WO2020213785A1 - Système pour générer automatiquement des phrases à base de texte sur la base de l'apprentissage profond afin d'obtenir une amélioration liée à l'infinité de modèles de prononciation - Google Patents

Système pour générer automatiquement des phrases à base de texte sur la base de l'apprentissage profond afin d'obtenir une amélioration liée à l'infinité de modèles de prononciation Download PDF

Info

Publication number
WO2020213785A1
WO2020213785A1 PCT/KR2019/006337 KR2019006337W WO2020213785A1 WO 2020213785 A1 WO2020213785 A1 WO 2020213785A1 KR 2019006337 W KR2019006337 W KR 2019006337W WO 2020213785 A1 WO2020213785 A1 WO 2020213785A1
Authority
WO
WIPO (PCT)
Prior art keywords
deep learning
intention
unit
text
output
Prior art date
Application number
PCT/KR2019/006337
Other languages
English (en)
Korean (ko)
Inventor
윤종성
김영준
김위백
양형원
이인구
김홍순
송민규
Original Assignee
미디어젠 주식회사
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 미디어젠 주식회사 filed Critical 미디어젠 주식회사
Publication of WO2020213785A1 publication Critical patent/WO2020213785A1/fr

Links

Images

Classifications

    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
    • G10L13/00Speech synthesis; Text to speech systems
    • G10L13/08Text analysis or generation of parameters for speech synthesis out of text, e.g. grapheme to phoneme translation, prosody generation or stress or intonation determination
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00Speech recognition
    • G10L15/04Segmentation; Word boundary detection
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00Speech recognition
    • G10L15/06Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
    • G10L15/063Training

