Prediction of Missing Frequency Response Functions Through Deep Image Prior

Malvermi, Raffaele; Antonacci, Fabio; Sarti, Augusto; Corradi, Roberto

doi:10.1109/waspaa52581.2021.9632759

Cited by 2 publications

(6 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recently, in [ 44 ] the Deep Image Prior (DIP) method has been successfully employed for solving inverse problems in the context of image restoration. Similarly, DIP was applied for the reconstruction of irregularly sampled seismic [ 48 ] and vibrational [ 49 ] data, problems akin to RIR reconstruction. In this context, we consider a deep neural network as a generator described by the parametric nonlinear function

so that

and the solution of the inverse problem is now given by rewriting ( 6 ) as

where

represents the learnable parameters of the network and

is a random noise realization given as input.…”

Section: Problem Formulationmentioning

confidence: 99%

“…Inspired by the effectiveness of U-Net-like networks [ 41 ] in existing deep prior applications [ 44 , 46 , 48 , 49 ], we decide to adopt the MultiResUNet [ 51 ] also for facing the RIR reconstruction problem discussed in this manuscript. MultiResUNet is first introduced in [ 51 ], where it shows improved results for the segmentation of multimodal medical images.…”

Section: Network Descriptionmentioning

confidence: 99%

“…The network is implemented in PyTorch [ 54 ] and trained using the Adam optimizer [ 55 ] for 3000 epochs with learning rate set to

. Similarly to [ 44 , 48 , 49 ], in order to reinforce convergence at each iteration the input tensor is perturbed with additional zero-mean white noise of variance

.…”

Section: Network Descriptionmentioning

confidence: 99%

“…Recently, Deep Image Prior (DIP) [ 44 ] has been introduced as a new framework for solving image restoration tasks (e.g., inpainting, denoising, super resolution, etc.). The same deep prior strategy has been applied also for the interpolation of seismic traces [ 46 , 47 , 48 ], vibrational data [ 49 ], and denoising of audio signals [ 50 ]. The main idea of DIP is to interpret restoration as an inverse problem, for which the adopted CNN provides a regularization prior [ 44 ].…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Deep Prior Approach for Room Impulse Response Reconstruction

Pezzoli

Perini

Bernardini

et al. 2022

Sensors

Self Cite

View full text Add to dashboard Cite

In this paper, we propose a data-driven approach for the reconstruction of unknown room impulse responses (RIRs) based on the deep prior paradigm. We formulate RIR reconstruction as an inverse problem. More specifically, a convolutional neural network (CNN) is employed prior, in order to obtain a regularized solution to the RIR reconstruction problem for uniform linear arrays. This approach allows us to avoid assumptions on sound wave propagation, acoustic environment, or measuring setting made in state-of-the-art RIR reconstruction algorithms. Moreover, differently from classical deep learning solutions in the literature, the deep prior approach employs a per-element training. Therefore, the proposed method does not require training data sets, and it can be applied to RIRs independently from available data or environments. Results on simulated data demonstrate that the proposed technique is able to provide accurate results in a wide range of scenarios, including variable direction of arrival of the source, room T60, and SNR at the sensors. The devised technique is also applied to real measurements, resulting in accurate RIR reconstruction and robustness to noise compared to state-of-the-art solutions.

show abstract

so that

and the solution of the inverse problem is now given by rewriting ( 6 ) as

where

represents the learnable parameters of the network and

is a random noise realization given as input.…”

Section: Problem Formulationmentioning

confidence: 99%

Section: Network Descriptionmentioning

confidence: 99%

“…The network is implemented in PyTorch [ 54 ] and trained using the Adam optimizer [ 55 ] for 3000 epochs with learning rate set to

. Similarly to [ 44 , 48 , 49 ], in order to reinforce convergence at each iteration the input tensor is perturbed with additional zero-mean white noise of variance

.…”

Section: Network Descriptionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Deep Prior Approach for Room Impulse Response Reconstruction

Pezzoli

Perini

Bernardini

et al. 2022

Sensors

Self Cite

View full text Add to dashboard Cite

show abstract

“…Through this approach, a network learns how to map a random noise realization into the solution of the assigned task, relying on a single data item. In addition to the image domain, deep prior has also been successfully applied in the context of seismic data [20], room impulse responses [21] and vibrometric data [22].…”

Section: Introductionmentioning

confidence: 99%

Deep Prior-Based Audio Inpainting Using Multi-Resolution Harmonic Convolutional Neural Networks

Miotello,

Pezzoli,

Comanducci

et al. 2024

IEEE/ACM Trans. Audio Speech Lang. Process.

Self Cite

View full text Add to dashboard Cite

Prediction of Missing Frequency Response Functions Through Deep Image Prior

Cited by 2 publications

References 18 publications

Deep Prior Approach for Room Impulse Response Reconstruction

Deep Prior Approach for Room Impulse Response Reconstruction

Deep Prior-Based Audio Inpainting Using Multi-Resolution Harmonic Convolutional Neural Networks

Contact Info

Product

Resources

About