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

Voice enhancement method based on double-ear sound source positioning and deep learning in double-ear hearing aid

A technology of deep learning and sound source localization, applied in speech analysis, instruments, etc., can solve the problems of high complexity, inability to improve the speech intelligibility and comfort of deaf people, and not satisfying the real-time performance of digital hearing aids, and achieve real-time Good performance and low power consumption

Active Publication Date: 2019-03-01
BEIJING UNIV OF TECH
View PDF9 Cites 17 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, in the above speech enhancement algorithms based on deep learning, either directly use all the collected data as the input of deep learning, but this method is very complex and does not meet the real-time requirements of digital hearing aids. The characteristic parameters of the data are used as the input of deep learning, but the characteristic parameters extracted by this method cannot describe the characteristics of speech and noise well, and cannot improve the speech intelligibility and comfort of the deaf.
However, the self-learning ability of deep neural network is unmatched by other methods.

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
  • Voice enhancement method based on double-ear sound source positioning and deep learning in double-ear hearing aid
  • Voice enhancement method based on double-ear sound source positioning and deep learning in double-ear hearing aid
  • Voice enhancement method based on double-ear sound source positioning and deep learning in double-ear hearing aid

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0026] In step 1, time-frequency analysis is performed on the input signal of the digital hearing aid using a Gammatone filter that can simulate the working mechanism of the basilar membrane and the auditory nerve in the human auditory system.

[0027] (1) Input signal x of digital hearing aid l (k), x r (k), after Gammatone filter

[0028] Divide the signal frequency band into 64 frequency bands to get the decomposed signal x f,l (k), x f,r (k). where m is the filter order, is the initial phase of the filter, U(t) is the unit step function, and B is the bandwidth. f is the index range of the frequency band from 1 to 64, f c is the center frequency of the filter, ranging from 50Hz to 8kHz, l and r are the left and right ear marks, and k is the number of samples.

[0029] (2) The decomposed signal x obtained by using the Hamming window pair f,l (k), x f,r (k) Perform frame-by-frame windowing. According to the short-term stationary characteristics of the speech signal,...

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

A vice enhancement method based on double-ear sound source positioning and deep learning in a double-ear digital hearing aid belongs to the field of voice signal processing. Firstly a two-stage deep neural network is used for accurately positioning a target voice, and noise in a direction which is different from the direction of target voice is eliminated according to spatial filtering. By means of a deep learning model in which a time delay control bidirectional long-short term memory deep neural network and a classifier are combined, an extracted multi-resolution hearing cepstrum coefficientis used as a characteristic input. Through nonlinear processing capability of deep learning, each time frequency unit of the noise-containing voice is classified to a voice time frequency unit or noise time frequency unit. Finally a voice waveform combining algorithm is used for eliminating the noise in the direction which is same with that of the target voice. The algorithm eliminates the noisein the direction which is different from the direction of the target voice and eliminates the noise in the direction that is same with the target voice, and finally obtains the enhanced voice which satisfies speech intelligibility and comfort of a deaf person. All deep learning models utilize offline training, thereby satisfying a requirement for real-time performance.

Description

technical field [0001] The invention belongs to the technical field of speech signal processing, and relates to two key speech signal processing technologies of target speech localization and speech enhancement in digital hearing aids. Background technique [0002] Hearing impairment is a chronic disease that seriously affects the quality of human life. In the United States, the incidence of hearing loss in the elderly over the age of 65 is about 30% to 40%, in Canada it is 20%, in Europe it is 35%, and in my country it is 35%. And with the increase of age, the incidence rate increases sharply. At present, the total number of elderly people over the age of 60 in the world has reached 600 million, of which my country accounts for nearly 30%, and only 22.28% of the elderly in my country have hearing thresholds within the normal range. The development of hearing aids has brought good news to these hearing-impaired patients. Hearing aid is a device that amplifies sound to sup...

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 Applications(China)
IPC IPC(8): G10L21/0208G10L21/0272G10L21/0232G10L25/30G10L25/24
CPCG10L21/0208G10L21/0232G10L21/0272G10L25/24G10L25/30
Inventor 李如玮李涛孙晓月杨登才潘冬梅张永亚
Owner BEIJING UNIV OF TECH
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