Small training set optimized decoding network based voice wake-up implementation method

A decoding network and voice wake-up technology, which is applied in voice analysis, voice recognition, instruments, etc., can solve the problems of low false wake-up effect, achieve the effects of simplifying complexity, improving adaptability, and reducing model volume

Inactive Publication Date: 2019-07-19
武汉水象电子科技有限公司
View PDF8 Cites 20 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0008] The technical problem to be solved by the present invention is to overcome the existing overall idea of ​​voice wake-up or the method of voice recognition using a large-scale vocabulary, which needs to use a large amount of training corpus to achieve good wake-up and low false wake-up effect defects, providing A voice wake-up implementation method based on a small training set and an optimized decoder that simplifies the algorithm implementation and improves the operation efficiency under the same wake-up rate, thereby realizing the purpose of voice wake-up and easy transplantation

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
  • Small training set optimized decoding network based voice wake-up implementation method
  • Small training set optimized decoding network based voice wake-up implementation method
  • Small training set optimized decoding network based voice wake-up implementation method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0031] A method for implementing voice wakeup based on a small training set optimized decoding network, which is characterized in that it includes the following steps:

[0032] S1 extracts speech eigenfeatures

[0033] According to the analysis of the stability and correlation of the wake-up words, the time window is designed to obtain the frame characteristic signal. The time window design here involves the window length, the shape, the amplitude of each point, and the weight between adjacent frames. Obtain the feature vector that distinguishes arousal words and non-arousal words;

[0034] S2 combines feature vectors to obtain feature phoneme alignment files

[0035] Select the time window according to the distribution of the phoneme of the wake-up word, classify the mapping of features and phonemes, and obtain labeled acoustic data; the alignment algorithm between features and phonemes in this step is mainly obtained by using a context-sensitive three-factor phoneme model. Accordin...

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 small training set optimized decoding network based voice wake-up implementation method. The method includes steps: S1, extracting voice intrinsic characteristics to obtain characteristic vectors enabling evident distinctiveness of wake-up words and non-wake-up words; S2, acquiring characteristic phoneme alignment files according to the characteristic vectors, selecting time windows according to wake-up phoneme distribution, and classifying characteristic and phoneme mappings to obtain acoustic data with tags; S3, calculating a frame-by-frame posterior probability model according to the acoustic data with the tags; S4, acquiring a phoneme-level posterior probability confidence coefficient calculation network by virtue of the obtained acoustic probability model; S5, establishing a wake-up word reconfirmation network. By simple model training strategies and steps of decoding network optimizing and the like, functions of voice wake-up and the like can be easily implemented on universal processors such as arm and dsp.

Description

Technical field [0001] The invention relates to a method for implementing voice wake-up based on a small training set optimized decoding network. A lightweight model refers to a small training data set, a trained model occupies a small space, and can be used on a mobile terminal with few hardware resources. The post-decoding network is optimized to reduce the offline false wake rate without increasing the complexity of the algorithm. Background technique [0002] Voice is the most convenient and efficient way for humans to communicate and communicate with each other. It is a dream goal for humans to allow machines to understand voices and perform related operations in accordance with human instructions. As a result, speech recognition technology came into being. Voice recognition technology is currently an important means of human-computer interaction, and voice wake-up is an important entry point for human-computer interaction. The smart voice device is normally in the standby...

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): G10L15/02G10L15/06G10L15/08G10L15/16G10L15/187G10L15/26
CPCG10L15/02G10L15/063G10L15/083G10L15/16G10L15/187G10L15/26G10L2015/025
Inventor 赵升
Owner 武汉水象电子科技有限公司
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