Definitions

  • the present invention relates to a deep learning-based text sentence automatic generation system for improving the infinity of speech patterns, and more particularly, by using speech patterns of texts and intention information of texts, which are deep learning learning data collected from native speakers. Perform deep learning training, derive deep learning model information for each intention, which is the result of the execution, and automatically generate and output a number of text sentences with specific intentions by applying specific intention information to the derived deep learning model information for each intention.
  • the present invention relates to a deep learning-based text sentence automatic generation system for improving the infinity of speech patterns.
  • deep learning training is performed using speech patterns of texts, which are deep learning learning data collected from native speakers, and common intention information of texts, and a deep learning model for each intention as a result of the execution
  • speech patterns of texts which are deep learning learning data collected from native speakers, and common intention information of texts
  • a deep learning model for each intention as a result of the execution
  • We propose a deep learning-based natural language speech pattern text sentence automatic generation system to automatically generate and output multiple text sentences with specific intentions by deriving information and applying specific intention information to the derived deep learning model information for each intention. was done.
  • a first object of the present invention is to perform deep learning training using speech patterns of native-speaking texts and intention information of the collected deep learning learning data. It aims to generate deep learning model information by intention.
  • a second object of the present invention is to automatically generate text sentences with specific intentions by reflecting specific intention information to the generated deep learning model information for each intention.
  • a third object of the present invention is to utilize the generated text sentences as text for evaluation of NLU (Natural Language Understanding) of an automatic voice recognition evaluation system.
  • NLU Natural Language Understanding
  • a deep learning-based text sentence automatic generation system for improving the infinity of speech patterns
  • Deep learning model information including a plurality of deep learning models is generated by receiving various texts, which are deep learning learning data having speech pattern information and intention information, and generated deep learning model information for each intention.
  • An output intention input unit 200 for providing the intention type information necessary for the text sentence generation by the test sentence automatic generation unit 200 to the test sentence automatic generation unit 200,
  • a text sentence corresponding to the intention type information is generated, output, and output. It characterized in that it comprises a test sentence automatic generation unit 300 for outputting the confidence value for the text sentence.
  • a deep learning-based text sentence automatic generation system for improving the infinity of speech patterns, uses previously collected data (various speech patterns with specific intentions collected for use as deep learning learning data).
  • the user's potential speech pattern that is, by automatically generating and outputting a new text sentence (text sentence with the same intention as the speech pattern of the collected text but different speech pattern) different from the collected data based on the It provides an effect that can improve the infinity of the potential firing pattern.
  • FIG. 1 is an overall configuration diagram schematically showing a system for automatically generating text sentences based on deep learning for improving the infinity of speech patterns according to a first embodiment of the present invention.
  • FIG. 2 is a block diagram of a deep learning modeling unit 100 of a system for automatically generating text sentences based on deep learning for improving the infinity of speech patterns according to the first embodiment of the present invention.
  • FIG. 3 is a diagram illustrating the basic structure of an RNN of a system for automatically generating text sentences based on deep learning for improving the infinity of speech patterns according to the first embodiment of the present invention.
  • FIG. 4 is a block diagram of an automatic test sentence generation unit 300 of a system for automatically generating text sentences based on deep learning for improving the infinity of speech patterns according to the first embodiment of the present invention.
  • FIG. 5 is a structural diagram of automatic text sentence generation generated by the test sentence automatic generation unit 300 of the deep learning-based automatic text sentence generation system for improving the infinity of speech patterns according to the first embodiment of the present invention.
  • FIG. 6 is an exemplary diagram of an RNN algorithm structure and training parameters of a system for automatically generating text sentences based on deep learning for improving the infinity of speech patterns according to the first embodiment of the present invention.
  • FIG. 7 is a diagram illustrating an example of training an RNN model of a system for automatically generating text sentences based on deep learning for improving the infinity of speech patterns according to the first embodiment of the present invention.
  • FIG. 8 is a diagram illustrating an example of an automatic generation result test generated by a deep learning-based text sentence automatic generation system for improving the infinity of a speech pattern according to the first embodiment of the present invention.
  • FIG. 9 is a diagram illustrating an example of outputting text sentences for each intention type generated by the deep learning-based automatic text sentence generation system for improving the infinity of speech patterns according to the first embodiment of the present invention.
