KR950034054A - Candidate Word Extractor and Extraction Method for Large Vocabulary Speech Recognition - Google Patents

Candidate Word Extractor and Extraction Method for Large Vocabulary Speech Recognition Download PDF

Info

Publication number
KR950034054A
KR950034054A KR1019940010682A KR19940010682A KR950034054A KR 950034054 A KR950034054 A KR 950034054A KR 1019940010682 A KR1019940010682 A KR 1019940010682A KR 19940010682 A KR19940010682 A KR 19940010682A KR 950034054 A KR950034054 A KR 950034054A
Authority
KR
South Korea
Prior art keywords
candidate word
codeword
word
candidate
words
Prior art date
Application number
KR1019940010682A
Other languages
Korean (ko)
Inventor
안영목
김회린
황규웅
Original Assignee
양승택
재단법인 한국전자통신연구소
조백제
한국전기통신공사
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 양승택, 재단법인 한국전자통신연구소, 조백제, 한국전기통신공사 filed Critical 양승택
Priority to KR1019940010682A priority Critical patent/KR950034054A/en
Publication of KR950034054A publication Critical patent/KR950034054A/en

Links

Landscapes

  • Document Processing Apparatus (AREA)

Abstract

본 발명은 대어휘 음성인식을 위한 후보단어 추출장치 및 추출 방법에 관한 것으로, 후보단어 추출기의 효과는 대어휘 인식시에 인식단어일 확률이 희박한 단어를 미리 제외 시킴으로써 소요되는 계산 시간을 대폭 줄일수 있다. 따라서 음성 인식 시스템의 속도를 증진시킬 것이며 최종 인식단어 추출부의 담색 영역을 대폭 줄일수 있다.The present invention relates to a candidate word extraction apparatus and extraction method for large vocabulary speech recognition, the effect of the candidate word extractor can significantly reduce the computation time required by excluding words that are less likely to be recognized words in large vocabulary recognition in advance. have. Therefore, the speed of the speech recognition system will be improved, and the pale color area of the final recognition word extract can be greatly reduced.

Description

대어휘 음성인식을 위한 후보단어 추출장치 및 추출 방법.Candidate Word Extraction Apparatus and Extraction Method for Large Vocabulary Speech Recognition.

본 내용은 요부공개 건이므로 전문내용을 수록하지 않았음Since this is an open matter, no full text was included.

제1도는 음성인식 시스템의 전체 구성도, 제2도는 후보단어 추출부가 첨가된 음성인식 시스템의 전체 구성도,제3도는 후보단어 추출부의 전체 흐름도.1 is an overall configuration diagram of a speech recognition system, FIG. 2 is an overall configuration diagram of a speech recognition system to which a candidate word extracting unit is added, and FIG.

Claims (3)

