN with an N × N orthogonal feedback matrix is isomorphic to a normalized digital waveguide network consisting of one scattering junction and a vector transformer joining N reflectively terminated bra...
متن کاملDifferentiable Invariants
Invariants that incrementally maintain the value of expressions under assignments to their variables are a natural abstraction to build high-level local search algorithms. But their functionalities are not sufficient to allow arbitrary expressions as constraints or objective functions as in constraint programming. Differentiable invariants bridge this expressiveness gap. A differentiable invari...
متن کاملDifferentiable Imbeddings
1. Terminology. V and M will be differentiable manifolds of dimension n and m respectively; differentiable meaning always of class C. For simplicity, we assume V compact and without boundary. We shall have to consider several categories of maps: (1) the category of continuous maps, (2) the category of topological imbeddings, (3) the category of topological immersions: a map ƒ: F—>M is a topolog...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: IEEE/ACM transactions on audio, speech, and language processing
سال: 2022
ISSN: ['2329-9304', '2329-9290']
DOI: https://doi.org/10.1109/taslp.2022.3193298