  • first and second may be used to describe various elements, but the elements may not be limited by terms.
  • a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element.
  • Deep learning model information including a plurality of deep learning models is generated by receiving various texts, which are deep learning learning data having speech pattern information and intention information, and generated deep learning model information for each intention.
  • An output intention input unit 200 for providing the intention type information necessary for the text sentence generation by the test sentence automatic generation unit 200 to the test sentence automatic generation unit 200,
  • a text sentence corresponding to the intention type information is generated, output, and output. It characterized in that it is configured to include a test sentence automatic generation unit 300 for outputting the confidence value for the text sentence.
  • a model text corpus unit 110 for receiving a plurality of texts having a specific speech pattern, which is a deep learning learning material, and providing the input texts to the training modeling unit 130;
  • model text corpus unit 110 To receive the intention of the texts input to the model text corpus unit 110, tag the input corresponding intention to the text, and provide intention tagging information for each text tagged with the intention for each text to the training modeling unit 130 A model intention input unit 120;
  • Deep learning model for each intention by performing deep learning training using a plurality of texts having a specific speech pattern provided by the model text corpus unit 110 and intention tagging information for each text provided by the model intention input unit 120 And a training modeling unit 130 for generating information and providing the generated deep learning model information for each intention to the automatic test sentence generation unit 300.
  • test sentence automatic generation unit 300 In addition, the test sentence automatic generation unit 300,
  • an output parameter providing module 311 for adjusting a parameter related to an output option when outputting the generated text sentence.
  • NLU Natural Language Understanding
  • FIG. 1 is an overall configuration diagram schematically showing a system for automatically generating text sentences based on deep learning for improving the infinity of speech patterns according to a first embodiment of the present invention.
  • the deep learning-based text sentence automatic generation system 1000 for improving the infinity of the speech pattern according to the present invention includes a deep learning modeling unit 100, an output intention input unit 200, and an automatic test sentence. It will be configured to include the generation unit 300.
  • the deep learning modeling unit 100 receives various texts, which are deep learning learning data having speech pattern information and intention information, and performs deep learning training to generate deep learning model information for each intention including a plurality of deep learning models. , It is a means for providing the generated deep learning model information for each intention to the test sentence automatic generation unit 300.
  • the deep learning learning material input to the deep learning modeling unit 100 is '1.
  • I want to know who Sherlock Holmes is', etc. utterance patterns of each input text
  • the information is "Panda_Restaurant_Find”, "I_Sherlock Holmes_Who is_I want to know”, and the intention information of each input text is'Restaurant Search' and'People Search'.
  • the deep learning modeling unit 100 performs deep learning training by receiving texts, which are a plurality of deep learning learning data having speech patterns and intention information, and deep learning model information for each intention (a plurality of deep learning models). Included) is provided to the test sentence automatic generation unit 300.
  • the output intention input unit 200 is a means for providing the intention type information required when the text sentence is generated by the test sentence automatic generating unit 300 to the test sentence automatic generating unit 300.
  • the automatic test sentence generation unit 300 by providing an intention type of'find a restaurant' to the automatic test sentence generation unit 300, the automatic test sentence generation unit 300 generates and outputs text sentences of various speech patterns with the intention of'finding a restaurant'. To do it.
  • the test sentence automatic generation unit 300 corresponds to intention type information by using deep learning model information for each intention provided by the deep learning modeling unit 100 and intention type information provided by the output intention input unit 200. This is a configuration that generates and outputs a text sentence that is written and outputs a confidence value for the text sentence that is output.
  • the intention type information provided by the output intention input unit 200 is'Find a restaurant'
  • the deep learning model information for each intention provided by the deep learning modeling unit 100 is a deep 'Restaurant search' with various speech patterns using the learning model information is to create and output text sentences of intention.
  • FIG. 2 is a block diagram of a deep learning modeling unit 100 of an automatic text sentence generation system with improved speech pattern infinity based on deep learning according to the first embodiment of the present invention.
  • the deep learning modeling unit 100 includes a model text corpus unit 110, a model intention input unit 120, and a training modeling unit 130.
  • the model text corpus unit 110 receives a plurality of texts having a specific speech pattern, which is a deep learning learning material, and provides the input texts to the training modeling unit 130.
  • a plurality of texts having a specific speech pattern to be input have respective intentions.
  • a plurality of texts having a specific speech pattern to be input are 1.Where is the ⁇ Panda Restaurant>?, 2.I want to look for ⁇ Panda Restaurant>, 3.Could you please find ⁇ Panda Restaurant>? If so, the above texts are texts intended to find a restaurant called'Panda'. That is, the texts are texts intended to be'Find Restaurant'.
