Unlock instant, AI-driven research and patent intelligence for your innovation.

Voice recognition method, system and device and medium

A speech recognition and speech technology, applied in speech recognition, speech analysis, instruments, etc., to achieve cross-recognition in multiple fields, reduce waste of computing resources, and reduce labor costs

Active Publication Date: 2021-08-24
SHANGHAI QIYUE INFORMATION TECH CO LTD
View PDF5 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004]Aiming at the above-mentioned defects in the prior art, the present invention provides a technical solution of a voice recognition method, system, device and medium, aiming at solving how to pass depth-based The learned dynamic language model implements the technical problem of adapting to different domain recognition services; further, it can also solve the problem of providing recognition services for different domains through deep learning-based dynamic language models based on domain information and / or judging the domain information of existing dialogues. Reduce the problem of excessive server deployment; further, it can also solve technical problems such as reducing the waste of computing resources, reducing labor costs, and effectively providing recognition results in different fields for long dialogue recognition

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
  • Voice recognition method, system and device and medium
  • Voice recognition method, system and device and medium
  • Voice recognition method, system and device and medium

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0043] Bonded below image 3 The prior flow diagram of one embodiment of the speech recognition method according to the present invention will be described with reference to the speech recognition implementation process of the present invention.

[0044] Step S110, according to the speech recognition service request, the acoustic characteristics of the voice to be identified and the domain information corresponding to the acoustic characteristics are obtained.

[0045] In one embodiment, the characteristic of the current sentence of the current sentence can be converted to the current sentence of the current sentence in the primary voice recognition service request, and the characteristics of the current sentence are extracted. Vector); According to the domain of the corpus, it is determined that the domain belonging to the current sentence, as the domain information corresponding to the acoustic characteristics.

[0046] Here, the result obtained by the acoustic model is a word, w...

Embodiment 2

[0080] The following will be combined Figure 4 A structural block diagram of one embodiment of the speech recognition system according to the present invention Figure 5 to 9 The identification service deployment principle, long dialog exemplary, and model construction, model / module connection, model map of model / module, model map of model / module, model map of model / module, and model map of the model / module are shown in accordance with the application of the model / module. Further explanation.

[0081] In one embodiment, the system can include: configured at the judge 410 connected to the judgment device 410, configured in the rear end language identification service device 420; the determination device 410 is used in a service request According to the field classification of the corpus, it is determined that the field belonging to the current sentence of the current sentence to be identified, as a domain information corresponding to the current, the language identificat...

Embodiment 3

[0120] The following will be combined Figure 7 to 11 Several schematic diagrams showing the application scenarius deploying and performing a model in line deployment and performing speech recognition in the specific application of the present invention, the technical solution of the present invention is constructed in actual application scenarios, and deploy speech recognition services, The process of identifying is further illustrated. It is only a specific application example, not the implementation of the present invention.

[0121] First, set the label corresponding to the corpus in various fields, for example: 'Area I' is 0, 'Area II' is 1, and so on.

[0122] Second, establish a corbanity change determination module, corpus class classification module, domain information decoding module, and language models that lead the domain. Figure 7 to 9 Indicated. Wherein represents the status of the Time of the feature extractor, the state is the state vectuation of the dialog, and th...

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 relates to the field of voice recognition, and provides a voice recognition method, system and device and a medium for overcoming the defects that in existing voice recognition, computing resources are wasted, hot switching cannot be achieved among multiple models, and a single domain model is not suitable for long dialogue recognition. The objective of the invention is to solve the technical problem of how to provide voice recognition services in different fields according to field information based on a dynamic language model of deep learning. Therefore, according to the method, the built voice recognition model is combined with the utilization of corpus field information in the prediction process, the voice recognition service suitable for effective hot switching and long dialogues in multiple fields is provided, the existing voice recognition service performance is improved, resource waste is effectively reduced, the method is suitable for correct recognition of cross and long dialogues in different fields, recognition of hot switching is achieved and moreover, the method is simple to implement, easy to operate, low in cost and high in efficiency.

Description

Technical field [0001] The present invention relates to the field of speech recognition, and in particular, a speech recognition method, a system, a device, and a medium are involved. Background [0002] In speech recognition, its main process is generally identified by an acoustic model, and then the corresponding text is translated by the language model according to the acoustic characteristics. Due to the presence of the homonym-like line, different language models can be trained to adapt to the specific field scenario. like figure 1 As shown: The speech to be identified into the unit of the corresponding acoustic model (pinyin, tone sequence number, etc.), extracting acoustic characteristics by acoustic model, such as identifying the acoustic model "jian3 yi4 gong1 zuo4", and then, Translation can be "simple work" with a generic language model, and an epidemic prevention language model translates can be "quarantine work". Thus, in order to provide support for many fields, man...

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): G10L15/14G10L15/26G10L15/18
CPCG10L15/142G10L15/26G10L15/18
Inventor 白蒙蒙
Owner SHANGHAI QIYUE INFORMATION TECH CO LTD