A language recognition method and system based on an adaptive central anchor

A language recognition and self-adaptive technology, applied in neural learning methods, natural language translation, special data processing applications, etc., can solve the problem of low accuracy and reliability of deep learning methods, improve accuracy and enhance feature expression capabilities. , the effect of improving the accuracy

Active Publication Date: 2021-12-10
北京快鱼电子股份公司
View PDF14 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] At present, the deep learning method used in language recognition model training can be divided into front-end, coding layer and back-end. The front-end represents the input features, the coding layer represents the backbone network, the back-end represents the loss function, and the back-end can be divided into measurement loss and classification. loss, the accuracy and reliability of the deep learning method using the backbone network combined with the metric loss or the backbone network combined with the classification loss are not high, and still need to be further improved

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 language recognition method and system based on an adaptive central anchor
  • A language recognition method and system based on an adaptive central anchor
  • A language recognition method and system based on an adaptive central anchor

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0053] Such as figure 1 As shown, this embodiment discloses a language recognition method based on an adaptive central anchor, including the following steps:

[0054] S100: Construct a language data set, and preprocess the voice data in the language data set;

[0055] The language data set (vocal database) adopts the existing language database. The sampling rate of the language data set is 16000HZ, including 35 languages, and 10 hours of voice data are randomly selected for each language;

[0056] Taking one of the language datasets as an example, the speech data in the language dataset is preprocessed, including:

[0057] S110: Extract all the voice data of the same language in the language data set for splicing; splicing all the voices extracted from the language data set is recorded as S src ;

[0058] S120: Calculate the continuous silent segment in the speech data after splicing, if the continuous silent segment is greater than the set threshold, remove the silent part ...

Embodiment 2

[0135] Such as Image 6 As shown, this embodiment discloses a language recognition system based on an adaptive central anchor, including a language data set construction module, an enhancement processing module, a feature extraction module, a first training module, a second training module, and a language recognition module;

[0136] The language data set building block is used to construct the language data set;

[0137] The enhanced processing module is used to perform enhanced processing on the voice data in the language data set;

[0138] Feature extraction module is used for extracting the feature of the speech data after described enhancement processing, generates feature data set;

[0139] The first training module is used to construct a deep neural backbone network, and train the deep neural backbone network in a supervised learning manner based on a classification loss function;

[0140] The second training module is used to further train the deep neural backbone ne...

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 language recognition method and system based on an adaptive central anchor, which includes constructing a language data set; performing enhanced processing on speech data in the language data set; extracting features of the enhanced speech data to generate a feature data set; Constructing a deep neural backbone network, and using a supervised learning method to train the deep neural backbone network based on the classification loss function; based on the metric loss function, using an adaptive central anchor method to further train the deep neural backbone network; based on the trained deep neural backbone network Carry out language recognition; this method expands the data set through data enhancement, and also increases the robustness of the data set; uses a multi-scale fusion residual neural network to enhance the feature expression ability of the backbone network, and uses classification loss and measurement successively The loss further improves the accuracy of recognition.

Description

technical field [0001] The invention relates to the technical field of speech processing, in particular to a language recognition method and system based on an adaptive central anchor. Background technique [0002] With the development of speech signal processing and artificial intelligence, especially the rapid development of deep learning in recent years, the reliability of language recognition technology has been further improved. Many intelligent voice assistants can use automatic language recognition technology to infer the language used by users; As a preprocessing part of many speech processing tasks, language recognition technology has a wide range of applications in the fields of multilingual speech recognition, cross-lingual communication, and machine translation. [0003] At present, the deep learning method used in language recognition model training can be divided into front-end, coding layer and back-end. The front-end represents the input features, the coding ...

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
IPC IPC(8): G06F16/33G06F16/35G06F40/56G06N3/04G06N3/08
CPCG06F16/3343G06F16/35G06F40/56G06N3/04G06N3/08
Inventor 马杰
Owner 北京快鱼电子股份公司
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