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

A lightweight speech recognition method for edge computing

An edge computing and speech recognition technology, applied in the field of deep learning, can solve the problem of few applications of lightweight speech recognition network models, and achieve the effect of improving accuracy

Active Publication Date: 2022-02-18
SOUTH CHINA NORMAL UNIVERSITY
View PDF6 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

So there are very few applications in the lightweight speech recognition network model before

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 lightweight speech recognition method for edge computing
  • A lightweight speech recognition method for edge computing
  • A lightweight speech recognition method for edge computing

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0036] The lightweight speech recognition method for edge computing in this embodiment is characterized in that it includes the following steps:

[0037] S1. Construct a lightweight speech recognition network model EdgeRNN for edge computing devices, the EdgeRNN is composed of 1-D CNN and RNN, the 1-D CNN is used to extract spatial advanced features on temporal features, and the RNN For the modeling of speech time series; The EdgeRNN includes an acoustic feature extraction layer, a dense block, a maximum pooling layer, an RNN layer, a self-attention layer and a classification layer;

[0038] In speech recognition, it is first necessary to extract the acoustic features of the original speech. The present invention conducts multiple experiments from the two aspects of accuracy and speed, and finally selects and extracts 128-dimensional mel spectrogram, 12-dimensional delta and 12-dimensional double -delta feature.

[0039] This embodiment takes the four emotions in Session1 of ...

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 lightweight speech recognition method oriented to edge computing, comprising the following steps: S1, constructing a lightweight speech recognition network model EdgeRNN oriented to edge computing devices, the EdgeRNN includes an acoustic feature extraction layer, a dense block , maximum pooling layer, RNN layer, self-attention layer and classification layer; S2, input the original speech audio to EdgeRNN for processing, S21, extract low-level acoustic features from the original audio; S22, design an edge computing Dense blocks; S23, using dense blocks to continuously extract advanced spatial features from low-level acoustic features; S24, using the largest pooling layer to eliminate noise in speech features; S25, fusing the original acoustic features and advanced spatial features Enter the RNN layer to extract time series information; S25, enter the lightweight attention mechanism layer, and obtain the final representation result of the speech level; S26, use the fully connected layer to analyze the category of speech. The invention improves the accuracy and efficiency of speech recognition.

Description

technical field [0001] The invention belongs to the technical field of deep learning, and in particular relates to a lightweight speech recognition method oriented to edge computing. Background technique [0002] In recent years, deep neural networks have achieved remarkable performance in computer vision, natural language processing, speech recognition, etc. But in the field of lightweight deep neural networks, only computer vision has achieved rapid development. This phenomenon is mainly due to two reasons: On the one hand, in the field of natural language processing and speech recognition, time series problems are mainly dealt with, which basically requires the use of recurrent neural networks (RNN), and RNN is computationally intensive. type and require a large amount of storage space, for example, RNN neurons require 8 times the number of weights and multiply-accumulate (MAC) operations of typical CNN units; on the other hand, progress in computer vision has benefited ...

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): G10L15/16G10L15/02G10L25/30G10L25/24G10L25/63
CPCG10L15/16G10L15/02G10L25/30G10L25/24G10L25/63
Inventor 龚征杨顺志叶开魏运根
Owner SOUTH CHINA NORMAL UNIVERSITY
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

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com