Title:
SPEECH ENHANCEMENT METHOD AND DEVICE USING FAST FOURIER CONVOLUTION
Document Type and Number:
WIPO Patent Application WO/2023/182765
Kind Code:
A1
Abstract:
Methods for processing and analyzing audio recordings and, in particular, for speech denoising are provided. A method for speech denoising using a fast Fourier convolution operator comprises: splitting channels of input tensor into local and global branches; using conventional convolutions for local updates of feature maps at the local branch; performing Fourier transform across frequency dimension of the global branch feature map; updating the global branch feature map in spectral domain by point-wise convolutions; applying an inverse Fourier transform to the updated global branch feature map; and summing local and global branches activations. The technical result consists in improving the quality of speech denoising and/or enhancement of speech component in a speech audio signal.
Inventors:
SHCHEKOTOV IVAN SERGEEVICH (RU)
ANDREEV PAVEL KONSTANTINOVICH (RU)
ALANOV AIBEK ARSTANBEKOVICH (RU)
IVANOV OLEG YURIEVICH (RU)
VETROV DMITRY PETROVICH (RU)
ANDREEV PAVEL KONSTANTINOVICH (RU)
ALANOV AIBEK ARSTANBEKOVICH (RU)
IVANOV OLEG YURIEVICH (RU)
VETROV DMITRY PETROVICH (RU)
Application Number:
PCT/KR2023/003711
Publication Date:
September 28, 2023
Filing Date:
March 21, 2023
Export Citation:
Assignee:
SAMSUNG ELECTRONICS CO LTD (KR)
International Classes:
G10L21/0208; G06N3/0455; G06N3/0464; G10L25/18; G10L25/24; G10L25/30
Domestic Patent References:
| WO2021251627A1 | 2021-12-16 |
Foreign References:
| CN113314140A | 2021-08-27 | |||
| CN113655986A | 2021-11-16 | |||
| CN108768542A | 2018-11-06 |
Other References:
CHI LU, JIANG BORUI, MU YADONG: "Fast Fourier Convolution", NIPS`20: PROCEEDINGS OF THE 34TH INTERNATIONAL CONFERENCE ON NEURAL INFORMATION PROCESSING SYSTEMS, 1 January 2020 (2020-01-01), XP093095623
Attorney, Agent or Firm:
KIM, Tae-hun et al. (KR)
Download PDF: