Word and pattern recognition through overlapping hierarchical tree defined by relational features
First Claim
1. A voice recognizing method, comprising the steps of:
- providing a plurality of voice recognition objective words;
structuring said voice recognition objective words into a hierarchical structure having a plurality of levels based upon correlation values indicative of similarities between each pair of voice recognition objective words;
inputting an unknown voice signal;
accumulating scores for the voice recognition objective words which are related in said hierarchical structure to a voice recognition word in a higher level, which word in said higher level receives a high score for said unknown input voice signal by using parameters which are calculated in advance; and
extracting and recognizing a corresponding voice recognition objective word to said unknown input voice signal.
1 Assignment
0 Petitions
Accused Products
Abstract
A voice recognizing method in which a plurality of voice recognition objective words are provided. Scores are accumulated for an unknown input voice signal as compared to the voice recognition objective words by using parameters which are calculated in advance. Upon receipt of an unknown voice signal, a corresponding voice recognition objective word is extracted and recognized. The voice recognition objective words are structured into an overlapping hierarchical structure by using correlation values between each pair of voice recognition objective words. This correlation may be computed from acoustic features, HMM parameters or the like. Score calculation is performed on the unknown input voice signal by using a dictionary of the voice recognition objective words structured in the hierarchical structure. Upon preliminary recognition, the dictionary of the voice recognition objective words is resorted without recalculation of the correlation values.
-
Citations
20 Claims
-
1. A voice recognizing method, comprising the steps of:
-
providing a plurality of voice recognition objective words; structuring said voice recognition objective words into a hierarchical structure having a plurality of levels based upon correlation values indicative of similarities between each pair of voice recognition objective words; inputting an unknown voice signal; accumulating scores for the voice recognition objective words which are related in said hierarchical structure to a voice recognition word in a higher level, which word in said higher level receives a high score for said unknown input voice signal by using parameters which are calculated in advance; and extracting and recognizing a corresponding voice recognition objective word to said unknown input voice signal. - View Dependent Claims (2)
-
-
3. A voice recognizing method, comprising the steps of:
-
providing a plurality of voice recognition objective words; structuring said voice recognition objective words into a hierarchical structure by using correlation values between each pair of voice recognition objective words; inputting an unknown voice signal; accumulating scores for the voice recognition objective words for said unknown input voice signal by using parameters which are calculated in advance; extracting and recognizing a corresponding voice recognition objective word to said unknown input voice signal; calculating scores for said unknown input voice signal compared to the objective words by calculating scores for voice recognition objective words of a proper, first hierarchy level in the hierarchical structure, calculating scores for words on a next hierarchy level immediately under the first hierarchy level associated with words having high scores in said first hierarchical level, calculating scores for words on a hierarchy level immediately under said next hierarchy level associated with words having high scores in said next hierarchical level, and repeating said score calculating until lowest hierarchy level is reached; and selecting at least a word having the highest score from the words whose scores are calculated. - View Dependent Claims (4, 5, 6)
-
-
7. A method of forming information of pattern recognition objects for recognizing patterns in which scores are accumulated for the pattern recognition objects when compared to an unknown input pattern by using parameters which are calculated in advance, thereby allowing a corresponding pattern recognition object to be recognized, comprising the steps of:
-
determining correlation values between each pair of recognition objects; grouping the pattern recognition objects in decreasing order, on the basis of said correlation values; selecting a pattern recognition object serving as a typical object of a group; regrouping said pattern recognition objects to form groups each having a relationship between a typical pattern recognition object and a set of pattern recognition objects belonging to the group containing said typical pattern recognition object; grouping each pattern recognition object which is not selected as a typical object of a group to belong to the group of the typical pattern recognition object having a small correlation value between the typical object and the non-selected object; grouping the pattern recognition objects obtained by performing the grouping and adding to the groups described above within a group; and repeating these grouping steps a predetermined number of times to structure the pattern recognition objects into a hierarchical structure, said hierarchical structure being formed so that when recognizing an unknown pattern, only pattern recognition objects on a lower level of said hierarchical structure related to a pattern recognition object receiving a high score on a higher level of said hierarchical structure are searched. - View Dependent Claims (8, 9, 10)
-
-
11. A voice recognizing apparatus comprising:
-
means for providing a plurality of voice recognition objective words; means for structuring said voice recognition objective words into a hierarchical structure having a plurality of levels based upon correlation values indicative of similarities between each pair of voice recognition objective words; means for inputting an unknown voice signal; means for accumulating scores for the voice recognition objective words which are related in said hierarchical structure to a voice recognition word in a higher level, which word in said higher level receives a high score for said unknown input voice signal by using parameters which are calculated in advance; and means for extracting and recognizing a corresponding voice recognition objective word to said unknown input voice signal. - View Dependent Claims (12)
-
-
13. A voice recognizing apparatus, comprising:
-
means for providing a plurality of voice recognition objective words; means for structuring said voice recognition objective words into a hierarchical structure by using correlation values between each pair of voice recognition objective words; means for inputting an unknown voice signal; means for accumulating scores for the voice recognition objective words for said unknown input voice signal by using parameters which are calculated in advance; means for extracting and recognizing a corresponding voice recognition objective word to said unknown input voice signal; calculating means for calculating scores for said unknown input voice signal compared to the objective words, said calculating means calculating scores for voice recognition objective words of a proper, first hierarchy level in the hierarchical structure, calculating scores for words on a next hierarchy level immediately under the first hierarchy level associated with words having high scores in said first hierarchical level, calculating scores for words on a hierarchy level immediately under said next hierarchy level associated with words having high scores in said next hierarchical level, and repeating said score calculating until lowest hierarchy level is reached; and means for selecting at least a word having the highest score from the words whose scores are calculated. - View Dependent Claims (14, 15, 16)
-
-
17. A recording medium for recording information thereon, said recording medium having recorded thereon information of pattern recognition objects for use with a computing apparatus, said information of pattern recognition objects allowing said computing apparatus to recognize patterns by accumulating scores for the pattern recognition objects when compared to an unknown input pattern by using parameters which are calculated in advance, thereby allowing a corresponding pattern recognition object to be recognized, wherein said information of pattern recognition objects is determined by:
-
determining correlation values between each pair of recognition objects; grouping the pattern recognition objects in decreasing order, on the basis of said correlation values; selecting a pattern recognition object serving as a typical object of each group; regrouping said pattern recognition objects to form groups each having a relationship between a typical pattern recognition object and a set of pattern recognition objects belonging to the group containing said typical pattern recognition object; grouping each pattern recognition object which is not selected as a typical object of a group to belong to the group of the typical pattern recognition object having a small correlation value between the typical object and the non-selected object; and grouping the pattern recognition objects obtained by performing the grouping and adding to the groups described above within a group; and wherein the pattern recognition objects are formed into a hierarchical structure, said hierarchical structure being formed so that when recognizing an unknown pattern, only pattern recognition objects on a lower level of said hierarchical structure related to a pattern recognition object receiving a high score on a higher level of said hierarchical structure are searched. - View Dependent Claims (18, 19, 20)
-
Specification