  • texts having various speech patterns corresponding to various intentions such as music search, address search, weather search, stock search, etc. are input through the model text corpus unit 110.
  • the input of a plurality of texts having a specific speech pattern, which is a deep learning learning data, into the model text corpus unit 110 is a manual input of the collected texts by a human and automatic input of the texts collected by the text collecting robot by the robot. It characterized in that it comprises a.
  • model intention input unit 120 receives the intention of the texts input to the model text corpus unit 110, tags the input corresponding intention to the corresponding text, and stores intention tagging information for each text whose intention is tagged for each text. It characterized in that it is provided to the training modeling unit 130.
  • model intention input unit 120 tags the intention of "Find a restaurant” in the input text "Where is the ⁇ Panda Restaurant>?", and the intention tag for each text tagged with the intention information of "Find a restaurant" Information (eg, "Where is the ⁇ Panda Restaurant>? _Find a restaurant") is provided to the training modeling unit 300.
  • the training modeling unit 130 uses a plurality of texts having a specific speech pattern provided by the model text corpus unit 110 and intention tagging information for each text provided by the model intention input unit 120 It performs a function of generating deep learning model information for each intention by performing learning training, and providing the generated deep learning model information for each intention to the test sentence automatic generation unit 300.
  • the training modeling unit 130 may utilize the conventional recurrent neural network (RNN) algorithm shown in FIG. 3 to generate deep learning model information for each intention.
  • RNN recurrent neural network
  • the RNN Recurrent Neural Network
  • the training modeling unit 130 is characterized in that it is configured to further include a modeling parameter providing module 131 that adjusts parameters related to modeling during deep learning modeling.
  • the modeling parameter providing module ( 131) provides modeling parameters related to these modeling options to the training modeling unit 130 so that modeling can be performed using the provided parameters.
  • FIG. 4 is a block diagram of an automatic test sentence generation unit 300 of a system for automatically generating text sentences based on deep learning for improving the infinity of speech patterns according to the first embodiment of the present invention.
  • FIG. 5 is a structural diagram of automatic text sentence generation generated by the test sentence automatic generation unit 300 of the deep learning-based automatic text sentence generation system for improving the infinity of speech patterns according to the first embodiment of the present invention.
  • the automatic test sentence generation unit 300 includes a calculation model unit 310, a corpus output unit 320, and a confidence value output unit 330.
  • the calculation model unit 310 uses the intention-specific deep learning model information 50 provided from the deep learning modeling unit 100 and the intention type information provided through the output intention input unit 200 to correspond to the intention type information. Text sentences of various speech patterns with intention are automatically generated.
  • 50,000 native-speaking texts are input to the deep learning modeling unit 100, and the input texts are divided into 150 intention types, and 50,000 native-speaking texts have specific speech patterns for each of 150 intention types
  • the training modeling unit 130 of the deep learning modeling unit 100 generates deep learning model information 50 for each intention by using a recurrent neural network (RNN) algorithm as described above and provides it to the computational model unit 310.
  • RNN recurrent neural network
  • the calculation model unit 310 which received deep learning model information 50 for each intention from the deep learning modeling unit 100 and intention type information from the output intention input unit 200, has an intention corresponding to the intention type information. It is to automatically generate text sentences with various speech patterns.
  • the calculation model unit 310 receiving deep learning model information 50 for each intention from the deep learning modeling unit 100 and intention type information from the output intention input unit 200 is an example.
  • text sentences with various speech patterns, which are output values, are automatically generated by using a representative RNN algorithm called GRU Cell (Gated Recurrent Unit).
  • GRU Cell Gated Recurrent Unit
  • the input data ⁇ STR> corresponds to the intention type information provided from the output intention input unit 200.
  • the intention type information is the intention of "Find a restaurant”
  • the input data ⁇ STR> is S1 If you pass Find, the output value of Find comes out, if you pass the input data Find through S2, you get an output value of me. If you pass me through S3, you get an output value called POI (means the destination you are looking for), and POI (means the destination you are looking for) is S4.
  • a period that ends the sentence is outputted, so a text sentence with a specific speech pattern with the intention of finding a restaurant called "Find me POI.” is created.
  • the corpus output unit 320 outputs a text sentence automatically generated by the calculation model unit 310.
  • the text sentences automatically generated by the computational modeling unit 310 are text sentences having the same intention as the text input to the deep learning modeling unit 100 as deep learning learning data, but having different speech patterns.
  • the text entered into the deep learning modeling unit 100 as a learning material for deep learning is 1. Where is the ⁇ Restaurant>?, 2. I want to look for ⁇ Restaurant>., 3. Can you please find ⁇ Restaurant>? If so, the text sentence automatically generated by the computational model unit 310 is 4. Find ⁇ Restaurant>., 5. I want to find ⁇ Restaurant>., 6. Can you please looking for ⁇ Restaurant>., 7. Where is the ⁇ Restaurant>. As such, text sentences with the same intention as "Find a restaurant" but different speech patterns are generated, and the generated text sentences are output through the corpus output unit 320.
