[미국특허]
Compressed speech lexicon and method and apparatus for creating and accessing the speech lexicon
원문보기
IPC분류정보
국가/구분
United States(US) Patent
등록
국제특허분류(IPC7판)
G06F-017/28
G06F-017/21
출원번호
US-0751871
(2000-12-29)
등록번호
US-7451075
(2008-11-11)
발명자
/ 주소
Mohammed,Yunus
출원인 / 주소
Microsoft Corporation
대리인 / 주소
Kelly,Joseph R.
인용정보
피인용 횟수 :
2인용 특허 :
38
초록▼
A compressed lexicon is built by receiving a word list, which includes word-dependent data associated with each word in the word list. A word is selected from the word list. A hash value is generated based on the selected word, and the hash value identifies an address in a hash table which, in turn,
A compressed lexicon is built by receiving a word list, which includes word-dependent data associated with each word in the word list. A word is selected from the word list. A hash value is generated based on the selected word, and the hash value identifies an address in a hash table which, in turn, is written with a location in lexicon memory that is to hold the compressed form of the selected word, and the compressed word-dependent data associated with the selected word. The word is then encoded, or compressed, as is its associated word-dependent data. This information is written at the identified location in the lexicon memory.
대표청구항▼
What is claimed is: 1. A method of building a compressed speech lexicon for use in a speech application, comprising: receiving a word list configured for use in the speech application, the word list including a plurality of words, with each word in the word list having associated word-dependent dat
What is claimed is: 1. A method of building a compressed speech lexicon for use in a speech application, comprising: receiving a word list configured for use in the speech application, the word list including a plurality of words, with each word in the word list having associated word-dependent data selected from the group consisting of a pronunciation and part-of-speech; selecting one of the words from the word list; generating an index entry identifying a location in a compressed speech lexicon memory for holding the selected word; encoding the selected word and its associated word-dependent data to obtain encoded words and associated encoded word-dependent data; and writing the encoded word and its associated word-dependent data at the identified location in the speech lexicon memory. 2. The method of claim 1 and further comprising: repeating the steps of selecting, generating, encoding and writing for each word in the word list and the associated word-dependent data. 3. The method of claim 2 and further comprising: writing codebooks corresponding to the encoded words and the encoded word-dependent data in the speech lexicon memory. 4. The method of claim 1 wherein receiving the word list comprises: counting the words in the word list; allocating a hash table memory based on a number of words in the word list; and allocating a speech lexicon memory based on the number of words in the word list. 5. The method of claim 1 wherein encoding comprises: providing a word encoder to encode the words in the word list and encoding the words with the word encoder; and providing word-dependent data encoders for each type of word-dependent data in the word list and encoding the word-dependent data with the word-dependent data encoders. 6. The method of claim 5 wherein encoding further comprises: Hufmann encoding the selected word and its associated word-dependent data. 7. The method of claim 1 wherein writing the encoded word and word-dependent data comprises: writing a data structure comprising: a word portion containing the encoded word; a word-dependent data portion containing the encoded word-dependent data; and wherein each word-dependent data portion has an associated last indicator portion and word-dependent data indicator portion, the last indicator portion containing an indication of a last portion of word-dependent data associated with the selected word, and the word-dependent data indicator portion containing an indication of the type of word-dependent data stored in the associated word dependent data portion. 8. The method of claim 7 wherein writing a data structure comprises writing the word portion and the word-dependent data portions as variable length portions followed by a separator. 9. The method of claim 1 wherein generating an index entry comprises: determining a next available location in the speech lexicon memory. 10. The method of claim 9 wherein generating an index entry comprises: calculating a hash value for the selected word; indexing into the hash table to an index location based on the hash value; and writing location data identifying the next available location in the speech lexicon memory into the index location in the hash table. 11. The method of claim 10 wherein writing location data comprises: writing an offset into the speech lexicon memory that corresponds to the next available location in the speech lexicon memory. 12. A method of accessing word information related to a word stored in a compressed speech lexicon, comprising: receiving the word; accessing an index to obtain a word location in the compressed speech lexicon that contains information associated with the received word including word-dependent data selected from the group consisting of a pronunciation and a part-of-speech; reading encoded word information from the word location; and decoding the word information for use in a speech application. 