마이크로폰에서 음성을 전기적 신호로 변환하여 공급하면 인간의 가청 주파수 대역의 신호만을 분리 통과시키는 대역 필터링수단(1); 상기 대역 필터링수단(1)을 통해 출력된 신호를 샘플율에 따라 샘플링하는 A/D변환수단(2); 상기 A/D변환수단(2)에 접속되어 샘플링된 신호를 분석하여 음성 구간에 해당되는 신호를 특징(Cepstrum)벡터를 추출하는 특징추출부와, 특징벡터(Cepstrum)가 코드북 중에서 가장 근접된 기준 벡터(코드워드)를 결정하는 코드워드 추출부와, 훈련 단계에서 만들어진 전체 후보단어들의 코드워드 분포도를 사용하여 발생할 확률이 높은 적은 수의 단어만을 후보단어 제한폭에 의해서 추출하는 후보단어 추출보와, 기준패턴과의 후보단어 추출부에서 선택된 적은 수의 단어를 통해 최종 인식단어를 골라내는 최종 인식단어 추출부(인식부)를 구바하는 제어수단(3); 상기 제어수단(3)의 최종 인식단어를 출력시키는 I/O디코딩수단(7); 상기 제어수단(3)의 제어 신호를 전달하는 어드레스 디코딩 수단(4); 상기 어드레스 디코딩 수단(4)을 통해 상기 제어 수단(3)의 제어신호를 전달받아코드북, 각 단어들의 코드워드 분포도, 후보 단어 제한률에 대한 정보를 저장하는 데이타 룸(6); 및 상기 어드레스 디코딩 수단(4)을 통해 상기 제어수단(3)의 제어신호를 전달받아 특징주출을 위한 프로그램, 후보단어 추출 프로그램, 최종 인식 프로그램을 구비하는 프로그램 롬(5)을 구비하는 것을 특징으로 하는 대어휘 음성인식을 위한 후보단어 추출 장치.Band filtering means (1) for converting and supplying a voice to an electric signal from a microphone and separating and passing only a signal of an audible frequency band of a human; A / D conversion means (2) for sampling the signal output through the band filtering means (1) according to the sample rate; A feature extractor which is connected to the A / D conversion means 2 and extracts a feature vector of a signal corresponding to a voice section by analyzing a sampled signal; and a reference of which a feature vector is the closest among codebooks. A codeword extraction unit for determining a vector (codeword), a candidate word extraction report for extracting only a small number of words having a high probability of occurrence by using a candidate word limit using a codeword distribution of all candidate words generated in a training step, Control means (3) for compensating for a final recognition word extraction unit (recognition unit) for selecting the final recognition word through a small number of words selected by the candidate word extraction unit with the reference pattern; I / O decoding means (7) for outputting a final recognized word of said control means (3); Address decoding means (4) for transmitting a control signal of said control means (3); A data room (6) which receives the control signal of the control means (3) via the address decoding means (4) and stores information on a codebook, codeword distribution of each word, and candidate word limit; And a program ROM 5 having a program for feature extraction, a candidate word extraction program, and a final recognition program by receiving the control signal of the control means 3 through the address decoding means 4. Candidate word extraction device for large vocabulary speech recognition. 대어휘 음성인식을 위한 후보단어 추출장치에 적용되는 후보단어 추출 방법에 있어서, 특징 추출부에서 특징 벡터가 양자화기에 입력되어 가장 근접되는 코드북상의 기준 벡터를 알아낸 뒤 이를 시간 영역 분할하도록 하는 제1단계; 상기 제1단계 수행 후, 시간 영역 분할 후 일정한 영역별로 코드워드 발생 정도를 기록하여 코드워드 비교를 하도록 하는 제2단계; 상기 제2단계 수행 후, 모든 인식 대상 후보 단어들의 코드워드 분포도를 이용하여 해당 단어가 발생될 확률값을 계산하고 이를 후보단어 추출부로 보내는 제3단계; 및 상기 제3단계 수행 후, 후보단어 제한률에 의해 해당 후보 단어를 찾아낸 목록을 최종 인식 추출부를 통해 내보내는 제4단계를 포함하는 것을 특징으로 하는 대어휘 음성인식을 위한 후보단어 추출 방법.A candidate word extraction method applied to a candidate word extraction apparatus for large vocabulary speech recognition, comprising: a feature extractor inputting a feature vector into a quantizer to find a reference vector on a codebook closest to the nearest one and then time-domain partitioning the feature vector; Stage 1; A second step of performing a codeword comparison by recording a codeword generation degree for each predetermined region after performing the first step; A third step of calculating a probability value of generating a corresponding word using a codeword distribution map of all candidate words to be recognized after performing the second step and sending it to the candidate word extracting unit; And a fourth step of, after performing the third step, exporting a list in which the candidate word is found by the candidate word limiting rate through a final recognition extracting unit. 제2항에 있어서, 상기 제3단계는, 실험 특정 벡터를 코드북의 기준 특징 벡터들중에서 가장 가까운 것을 하나만 골라내어 출력시키는 제1과정; 시간 영역 분할에서 심볼열을 일정한 시간에 의해 모든 단어가 동일한 영역수를 갖도록 분리하는 제2과정;시간 영역별 발생 코드워드 기록에서 각 단어이 해당 구간에 발생된 코드워드를 기록하고 전체 후보단어의 학습데이타가 모두 입력될때까지 반복 수행하는 제3과정을 포함하고 있는것을 특징으로 하는 음성인식을 후보단어 추출 방법.3. The method of claim 2, wherein the third step comprises: a first step of selecting and outputting one experimental nearest vector from the reference feature vectors of the codebook; A second step of separating the symbol strings in a time domain division so that all words have the same number of domains by a predetermined time; in the occurrence codeword recording for each time domain, each word is recorded in the corresponding codeword and the entire candidate words are learned. And a third process of repeatedly performing data until all data is input. 참고사항 : 최초출원 내용에 의하여 공개하는 것임.Note: The disclosure is based on the original application.
KR1019940010682A 1994-05-16 1994-05-16 Candidate Word Extractor and Extraction Method for Large Vocabulary Speech Recognition KR950034054A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
KR1019940010682A KR950034054A (en) 1994-05-16 1994-05-16 Candidate Word Extractor and Extraction Method for Large Vocabulary Speech Recognition

