Voice noise reduction method and device, equipment and medium

A speech noise reduction and speech technology, applied in the field of machine learning, can solve the problems of high system requirements, difficult to control the model scale, occupying a large amount of resources, etc., to achieve high noise reduction efficiency, reduce computational complexity, and reduce the amount of computation.

Inactive Publication Date: 2020-07-17
ZHEJIANG UTRY INFORMATION TECH CO LTD
View PDF11 Cites 8 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, the deep learning model requires a lot of resources, resulting in a waste of resources
For example, some methods of noise suppression use layers with thousands of neurons and tens of millions of weights to perform noise suppression, resulting in huge computational costs for the model to run the network, the scale of the model itself is difficult to control, and the need to store data Thousands of lines of code and tens of megabytes of neuron weights; when using these methods for speech noise reduction, the requirements for the system are high and the amount of calculation is large

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 noise reduction method and device, equipment and medium
  • Voice noise reduction method and device, equipment and medium
  • Voice noise reduction method and device, equipment and medium

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0050] Embodiment 1 provides a speech noise reduction method, aiming to realize speech noise reduction through frequency band gain coefficients.

[0051] Please refer to figure 1 Shown, a kind of speech denoising method comprises the following steps:

[0052] S110, acquiring voice data;

[0053] In order to realize real-time speech noise reduction, a frame of speech data is collected every 10ms in this embodiment, and the sampling rate is 48kHz.

[0054] Of course, in the case of non-real-time speech noise reduction, it is only necessary to divide the speech data into frames and perform noise reduction processing on the speech data frame by frame.

[0055] The source of the voice data is, for example, a voice data stream in a noisy environment obtained by a robot microphone, and this embodiment does not limit the specific source.

[0056] S120. Perform preprocessing on the voice data, and extract multi-dimensional features of the preprocessed voice data;

[0057] The above...

Embodiment 2

[0089] Embodiment 2 mainly explains and explains the construction process of the preset speech noise reduction model, and aims to maintain all necessary basic signal processing without neural network simulation by combining traditional signal processing methods and deep learning methods of recurrent neural networks , and learn all the work that needs repeated parameter adjustment through the neural network to realize the construction of the speech noise reduction model.

[0090] Compared with other deep learning neural networks, the recurrent neural network (RNN) adds time series and can be better applied in the field of speech processing technology; therefore, this embodiment selects the recurrent neural network as the preset speech noise reduction model.

[0091] Please refer to image 3 As shown, the training process of the preset speech noise reduction model includes the following steps:

[0092] S210. Obtain a pre-built cyclic neural network, which includes 3 fully conne...

Embodiment 3

[0111] Embodiment 3 discloses a device corresponding to the speech noise reduction method of the above embodiment, which is the virtual device structure of the above embodiment, please refer to Figure 4 shown, including:

[0112] Obtaining module 310, for obtaining voice data;

[0113] The filtering module 320 is used to preprocess the voice data, extract the multi-dimensional features and voice activity detection parameters of the preprocessed voice data; when the voice activity detection parameter is 1, divide the voice data into For several frequency bands, filter the noise data in the frequency band according to the frequency band gain coefficient; when the voice activity detection parameter is 0, set the frequency band gain coefficient to 0, and filter the noise data in the frequency band;

[0114] The output module 330 is configured to restore the filtered voice data into a voice data stream, and output the voice data stream.

[0115] Preferably, the preset speech noi...

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 voice noise reduction method and relates to the field of machine learning, and the method is used for solving the problems of large calculation amount and large resource occupation of the existing voice noise reduction. The method comprises the following steps of obtaining voice data; preprocessing the voice data, and extracting multi-dimensional features of the preprocessed voice data; inputting the multi-dimensional features into a preset voice noise reduction model to obtain a frequency band gain coefficient; dividing the voice data into a plurality of frequency bands, and filtering noise data in the frequency bands according to the frequency band gain coefficient; and recovering the filtered voice data into a voice data stream and outputting the voice data stream. The invention further discloses a voice noise reduction device, electronic equipment and a computer storage medium. According to the invention, the frequency band gain coefficient is calculated to realize voice noise reduction.

Description

technical field [0001] The present invention relates to the technical field of machine learning, in particular to a speech noise reduction method, device, equipment and medium. Background technique [0002] Noise suppression has been a topic of high interest since the 1970s. Traditional noise suppression algorithms require a noise spectrum estimator. The noise spectrum estimator itself is driven by a voice activity detector (VAD) or a similar algorithm. Each component of the noise spectrum estimator requires an accurate estimator, which requires high precision. It requires a lot of manual parameter adjustment work, and the efficiency is low. As long as one parameter is not accurate enough, it will easily affect the noise reduction effect. [0003] Existing technologies have begun to use deep learning technology for noise suppression. The common practice is to introduce deep neural networks into engineering problems. This method is called end-to-end—neurons receive and trans...

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): G10L21/0208G10L21/0232G10L21/0264
CPCG10L21/0208G10L21/0232G10L21/0264
Inventor 丁大为王哲嵇望
Owner ZHEJIANG UTRY 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