13. The method of claim 12 and further comprising: prior to reading the encoded word information, reading an encoded word from the word location; decoding the encoded word; and verifying that the decoded word is the same as the received word. 14. The method of claim 12 wherein decoding the word information comprises: initializing decoders associated with the word and its associated information. 15. The method of claim 12 wherein accessing an index comprises: calculating a hash value based on the received word; finding an index location in the index based on the hash value; and reading from the index location a pointer value pointing to the word location in the compressed lexicon. 16. The method of claim 12 wherein reading the encoded word information comprises: reading a plurality of fields from the word location containing variable length word information. 17. The method of claim 16 wherein reading a plurality of fields comprises: prior to reading each field, reading data type header information indicating a type of word information in an associated field. 18. The method of claim 17 wherein reading a plurality of fields comprises: reading a last field indicator indicating whether an associated one of the plurality of fields is a last field associated with the received word. 19. A compressed speech lexicon builder for building a compressed speech lexicon for use in a speech application based on a word list containing a plurality of domains, the domains including words and word-dependent data associated with each of the words, the compressed speech lexicon builder comprising: a plurality of domain encoders, one domain encoder being associated with each domain in the word list, the domain encoders being configured to compress the words and the associated word-dependent data selected from the group consisting of a pronunciation and a part-of-speech, to obtain compressed words and compressed word-dependent data; a hashing component configured to generate a hash value for each word in the word list; a hash table generator, coupled to the hashing component, configured to determine a next available location in a speech lexicon memory and write, at an address in a hash table identified by the hash value, the next available location in the speech lexicon memory; and a speech lexicon memory generator, coupled to the domain encoders and the hash table generator, configured to store in the speech lexicon memory, for use by the speech application, the compressed words and compressed word-dependent data, each compressed word and its associated compressed word-dependent data being stored at the next available location in the speech lexicon memory written in the hash table at the hash table address associated with the compressed word. 20. The compressed speech lexicon builder of claim 19 and further comprising: a codebook generator generating a codebook associated with each domain encoder.
Deligne Sabine ; Sagisaka Yoshinori,JPX ; Nakajima Hideharu,JPX, Apparatus for generating a statistical sequence model called class bi-multigram model with bigram dependencies assumed between adjacent sequences.
Braden-Harder Lisa C. (Somers NY) Kim Michelle Y. L. (Scarsdale NY) Klavans Judith L. (Hastings-on-Hudson NY) Zadrozny Wlodek W. (Mohegan Lake NY), Archiving and retrieving multimedia objects using structured indexes.
Mitchell John C.,GB2 ; Heard Alan James,GB2 ; Corbett Steven Norman,GB2 ; Daniel Nicholas John,GB2, Automated proofreading using interface linking recognized words to their audio data while text is being changed.
Lewis G. Pringle ; Robert W. Swerdlow ; Alec Wysoker, Automated translation of annotated text based on the determination of locations for inserting annotation tokens and linked ending, end-of-sentence or language tokens.
Shieber Stuart M. ; Armstrong John ; Baptista Rafael Jose ; Bentz Bryan A. ; Ganong ; III William F. ; Selesky Donald Bryant, Command parsing and rewrite system.
Young Jonathan Hood ; Parmenter David Wilsberg ; Roth Robert ; Dubach Joev ; Gadbois Gregory J. ; Van Even Stijn, Error correction in speech recognition.
Dahan Jean-Guy ; Gupta Vishwa,CAX, Method and apparatus for performing speech recognition utilizing a supplementary lexicon of frequently used orthographies.
Daniel M. Coffman ; Popani Gopalakrishnan ; Ganesh N. Ramaswamy ; Jan Kleindienst CZ; Chalapathy V. Neti, Method and system for multi-client access to a dialog system.
Beattie Valerie L. ; Miller David R. H. ; Edmondson Shawn Eric ; Patel Yogen N. ; Talvola Geoffrey A., Multi-dialect speech recognition method and apparatus.
Loatman Robert B. (Vienna VA) Post Stephen D. (McLean VA) Yang Chih-King (Rockville MD) Hermansen John C. (Catharpin VA), Natural language understanding system.
Andrew J. Hunt ; William D. Walker ; Johan Wouters, System and method for interfacing speech recognition grammars to individual components of a computer program.
Hantler, Sidney L.; Laker, Meir M.; Lenchner, Jonathan; Milch, Daniel, Methods and apparatus for performing spelling corrections using one or more variant hash tables.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.