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

Speech recognition method adopting rectified deep auto-encoder network

A technology of self-encoding network and speech recognition, which is applied in the field of corrected linear deep self-encoding network speech recognition, to achieve the effect of optimizing model training algorithm, improving training efficiency, enhancing recognition performance and anti-noise ability

Inactive Publication Date: 2017-05-31
TAIYUAN UNIV OF TECH
View PDF8 Cites 33 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0006] The specific technical problem to be solved by the present invention is: how to establish a deep learning speech recognition system method in a noisy environment, and obtain good anti-noise performance

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
  • Speech recognition method adopting rectified deep auto-encoder network
  • Speech recognition method adopting rectified deep auto-encoder network
  • Speech recognition method adopting rectified deep auto-encoder network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0033] The specific embodiment of the present invention is further described as follows.

[0034] Since autoencoders can be quickly trained by an unsupervised layer-by-layer greedy training algorithm, this approach bypasses the high complexity of directly training a DAE as a whole and reduces it to training multiple autoencoders question. After training in this way, the network is fine-tuned through the traditional global learning algorithm, and the maximum likelihood function is used as the objective function to make the network optimal. This learning algorithm is essentially equivalent to pre-training the layer-by-layer autoencoder to obtain a better initial parameter value for the model, and then further training and optimizing the network by using a small number of traditional learning algorithms.

[0035] During training, for the DAE model with the traditional Sigmoid function as the activation function, when using the backpropagation algorithm to transfer the error from...

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 speech recognition method adopting a rectified deep auto-encoder network. The method comprises the following steps: first, carrying out training on the deep auto-encoder network by adopting a rectification linear unit as an activation function which replaces the traditional Sigmoid function, then, introducing L2 regularization for optimizing the overfitting problem easily occurring during the deep model training process, and finally, carrying out the layer-by-layer greedy and unsupervised 'pretraining' and supervised 'fine tuning', thus obtaining weights with character representation for speech recognition. According to the method provided by the invention, the strong capacity of learning substantive characteristics of data sets from the minority of samples of the deep neural network is fully utilized, thus the problems of gradient disappear and overfitting during the training process are solved, and the recognition precision of the system under the noise environment is improved.

Description

technical field [0001] The invention relates to a method for recognizing speech by using a deep learning network model, which belongs to the field of speech signal processing, and in particular to a method for correcting linear deep self-encoding network speech recognition. Background technique [0002] Due to the large difference between the theoretical assumptions of traditional speech recognition methods and the actual situation, it is difficult to achieve the expected performance in practical applications, and a breakthrough in theory is urgently needed. Deep learning is currently an important machine learning theory for big data, and it has a wide range of applications in speech, image, text and other fields. At present, deep learning algorithms have developed to a certain extent in speech recognition. It simulates the principle of human neuron activity, and has the ability of self-learning, association, comparison, reasoning and generalization. a new way. However, as...

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): G10L15/06G10L15/02
CPCG10L15/063G10L15/02
Inventor 黄丽霞张雪英孙颖娄英丹
Owner TAIYUAN UNIV OF TECH
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