Voice recognition method and apparatus by updating reference patterns

A voice recognition method comprising the steps of

• extracting patterns from an input voice signal,
• tentatively identifying each of said extracted patterns as representing a syllable by calculating similarity levels of said each extracted pattern with a plurality of standard patterns each of which is currently registered in a memory means, said memory means also storing phonological information on an environment associated with each of said currently registered standard patterns,
• determining which of said extracted patterns were correctly and incorrectly identified in said tentatively identifying step,
• assigning a new syllable to each of said patterns which were determined as having been incorrectly identified in said determining step,
• storing in a first buffer temporal sequences each associated with a syllable and showing the order in which patterns are determined in said determining step to have been correctly and incorrectly identified in said tentatively identifying step as representing said syllable,
• storing in a second buffer the temporal sequence in which all of said extracted patterns have been determined in said determining step as having been correctly and incorrectly identified in said tentatively identifying step,
• referencing the contents of said first and second buffers, when a new voice signal with new patterns is received, to obtain therefrom a recent recognition record of each syllable backward in time from when said new voice signal is received, and
• determining according to said recognition record whether or not one or more of said standard patterns is to be updated by one or more of said new patterns.

A voice recognition method comprising the steps of

• (a) setting a maximum number allowable for standard patterns associated with each of specified syllables,
• (b) initially registering in a memory means a minimum number of standard patterns required for pattern recognition,
• (c) extracting patterns from an input voice signal,
• (d) tentatively identifying each of said extracted patterns as representing a syllable by calculating similarity levels of said each extracted pattern with a plurality of standard patterns each of which is currently registered in said memory means,
• (e) determining which of said extracted patterns were incorrectly identified in said tentatively identifying step,
• (f) assigning a new syllable to each of said patterns which were determined as having been incorrectly identified in said determining step, and
• (g) thereafter automatically updating said standard patterns in said memory means, whenever a new voice signal is inputted, by repeating steps (c)-(f) to obtain new patterns and selectively registering said new patterns in said memory means within the limit of not exceeding said maximum number.

A voice recognition apparatus for receiving an input voice signal and outputting a syllable array corresponding thereto, said apparatus comprising

• memory means for storing standard patterns associated with syllables together with phonological environmental information,
• analyzing means for extracting patterns from an input voice signal and tentatively identifying each of said extracted patterns as representing a syllable by calculating similarity levels of said each extracted pattern with a plurality of standard patterns currently stored in said memory means,
• a control unit,
• input means for causing said control unit to assign a selected new syllable to each pattern determined to have been incorrectly identified by said analyzing means, and
• a voice recognition means which serves under the control of said control unit to associate each of said standard patterns with phonological information on at least one neighborhood immediately before or after in a voice signal from which said each standard pattern was derived, and to update said memory means by registering therein said extracted patterns for which a new syllable has been assigned by said input means, each of said extracted patterns replacing one of said standard patterns having the same phonological information on a neighborhood thereof as do said extracted patterns.

A voice recognition apparatus for receiving an input voice signal and outputting a syllable array corresponding thereto, said apparatus comprising

• memory means for storing standard patterns associated with syllables together with phonological environmental information,
• analyzing means for extracting patterns from an input voice signal and tentatively identifying each of said extracted patterns as representing a syllable by calculating similarity levels of said each extracted pattern with a plurality of standard patterns currently stored in said memory means,
• input means for assigning a selected new syllable to each pattern determined to have been incorrectly identified by said analyzing means, the remaining ones of said extracted patterns being considered as correctly identified patterns,
• a first buffer storing temporal sequences, each corresponding to a syllable and showing the order in which said correctly and incorrectly identified patterns associated with said corresponding syllable have been identified,
• a second buffer storing the temporal sequence in which all of said correctly and incorrectly identified patterns have been identified, and
• a control unit which serves to reference the contents of said first and second buffers, when a new voice signal with new patterns is received, to obtain therefrom a recent recognition record of each syllable backward in time from when said new voice signal is received and to determine according to said recognition record whether or not one or more of said standard patterns is to be updated by one or more of said new patterns.

A voice recognition apparatus for receiving an input voice signal and outputting a syllable array corresponding thereto, said apparatus comprising

• memory means for storing standard patterns associated with syllables,
• analyzing means for extracting patterns from an input voice signal and tentatively identifying each of said extracted patterns as representing a syllable by calculating similarity levels of said each extracted pattern with a plurality of standard patterns currently stored in said memory means,
• input means for assigning a selected new syllable to each pattern determined to have been incorrectly identified by said analyzing means, the remaining ones of said extracted patterns being considered as correctly identified patterns, and
• control means programmed to set a maximum number allowable for standard patterns associated with each of specified syllables, to initially register in said memory means a minimum number of standard patterns required for pattern recognition, and to thereafter automatically update the contents of said memory means, whenever a new voice signal is inputted, by registering in said memory means new patterns extracted from said inputted voice signal within the limit of not exceeding said maximum number.