Supercharge Your Innovation With Domain-Expert AI Agents!

Model training methods and device, electronic equipment and memory medium

A technology of model training and training samples, applied in the field of computer and model training, can solve the problems of poor timeliness and high cost

Active Publication Date: 2020-02-11
BAIDU ONLINE NETWORK TECH (BEIJIBG) CO LTD
View PDF8 Cites 12 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Using the existing training method of the speech synthesis front-end model based on the pre-trained language model, not only the timeliness is poor, but also the cost is high

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
  • Model training methods and device, electronic equipment and memory medium
  • Model training methods and device, electronic equipment and memory medium
  • Model training methods and device, electronic equipment and memory medium

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0046] figure 1 It is a schematic flow chart of the model training method provided in Embodiment 1 of the present application. The method can be executed by a model training device or an electronic device. The device or electronic device can be implemented by software and / or hardware. The device or electronic device can be Integrate in any smart device with network communication function. Such as figure 1 As shown, the model training method may include the following steps:

[0047] S101. In the first stage of fine-tuning training, input each training sample of the first sample type into a shared layer module of the model to be trained.

[0048] In a specific embodiment of the present application, the training process of the speech synthesis front-end model based on the pre-trained language model may only include the first stage and fine-tuning training, or may include the first-stage fine-tuning training and the second-stage fine-tuning training; More stages of fine-tuning ...

Embodiment 2

[0059] image 3 It is a schematic flow chart of the training method of the speech synthesis front-end model based on the pre-trained language model provided in Embodiment 2 of the present application. Such as image 3 As shown, the training method of the speech synthesis front-end model based on the pre-trained language model may include the following steps:

[0060] S301. In the first stage of fine-tuning training, input each training sample of the first sample type into a shared layer module of the model to be trained.

[0061] In a specific embodiment of the present application, the training process of the speech synthesis front-end model based on the pre-trained language model may only include the first stage and fine-tuning training, or may include the first-stage fine-tuning training and the second-stage fine-tuning training; More stages of fine-tuning training can be included. In the first stage of fine-tuning training, the electronic device may input each training s...

Embodiment 3

[0078] Image 6 It is a schematic structural diagram of a training device for a speech synthesis front-end model based on a pre-trained language model provided in Embodiment 3 of the present application. Such as Image 6 As shown, the device 600 includes: a first input module 601, a first training module 602, a second input module 603 and a second training module 604; wherein,

[0079] The first input module 601 is configured to input training samples of each first sample type into the shared layer module of the model to be trained in the first stage of fine-tuning training;

[0080] The first training module 602 is configured to adjust the model parameters in the shared layer module based on training samples of each first sample type;

[0081] The second input module 603 is configured to input the training samples of each first sample type into the task layer module corresponding to each task type of the model to be trained through the shared layer module;

[0082] The sec...

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 model training methods and device, electronic equipment and a memory medium and relates to the field of model training. A specific realization scheme comprises the steps of infine adjustment training of a first stage, inputting each training sample of a first sample type into shared layer modules of a to-be-trained model; adjusting model parameters in the shared layer modules on the basis of each training sample of the first sample type; inputting each training sample of the first sample type into a task layer module corresponding to each task type of the to-be-trainedmodel through utilization of the shared layer modules; extracting own matched training data through utilization of each task layer module; and on the basis of the training data matched with each tasklayer module, adjusting the model parameters in each task layer module. According to the embodiments of the invention, the shared layer modules can be trained uniformly, each task layer module also can be independently trained, single-task performance is improved, and moreover, multi-task training effect is also reserved.

Description

technical field [0001] The present application relates to the field of computer technology, and further relates to the field of model training, especially a model training method, device, electronic equipment and storage medium. Background technique [0002] In the training process of the existing speech synthesis front-end model based on the pre-trained language model, different language training models are used for different types of input data, which are independent of each other. For example, input the data of polyphonic characters into the polyphonic character model, and use the data of polyphonic characters to train the polyphonic character model; input the data of prosody type into the prosody model, and use the data of prosody type to train the prosody model. Using the existing training method of the speech synthesis front-end model based on the pre-trained language model not only has poor timeliness, but also has high cost. Contents of the invention [0003] In v...

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): G10L15/06
CPCG10L15/063G10L2015/0635
Inventor 潘政林聂志朋白洁
Owner BAIDU ONLINE NETWORK TECH (BEIJIBG) CO LTD
Features
  • R&D
  • Intellectual Property
  • Life Sciences
  • Materials
  • Tech Scout
Why Patsnap Eureka
  • Unparalleled Data Quality
  • Higher Quality Content
  • 60% Fewer Hallucinations
Social media
Patsnap Eureka Blog
Learn More

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

© 2025 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com