  • the text sentence generated and output by the test sentence automatic generation unit 300 has the same intention type as the texts input to the deep learning modeling unit 100 to be used as deep learning learning data, but has a different speech pattern. Characterized in that it is a sentence.
  • the confidence value output unit 330 outputs a confidence value indicating how similar the text sentence output through the corpus output unit 320 is to the intention of the intention type information input through the output intention input unit 200. do.
  • a function of outputting a confidence value, that is, a probability value, that can confirm how similar the generated and output text sentence is to the intention of the intention type information input through the output intention input unit 200 is performed.
  • Calculation of the confidence value (probability value) between the output text sentence and the intention type can generally use a probabilistic distance value, and a detailed description of this can be found in Korean Patent No. 10-1890704,'Speech Recognition' It is described in detail in'a simple message output device and output method using language modeling' and'A device for improving language comprehension performance based on domain extraction and a method for improving performance' in Korean Patent No. 10-1913191.
  • the technology to calculate the confidence value is a technology mainly used in statistics or speech recognition technology, and it is a technology commonly applied in topic models, opinion mining, text summarization, data analysis, and public opinion surveys, so the principle of calculating the confidence value is specifically explained. It is obvious that even if not, those skilled in the art can fully understand the above meaning.
  • the calculation model unit 310 is characterized in that it is configured to further include an output parameter providing module 311 for adjusting a parameter related to an output option when the generated text sentence is output.
  • a parameter related to an output option that can be adjusted by the output parameter providing module 311 may be an intention type and an output text sentence quantity for each intention type.
  • the intention type is designated (eg, directions, radio search, address search, etc.) through the output parameter providing module 311 and the number of output text sentences for each designated intention type is adjusted to five, as shown in FIG. , 5 automatically generated text sentences for each type of intention are output.
  • the intention type eg, directions, radio search, address search, etc.
  • test sentence automatic generation unit 300 pure natural language for verification that is not used for deep learning model generation is uttered.
  • a pattern text sentence can be created, and the effect can be used as a test text sentence for a speech recognition device.
  • the text sentence generated and output by the test sentence automatic generation unit 300 may be used as an evaluation text for NLU (Natural Language Understanding) of the automatic speech recognition evaluation system.
  • NLU Natural Language Understanding
  • the following describes an example of testing the appropriateness of text sentences output for each intention, that is, reliability and redundancy of input sentences, by a deep learning-based automatic text sentence generation system for improving the infinity of speech patterns of the present inventors. Do it.
  • the redundancy of the input sentence is 40%, 40%, 80%, 40%, 0%, 60%, 20%, 40%, 20%, and 20%, respectively, and the redundancy of the average input sentence reaches 36%. It was confirmed that it has the advantage of securing new data for pure verification that was not used for generation.
  • a text sentence of a newly automatically generated natural language speech pattern can be generated and output for Navigate, which is an intention type 1, as follows, and this can be used as an NLU evaluation text.
  • Generated Pattern 1 navigate to a ⁇ POI> on ⁇ STREET> and ⁇ STREET>,
  • Generated Pattern 2 navigate me over to a ⁇ POI> on ⁇ STREET> and ⁇ STREET>,
  • Generated Pattern 4 navigate to the ⁇ STREET> and ⁇ STREET> intersection
  • Generated Pattern 5 navigate to ⁇ POI> on the corner of ⁇ STREET> and ⁇ STREET>.
  • the probability of generating non-grammatical sentences for each type of intention is different, but it shows a level of matching that is sufficiently utilized, and the new sentences outputted in this way are refined (remove original duplicate sentences, After the lack of domain information is removed), it can be used as NLU evaluation text.
  • the deep learning modeling unit 100 receives native-speaker texts having speech patterns and intention information, performs deep learning training, and automatically generates a test sentence for deep learning model information for each intention as a result of the execution. If provided as 300, the test sentence automatic generation unit 300 uses the deep learning model information for each intention and the intention type provided through the output intention input unit 200 to have different speech patterns corresponding to the intention type. By generating and outputting text sentences, it is possible to improve the infinity of the user's potential speech pattern.
  • the present inventor's deep learning-based text sentence automatic generation system for improving the infinity of speech patterns is based on previously collected data (texts having various speech patterns with specific intentions collected for use as deep learning learning data).
  • new text sentences text sentences that have the same intention as the utterance patterns of the collected text, but different utterance patterns
  • the user's potential utterance pattern that is, the infinity of potential utterance patterns
  • it provides an effect to improve (infinity)
  • it has high industrial applicability.

