Speech recognition method and system for recognizing single or un-correlated Chinese characters

• US 6,163,767 A
• Filed: 08/28/1998
• Issued: 12/19/2000
• Est. Priority Date: 09/19/1997
• Status: Expired due to Term

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: International Business Machines Corporation
• Inventors: Tang, Donald T., Shen, Li Qin, Zhu, Xiao Jin
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): ABEBE, DANIEL DEMELASH
• Application Number:
  US09/141,901
• Time in Patent Office: 844 Days
• Field of Search: 704/235, 704/260, 704/256, 704/257, 704/251, 704/231, 707/535, 707/536
• US Class Current: 704/231
• CPC Class Codes: G10L 15/193 Formal grammars, e.g. finite state automata, context free grammars or word networks

Abstract

A Chinese speech recognition (SR) method and system for single or un-correlated Chinese character(s). The method uses various types of Character Description Language (CDL) to describe the single or un-correlated Chinese character(s) to be inputted. The SR system uses CDL grammar directed speech recognizer to accept CDLs, which are inputted by voice. On the basis of analysis of CDL parser, the character generator gives a corresponding character. Therefore, recognition of single or un-correlated Chinese character(s) out of context can be made reliably.

Claims

1. A recognition method for single or un-correlated Chinese character(s) in a Chinese speech recognition system, characterized by the steps of:

describing the single or un-correlated Chinese character(s) to be inputted by means of various types of character Description Language wherein the various types of Character Description Language comprise;

A) describing a character by an ad hoc word which people usually use;

B) describing a character with the help of a related word;

C) describing a character taking advantage of Chinese character'"'"'s structure information;

D) describing a character by means of semantics;

E) describing a character through stroker; and

F) describing single English letter switch Chinese Pinyin, so that a single or un-correlated Chinese character(s) can be described by one of said various types of character Description Language or by combination of them;

parsing the inputting various types of Character Description Language; and

generating a corresponding character on the basis of parsing analysis.

2. The method of claim 1 including the steps of analyzing the grammar of the Chinese Character Description Language and generating the single or un-correlated Chinese characters as follows;

looking up in a table the specific word and its corresponding character to get the single character if the current reduction uses type A;

getting the single character which is acoustically close to the character in the word if the current reduction uses type B;

looking up in a dictionary containing the characters and their components and radicals to get the character if the current reduction uses type C;

looking up in a dictionary containing the characters and their meanings and synonyms and identify the single character by its pronunciation if the current reduction uses type D;

looking up in a dictionary containing the characters and their stoke information and identify the single character by its pronunciation if the current reduction uses type E; and

getting the first or last alphabet of the Pinyin of this word if the current reduction uses type F.

3. The method of claim 1 including the steps of:

inputting a word string consisting of previously recognized words h1, h2, . . . , hi;

finding the grammar rules R1, R2, . . . , Rj that h1, h2, . . . , hi in CDL grammar;

finding for each one of the grammar rules R1, R2, . . . , Rj, the next possible word hi+1 of string h1, h2, . . . , hi;

forming a set of all the possible hi+1'"'"'s;

modifying the probability estimation of each word w in the vocabulary by the language model;

setting P(w|h1, h2, . . . , h1) to zero if w is not in S; and

normalizing the probability estimations to give a final output P (W|h1, h2, . . . , hi).

4. A Chinese speech recognition system for single or un-correlated Chinese character(s), characterized by comprising:

CDL grammar directed speech recognizer for accepting various types of Character Description Language to be inputted by voice wherein the various types of Character Description Language comprise at least two of the following types;

A) describing a character by an ad hoc word which people usually use;

B) describing a character with the help of a related word;

C) describing a character taking advantage of Chinese character'"'"'s structure information;

D) describing a character by means of semantics;

E) describing a character through strokes; and

F) describing a single English letter with Chinese Pinyin;

CDL parser for parsing the Character Description Language outputted by said CDL grammar directed speech recognizer;

character generator for generating a corresponding character on the basis of analysis of parser.

5. The speech recognition system of claim 4 wherein said various types of character descriptions comprise all of the types A to F.