Unlock instant, AI-driven research and patent intelligence for your innovation.

Voice super-resolution method based on cyclic frame sequence gating cyclic unit network

A cycle unit and super-resolution technology, applied in the research field of high-resolution speech, can solve difficult problems, achieve the effect of improving quality, increasing calculation cost, and high signal-to-noise ratio

Active Publication Date: 2021-03-26
HARBIN ENG UNIV
View PDF10 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

It is difficult for the hearing-impaired to hear speech at a lower sampling rate

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 super-resolution method based on cyclic frame sequence gating cyclic unit network
  • Voice super-resolution method based on cyclic frame sequence gating cyclic unit network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0023] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0024] The present invention comprises the following steps in the realization process:

[0025] (1) Preprocessing the original voice signal: 1. Pre-emphasizing the original voice signal; 2. Framing the pre-emphasized voice signal;

[0026] (2) It is proposed to construct a CFS-GRU model: ①Construct two kinds of GRUs that increase and decrease the sampling rate of the characteristic parameters per unit time step; ②Combine the two GRUs so that the time step and the characteristic parameters are cross-multiplied Sampling magnification and can be input cyclically to build a CFS-GRU model;

[0027] (3) Complete the speech super-resolution based on the cyclic frame sequence network: ① input the pre-emphasized and frame-divided speech signal into the CFS-GRU model; ② use the SegSNRLoss loss function and use the high-resolution 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

The invention provides a voice super-resolution method of a gated cyclic unit network based on a cyclic frame sequence. The voice super-resolution method comprises the following steps: (1) preprocessing an original voice signal; (2) proposing to construct a CFS-GRU model; and (3) completing the voice super-resolution based on the cyclic frame sequence network. According to the method, a voice signal sequence is directly used as input based on a cyclic structure model established by a GRU, so that the calculation cost is reduced to a great extent, and the method has a better super-resolution effect compared with a traditional method; compared with LSTM, the GRU model has fewer model parameters, and the CFS-GRU model built through the GRU can be trained and converged more quickly. A CFS-GRUmodel trained by using SegSNRLoss as a loss function can be converged more quickly, an output frame sequence can have a high signal-to-noise ratio, and the quality of a super-resolution voice signal is improved.

Description

technical field [0001] The present invention relates to the field of speech super-resolution, in particular to a research on converting low-sampling-rate speech into high-resolution speech without affecting speech content. The present invention proposes a voice super-resolution method based on a cyclic frame sequence gated cyclic unit network, and obtains higher voice super-resolution processing performance with a smaller calculation volume. Background technique [0002] Speech Super-Resolution (SSR), also known as Speech Bandwith Expansion (BWE), aims to improve the quality of speech by upsampling speech through certain technologies. [0003] With the application of deep learning in the direction of speech, people have gradually found that the neural network trained under a certain sampling rate training set has a reduced effect on speech at other sampling rates. For some speech systems, once trained, they cannot dynamically Change the sampling rate of the voice to adapt t...

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): G10L21/003G10L25/18G10L25/24
CPCG10L21/003G10L25/18G10L25/24
Inventor 关键柳友德肖飞扬芦瑶兰宇晨田左王恺瀚谢明杰董喆
Owner HARBIN ENG UNIV