Abstract

La présente invention concerne un système pour générer automatiquement des phrases à base de texte sur la base de l'apprentissage profond afin d'obtenir une amélioration liée à l'infinité de modèles de prononciation et, plus particulièrement, un système pour générer automatiquement, sur la base de l'apprentissage profond, des phrases à base de texte ayant des modèles de prononciation de langage naturel. Le système est conçu pour : recevoir, par le biais d'une unité de modélisation d'apprentissage profond (100), une entrée de modèles de prononciation de textes et des informations à propos des objets communs des textes et réaliser un entraînement pour l'apprentissage profond à partir de ceux-ci ; fournir à une unité de génération automatique de phrases de test (300), par le biais de l'unité de modélisation d'apprentissage profond, des informations à propos des modèles d'apprentissage profond entraînés, classées en fonction des objets ; acquérir, par le biais de l'unité de génération automatique de phrases de test (300), des informations à propos des modèles d'apprentissage profond classées en fonction objets, fournies par l'unité de modélisation d'apprentissage profond ; entrer un type d'objet fourni par une unité d'entrée d'objet de sortie (200), aux informations correspondantes à propos des modèles d'apprentissage profond classées en fonction des objets et délivrer en sortie une phrase textuelle ayant un modèle de prononciation en langage naturel, la phrase étant générée automatiquement en fonction du type d'objet d'entrée ; et délivrer en sortie une valeur de crédibilité pour la phrase textuelle de sortie.
PCT/KR2019/006337 2019-04-15 2019-05-27 Système pour générer automatiquement des phrases à base de texte sur la base de l'apprentissage profond afin d'obtenir une amélioration liée à l'infinité de modèles de prononciation WO2020213785A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
KR1020190043515A KR102017229B1 (ko) 2019-04-15 2019-04-15 발화 패턴의 무한성 개선을 위한 딥러닝 기반의 텍스트 문장 자동 생성시스템
KR10-2019-0043515 2019-04-15

Publications (1)

Publication Number Publication Date
WO2020213785A1 true WO2020213785A1 (fr) 2020-10-22

Family

ID=67951389

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/KR2019/006337 WO2020213785A1 (fr) 2019-04-15 2019-05-27 Système pour générer automatiquement des phrases à base de texte sur la base de l'apprentissage profond afin d'obtenir une amélioration liée à l'infinité de modèles de prononciation

Country Status (2)

Country Link
KR (1) KR102017229B1 (fr)
WO (1) WO2020213785A1 (fr)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113179352A (zh) * 2021-05-25 2021-07-27 北京捷通华声科技股份有限公司 通话质量的确定方法、确定装置以及计算机可读存储介质

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102173382B1 (ko) 2020-02-25 2020-11-03 휴멜로 주식회사 텍스트 생성 장치 및 방법
CN113240088A (zh) * 2021-05-17 2021-08-10 上海中通吉网络技术有限公司 文本意图识别模型的训练方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2010224194A (ja) * 2009-03-23 2010-10-07 Sony Corp 音声認識装置及び音声認識方法、言語モデル生成装置及び言語モデル生成方法、並びにコンピューター・プログラム
KR20170078788A (ko) * 2015-07-02 2017-07-07 바이두 온라인 네트웍 테크놀러지 (베이징) 캄파니 리미티드 차량용 음성 명령어 인식 방법, 장치 및 저장 매체
KR20180071931A (ko) * 2016-12-20 2018-06-28 삼성전자주식회사 전자 장치, 그의 사용자 발화 의도 판단 방법 및 비일시적 컴퓨터 판독가능 기록매체
KR20180124716A (ko) * 2017-05-11 2018-11-21 경희대학교 산학협력단 효과적인 대화 관리를 위한 의료 시스템에서의 의도-컨텍스트 융합 방법
KR20190019294A (ko) * 2017-08-17 2019-02-27 주식회사 엘지유플러스 인공지능 장치의 자연어 처리를 검증하는 장치 및 방법

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20040073291A (ko) 2004-01-08 2004-08-19 정보통신연구진흥원 외국어 발음 평가 시스템 및 그 평가 방법

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2010224194A (ja) * 2009-03-23 2010-10-07 Sony Corp 音声認識装置及び音声認識方法、言語モデル生成装置及び言語モデル生成方法、並びにコンピューター・プログラム
KR20170078788A (ko) * 2015-07-02 2017-07-07 바이두 온라인 네트웍 테크놀러지 (베이징) 캄파니 리미티드 차량용 음성 명령어 인식 방법, 장치 및 저장 매체
KR20180071931A (ko) * 2016-12-20 2018-06-28 삼성전자주식회사 전자 장치, 그의 사용자 발화 의도 판단 방법 및 비일시적 컴퓨터 판독가능 기록매체
KR20180124716A (ko) * 2017-05-11 2018-11-21 경희대학교 산학협력단 효과적인 대화 관리를 위한 의료 시스템에서의 의도-컨텍스트 융합 방법
KR20190019294A (ko) * 2017-08-17 2019-02-27 주식회사 엘지유플러스 인공지능 장치의 자연어 처리를 검증하는 장치 및 방법

