-
公开(公告)号：JPS62231993A
公开(公告)日：1987-10-12
申请号：JP6503086
申请日：1986-03-25
Applicant: IBM
Inventor： KURODA AKIHIRO , NISHIMURA MASAFUMI , SUGAWARA KAZUHIDE
-
Patent Agency Ranking
公开(公告)号：JPH01167898A
公开(公告)日：1989-07-03
申请号：JP30609187
申请日：1987-12-04
Applicant: IBM
Inventor： SUGAWARA KAZUHIDE
Abstract: PURPOSE: To select a candidate word stably and preliminarily by evaluating the former half part and latter half part of fixed length of vocalization with individual score tables, making good use of time information, and performing a smoothing process along its time base. CONSTITUTION: Spoken words are divided into the former half part and latter half part of fixed length and the individual score tables are used to perform the likelihood calculation 15 of the preliminary choice 10, so processes can be performed in frame synchronism without waiting for all the words to be spoken. Further, smoothing is done when the score tables for the former half part and latter half part are generated, so the preliminary choice 10 can be made stably against fluctuations of vocalization along the time base. Further, smoothing between labels is performed, so excellent score tables can be generated even by small learning. Here, a preliminary selection part 10, a detailed recognition part 11, a word segmenting circuit 8, and a gate circuit 9 are actualized by software on a personal computer and a feature extracting circuit 4, a labeling circuit 6, and a power extracting circuit 5 are actualized by signal processing boards mounted on the personal computer.
-
公开(公告)号：CA1256562A
公开(公告)日：1989-06-27
申请号：CA528993
申请日：1987-02-04
Applicant: IBM
Inventor： KURODA AKIHIRO , NISHIMURA MASAFUMI , SUGAWARA KAZUHIDE
Abstract: Speaker adaptation is provided which easily enables a person to use a Hidden Markov model type recognizer previously trained by other particular person or persons. During training process, parameters of Markov models are calculated iteratively for example using Forward-Backward algorithm. The adaptation comprises storing and utilizing the intermediate results or probabilistic frequences of the last iteration. During the adaptation process, parameters are calculated by interpolation of the weighted sum of the stored probabilistic frequences and the ones obtained using new training data.
-
公开(公告)号：DE3878852T2
公开(公告)日：1993-09-23
申请号：DE3878852
申请日：1988-10-28
Applicant: IBM
Inventor： SUGAWARA KAZUHIDE
Abstract: A method and apparatus for automated speech recognition wherein an unknown speech input is divided into words on merit, the words are divided into fixed length leading portions and consequentially variable length remaining portions, both portions being further subdivided into fixed length plural frames, each frame being matched against a fixed set of features, determined to correspond to one of the same and tagged with a label representing that feature, the fixed length label string generated from the leading portion of a word being used to aggregate a probability value for the string by accessing individual probability values of occurrence from a leading portion table of labels against first portions of known words and the remaining portion label string being correspondingly processed against a separate remaining portion table, the two sets of aggregate values being compounded by word portion pairs and the words corresponding to the highest compounded values being selected as candidate words for further processing to produce a result, whereby execution of the first portion processing of a word can be undertaken while the remaining portion of the word is being received and the selection of candidates includes a measure of the time of occurrence of a determined feature in a word. The tables are established in a learning phase in which known words are input, divided, framed, featurised and labelled, the probability values for the labelled features being derived by accumulating time dependent weighting factors for given labels in given word, normalising the totals and smoothing the two sets of results.
-
公开(公告)号：DE3773039D1
公开(公告)日：1991-10-24
申请号：DE3773039
申请日：1987-03-25
Applicant: IBM
Inventor： KURODA AKIHIRO , NISHIMURA MASAFUMI , SUGAWARA KAZUHIDE
Abstract: The present invention relates to a speech recognition system of the type comprising a plurality of probabilistic finite state models each having a number of states and the ability of undergoing transitions from one state to another and producing a corresponding output representing a speech element, and means for defining for each model probability parameters each representing the probability that the model will undergo a transition from one predetermined state to another predetermined state and produce a corresponding output. Such a system can be used to recognise input speech data by initially dividing the input speech data into individual speech elements (4, 5, 6) and then applying the input speech elements to the models, and utilising the probability parameters of the models to recognise the input speech elements. According to the invention the speech recognition system is characterised in that it comprises training means (8) for supplying training speech data to the models in order to train the models and to define initial values for the probability parameters for each of the models, and adaptation means (9) for supplying adaptation speech data to the models in order to adapt the models and to define adapted values of the probability parameters for each of the models. The adapted values of the probability parameters are used to recognise the input speech elements (10).
-
公开(公告)号：DE3878852D1
公开(公告)日：1993-04-08
申请号：DE3878852
申请日：1988-10-28
Applicant: IBM
Inventor： SUGAWARA KAZUHIDE
Abstract: A method and apparatus for automated speech recognition wherein an unknown speech input is divided into words on merit, the words are divided into fixed length leading portions and consequentially variable length remaining portions, both portions being further subdivided into fixed length plural frames, each frame being matched against a fixed set of features, determined to correspond to one of the same and tagged with a label representing that feature, the fixed length label string generated from the leading portion of a word being used to aggregate a probability value for the string by accessing individual probability values of occurrence from a leading portion table of labels against first portions of known words and the remaining portion label string being correspondingly processed against a separate remaining portion table, the two sets of aggregate values being compounded by word portion pairs and the words corresponding to the highest compounded values being selected as candidate words for further processing to produce a result, whereby execution of the first portion processing of a word can be undertaken while the remaining portion of the word is being received and the selection of candidates includes a measure of the time of occurrence of a determined feature in a word. The tables are established in a learning phase in which known words are input, divided, framed, featurised and labelled, the probability values for the labelled features being derived by accumulating time dependent weighting factors for given labels in given word, normalising the totals and smoothing the two sets of results.
-
-
-
-
-
Search Results
6
records
(took 0.015 seconds)
