Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

An End-to-End Speech Enhancement Method Based on RefineNet

A speech enhancement and speech signal technology, applied in speech analysis, instruments, etc., can solve the problems of ignoring phase information, enhancing speech clarity and intelligibility, etc.

Active Publication Date: 2021-04-06
UNIV OF ELECTRONICS SCI & TECH OF CHINA
View PDF8 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] In view of the above-mentioned deficiencies in the prior art, the end-to-end speech enhancement method based on RefineNet provided by the present invention solves the problem that the existing speech enhancement method will ignore the phase information and enhance speech clarity and intelligibility.

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
  • An End-to-End Speech Enhancement Method Based on RefineNet
  • An End-to-End Speech Enhancement Method Based on RefineNet
  • An End-to-End Speech Enhancement Method Based on RefineNet

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0102] The specific embodiments of the present invention are described below so that those skilled in the art can understand the present invention, but it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes Within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions and creations using the concept of the present invention are included in the protection list.

[0103] Such as figure 1 As shown, an end-to-end speech enhancement method based on RefineNet includes the following steps:

[0104] S1. Transform the original noisy speech signal into a feature map containing time-frequency information through the TFANet time-frequency analysis network, and input it into the RefineNet network;

[0105] S2. Analyze the feature map through the RefineNet network to determine the feature map c...

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 an end-to-end speech enhancement method based on RefineNet. Firstly, a time-frequency analysis network is constructed to encode and analyze the speech signal, and then the feature mapping from noisy speech to pure speech is learned by using the RefineNet network, and finally the enhanced speech is generated by decoding. Signal. On this basis, we propose an improved method that fuses the evaluation index with the training loss function and a multi-objective fusion learning strategy that takes both STOI and SDR as optimization objectives. In tests under different noise environments and different signal-to-noise ratios, the method proposed by the present invention is significantly better than representative traditional methods, non-end-to-end and end-to-end deep learning methods in terms of STOI, PESQ and SDR, Can better improve the clarity and intelligibility of speech; get better speech enhancement effect.

Description

technical field [0001] The invention belongs to the technical field of speech signal processing, and specifically designs an end-to-end speech enhancement method based on RefineNet. Background technique [0002] The main goal of speech signal enhancement is to extract the original speech signal from noisy speech, and to improve the perceived quality and intelligibility of speech by suppressing or separating noise. application. After several decades of development, many speech enhancement algorithms have been proposed one after another. Classical speech enhancement techniques mainly include spectral subtraction, Wiener filtering, methods based on statistical models, etc. These methods are often based on the assumption that the noise is stationary. The enhancement effect deteriorates sharply in the case of smooth noise. [0003] The rise of deep learning and its successful application in the fields of image classification, speech recognition, and natural speech processing ha...

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): G10L19/02G10L21/0224G10L21/0232G10L25/27
CPCG10L19/02G10L21/0224G10L21/0232G10L25/27
Inventor 蓝天彭川李森刘峤钱宇欣叶文政李萌惠国强吕忆蓝
Owner UNIV OF ELECTRONICS SCI & TECH OF CHINA
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products