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113179352A (zh) * 2021-05-25 2021-07-27 北京捷通华声科技股份有限公司 通话质量的确定方法、确定装置以及计算机可读存储介质
CN113179352B (zh) * 2021-05-25 2023-04-07 北京捷通华声科技股份有限公司 通话质量的确定方法、确定装置以及计算机可读存储介质

Also Published As

Publication number Publication date
KR102017229B1 (ko) 2019-09-02

Similar Documents

Publication Publication Date Title
WO2020213785A1 (fr) Système pour générer automatiquement des phrases à base de texte sur la base de l'apprentissage profond afin d'obtenir une amélioration liée à l'infinité de modèles de prononciation
WO2019168253A1 (fr) Dispositif conversationnel de conseil interactif et procédé de compréhension hiérarchique de l'expression d'un utilisateur et de génération de réponse
WO2018151464A1 (fr) Système de codage et procédé de codage utilisant la reconnaissance vocale
WO2014106979A1 (fr) Procédé permettant de reconnaître un langage vocal statistique
WO2018128238A1 (fr) Système et procédé de consultation virtuelle utilisant un dispositif d'affichage
The SRI MAESTRO Team MAESTRO: Conductor of multimedia analysis technologies
WO2014025135A1 (fr) Procédé permettant de détecter des erreurs grammaticales, appareil de détection d'erreurs correspondant, et support d'enregistrement lisible par ordinateur sur lequel le procédé est enregistré
WO2020139058A1 (fr) Reconnaissance d'empreinte vocale parmi des dispositifs
WO2021162362A1 (fr) Procédé d'apprentissage de modèle de reconnaissance vocale et dispositif de reconnaissance vocale entraîné au moyen de ce procédé
WO2016021937A1 (fr) Procédé de sortie de contenu de données de texte sous forme de voix d'un émetteur de données de texte
WO2010150986A2 (fr) Appareil et procédé d’étude tout au long de la vie de mots en langue étrangère
WO2015023035A1 (fr) Procédé de correction d'erreurs de préposition et dispositif le réalisant
WO2019209040A1 (fr) Multimodèles qui comprennent des phrases en langage naturel
WO2017142109A1 (fr) Système et procédé d'amélioration des performances d'extraction d'un texte non structuré
WO2021071137A1 (fr) Procédé et système de génération automatique de questions d'inférence d'espace vide pour une phrase en langue étrangère
WO2018088664A1 (fr) Dispositif de détection automatique d'erreur de corpus d'étiquetage morphosyntaxique au moyen d'ensembles approximatifs, et procédé associé
WO2018131955A1 (fr) Procédé d'analyse de contenus numériques
WO2021251539A1 (fr) Procédé permettant de mettre en œuvre un message interactif en utilisant un réseau neuronal artificiel et dispositif associé
WO2022075714A1 (fr) Procédé et système d'extraction d'intégration de locuteur utilisant une technique de regroupement basée sur un système de reconnaissance de la parole pour l'identification de locuteur, et leur support d'enregistrement
WO2020204256A1 (fr) Système multimédia automatique d'évaluation de reconnaissance de parole utilisant un moteur de synthèse de parole
WO2014200187A1 (fr) Appareil pour apprendre l'apophonie et procédé associé
WO2021137534A1 (fr) Procédé et système d'apprentissage de la prononciation coréenne par analyse vocale
WO2018169276A1 (fr) Procédé pour le traitement d'informations de langue et dispositif électronique associé
WO2016072569A1 (fr) Système de service d'apprentissage du vocabulaire anglais permettant à un utilisateur de se familiariser avec le vocabulaire anglais via les syllabes et les accentuations
WO2020149541A1 (fr) Procédé et dispositif pour générer automatiquement un ensemble de données de questions-réponses pour un thème spécifique

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19925290

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19925290

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 29.04.2022)

122 Ep: pct application non-entry in european phase

Ref document number: 19925290

Country of ref document: EP

Kind code of ref document: A1