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
KR1019940010682A KR950034054A (en) 1994-05-16 1994-05-16 Candidate Word Extractor and Extraction Method for Large Vocabulary Speech Recognition

Publications (1)

Publication Number Publication Date
KR950034054A true KR950034054A (en) 1995-12-26

Family

ID=66682444

Family Applications (1)

Application Number Title Priority Date Filing Date
KR1019940010682A KR950034054A (en) 1994-05-16 1994-05-16 Candidate Word Extractor and Extraction Method for Large Vocabulary Speech Recognition

Country Status (1)

Country Link
KR (1) KR950034054A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR100322202B1 (en) * 1999-09-06 2002-02-06 윤장진 Device and method for recognizing voice sound using nervous network
KR100450396B1 (en) * 2001-10-22 2004-09-30 한국전자통신연구원 Tree search based speech recognition method and speech recognition system thereby

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR100322202B1 (en) * 1999-09-06 2002-02-06 윤장진 Device and method for recognizing voice sound using nervous network
KR100450396B1 (en) * 2001-10-22 2004-09-30 한국전자통신연구원 Tree search based speech recognition method and speech recognition system thereby

Similar Documents

Publication Publication Date Title
CN108447486B (en) Voice translation method and device
EP0140777B1 (en) Process for encoding speech and an apparatus for carrying out the process
CN112151005B (en) Chinese and English mixed speech synthesis method and device
CN110767210A (en) Method and device for generating personalized voice
JPS6147440B2 (en)
CN112581963B (en) Voice intention recognition method and system
Shaikh Naziya et al. Speech recognition system—a review
Rudresh et al. Performance analysis of speech digit recognition using cepstrum and vector quantization
Mishra et al. An Overview of Hindi Speech Recognition
Manjutha et al. Automated speech recognition system—A literature review
KR950034054A (en) Candidate Word Extractor and Extraction Method for Large Vocabulary Speech Recognition
CN113990325A (en) Streaming voice recognition method and device, electronic equipment and storage medium
CN114446278A (en) Speech synthesis method and apparatus, device and storage medium
CN112927677A (en) Speech synthesis method and device
Reddy et al. Transcription of Telugu TV news using ASR
Bisikalo et al. Method of automated transcribing of speech signals for information technology of text-dependent authentication of a person by voice
CN113763924B (en) Acoustic deep learning model training method, and voice generation method and device
JPH10133678A (en) Voice reproducing device
JP2806048B2 (en) Automatic transcription device
CN113362805B (en) Chinese and English speech synthesis method and device with controllable tone and accent
KR102480360B1 (en) Apparatus, method and computer program for generating synthesized sound source using learning through image
KR20220112560A (en) Speech recognition and speech dna generation system in real time end-to-end
KR100608643B1 (en) Pitch modelling apparatus and method for voice synthesizing system
Mukherjee et al. Leveraging Cross Lingual Speech Representations To Build ASR For Under-resourced Languages
CN117746831A (en) Emotion controllable speech synthesis method and system based on few samples of specific characters

Legal Events

Date Code Title Description
A201 Request for examination
N231 Notification of change of applicant
E902 Notification of reason for refusal
E601 Decision to refuse application