A Real-time Speech Noise Reduction Method Based on Skip Network

A real-time speech and noise reduction technology, applied in speech analysis, neural learning methods, biological neural network models, etc., can solve the problems of large deep neural network parameters, complex models, and long audio processing time, and achieve good speech noise reduction Effect, good noise reduction effect, small distortion effect

Active Publication Date: 2021-11-02
GUANGZHOU YIFANG INFORMATION TECH CO LTD
View PDF6 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Ordinary deep neural network has a large number of parameters and a complex model, so it takes a long time to process audio

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • A Real-time Speech Noise Reduction Method Based on Skip Network
  • A Real-time Speech Noise Reduction Method Based on Skip Network
  • A Real-time Speech Noise Reduction Method Based on Skip Network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0041] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments It is a part of embodiments of the present invention, but not all embodiments.

[0042] like Figure 1-2 As shown, the speech noise reduction method of the specific embodiment of the present invention is based on the multi-layer short-time Fourier transform loss function, including:

[0043] S1: Use the frequency band masking and signal reverberation data enhancement method to construct an audio training set for network training. The frequency band masking is to let the audio pass through the band-stop filter to remove some frequencies in the audio, and the signal reverberation is to continuously improve the audio frequency. At...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention discloses a real-time speech noise reduction method of jumping network. The method is based on a multi-layer short-time Fourier transform loss function, including: constructing an audio training set for network training by using a frequency band mask and a signal reverberation data enhancement method ; Build a jumping Unet lightweight network structure; use the multi-layer short-time Fourier transform loss function to train the model, and use the trained model for noise reduction. The invention adopts the jumping Unet network structure to reduce the weight of the model, and utilizes the loss function based on multi-layer short-time Fourier transform, noise shift, signal reverberation and other data enhancement methods to greatly improve the model's ability to process different noise types. Generalization.

Description

technical field [0001] The invention relates to a speech noise reduction method, in particular to a speech noise reduction method for jumping networks. Background technique [0002] Speech enhancement technology has always been a hot research field. It has great practicability in daily life, such as video conferencing, voice communication, etc. The use of speech enhancement and noise reduction technology can greatly improve the quality of people's voice and video calls. Traditional speech noise reduction methods mainly use spectral subtraction and methods based on statistical models. Such algorithms often cannot achieve good results in dealing with non-stationary noise signals. Traditional methods such as Wiener filtering are difficult to deal with noise signals that are non-stationary or multi-person conversations. The denoising method of the deep neural network that appeared later has improved this, but the processing speed is often slow and it is difficult to play an effe...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Patents(China)
IPC IPC(8): G10L21/0208G10L21/0264G06F17/14G06N3/04G06N3/08
CPCG10L21/0208G10L21/0264G06F17/14G06N3/08G06N3/044G06N3/045
Inventor 黄祥康吴庆耀白剑黄海亮梁瑛玮张海林鲁和平李长杰陈焕然李乐王浩洪行健冷冬丁一
Owner GUANGZHOU YIFANG INFORMATION TECH CO LTD
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap