Character and phoneme recognition based on probability clustering

• US 5,075,896 A
• Filed: 10/25/1989
• Issued: 12/24/1991
• Est. Priority Date: 10/25/1989
• Status: Expired due to Term

• Current Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Wilcox, Lynn D., Spitz, A. Lawrence
• Primary Examiner(s): BRINICH, STEPHEN M
• Application Number:
  US07/427,148
• Time in Patent Office: 790 Days
• Field of Search: 381/41, 381/43, 382/39, 382/14
• US Class Current: 382/225
• CPC Class Codes: G10L 15/187 Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams

Abstract

Prior to character or phoneme recognition, a classifier provides a respective probability list for each of a sequence of sample characters or phonemes, each probability list indicating the respective sample'"'"'s probability for each character or phoneme type. These probability lists are clustered in character or phoneme probability space, in which each dimension corresponds to the probability that a character or phoneme candidate is an instance of a specific character or phoneme type. For each resulting cluster, data is stored indicating its cluster ID and a probability list indicating the probability of each type at the cluster'"'"'s center. Then, during recognition, a probability cluster identifier compares the probability list for each candidate with the probability list for each cluster to find the nearest cluster. The cluster identifier then provides the nearest cluster'"'"'s cluster ID to a constraint satisfier that attempts to recognize the candidate based on rules, patterns, or a combination of rules and patterns. If necessary, the constraint satisfier uses the cluster ID to retrieve the stored probability list of the cluster to assist it in recognition.

Claims

1. A method comprising steps of:

obtaining a set of probability values for a character candidate, the set of probability values including a respective probability value for each of a set of character types, each character type'"'"'s probability value indicating a probability that the character candidate is an instance of the character type; and

using the set of probability values to identify which of a number of character probability clusters is nearest to the candidate;

the character probability clusters being regions of a space;

the space having a respective dimension for each of the character types;

the character candidate'"'"'s respective probability value for each character type indicating the character candidate'"'"'s position in the character type'"'"'s dimension so that the set of probability values define the character candidate'"'"'s position in the space.

2. The method of claim 1, further comprising a step of storing respective probability data for each of the character probability clusters prior to the step of using the set of probability values.

3. The method of claim 2 in which the storing step comprises clustering sample probability data in order to obtain the character probability clusters.

4. The method of claim 2, further comprising a step of recognizing the candidate based on the respective probability data for the identified nearest probability cluster.

5. The method of claim 3 in which the respective probability data of each cluster is a set of probability values, the set of probability values including a respective probability value for each of a set of recognizable character types.

6. The method of claim 1 in which the step of using the set of probability values comprises a substep of obtaining data identifying the nearest probability cluster, the method further comprising a step of obtaining data identifying a character type of the candidate character based on the data identifying the nearest cluster.

7. The method of claim 6 in which the cluster identifying data is a cluster ID.

8. A system comprising:

a classifier for providing a set of probability values for a character candidate;

the set of probability values including a respective probability value for each of a set of character types, each character type'"'"'s probability value indicating a probability that the character candidate is an instance of the character type;

a probability cluster identifier for using the set of probability values to produce data identifying which of a number of character probability clusters is nearest to the character candidate;

the character probability clusters being regions of a space;

the space having a respective dimension for each of the character types;

the character candidate'"'"'s respective probability value for each character type indicating the character candidate'"'"'s position in the character type'"'"'s dimension so that the set of probability values define the character candidate'"'"'s position in the space; and

a constraint satisfier for providing data identifying a character type of the character candidate based on the cluster identifying data from the probability cluster identifier.

9. The system of claim 8 in which the probability cluster identifier produces a cluster ID identifying the nearest cluster.

10. The system of claim 8, further comprising stored probability data including respective probability data for each of the character probability clusters, the constraint satisfier using the cluster identifying data to access the respective probability data of the identified nearest cluster.

11. The system of claim 10 in which the respective probability data for each cluster includes a respective set of probability values for the cluster, each cluster'"'"'s respective set of probability values including a respective probability value for each of the character types.

12. The system of claim 10 in which the classifier is further for providing respective probability data for a set of sample characters;

the system further comprising a trainer for storing the respective probability data for each of the character probability clusters based on the respective probability data for the sample characters.

13. The system of claim 12 in which the trainer is further for obtaining the character probability clusters and the respective probability data for each of the clusters by clustering the probability data for the sample characters.

14. A method comprising steps of:

obtaining a set of probability values for a phoneme candidate, the set of probability values including a respective probability value for each of a set of phoneme types, each phoneme type'"'"'s probability value indicating a respective probability that the phoneme candidate is an instance of the phoneme type; and

using the set of probability values to identify which of a number of phoneme probability clusters is nearest to the candidate;

the phoneme probability clusters being regions of a space;

the space having a respective dimension for each of the phoneme types;

the phoneme candidate'"'"'s respective probability value for each phoneme type indicating the phoneme candidate'"'"'s position in the phoneme type'"'"'s dimension so that the set of probability values define the phoneme candidate'"'"'s position in the space.

15. The method of claim 14, further comprising a step of storing respective probability data for each of the phoneme probability clusters prior to the step of using the set of probability values.

16. The method of claim 15 in which the storing step comprises clustering sample probability data in order to obtain the phoneme probability clusters.

17. The method of claim 15, further comprising a step of recognizing the candidate based on the respective probability data for the identified nearest probability cluster.

18. The method of claim 15 in which the respective probability data of each cluster is a set of probability values, the set of probability values including a respective probability value for each of a set of recognizable phoneme types.

19. The method of claim 14 in which the step of using the set of probability values comprises a substep of obtaining data identifying the nearest probability cluster, the method further comprising a step of obtaining data identifying a phoneme type of the candidate based on the data identifying the nearest cluster.

20. The method of claim 19 in which the cluster identifying data is a cluster ID.

21. A system comprising:

a classifier for providing a set of probability values for a phoneme candidate;

the set of probability values including a respective probability value for each of a set of phoneme types, each phoneme type'"'"'s probability value indicating a probability that the phoneme candidate is an instance of the phoneme type;

a probability cluster identifier for using the set of probability values to produce data identifying which of a number of phoneme probability clusters is nearest to the phoneme candidate;

the phoneme probability clusters being regions of a space;

the space having a respective dimension for each of the phoneme types;

the phoneme candidate'"'"'s respective probability value for each phoneme type indicating the phoneme candidate'"'"'s position in the phoneme type'"'"'s dimension so that the set of probability values define the phoneme candidate'"'"'s position in the space; and

a constraint satisfier for providing data identifying a phoneme type of the phoneme candidate based on the cluster identifying data from the probability cluster identifier.

22. The system of claim 21 in which the probability cluster identifier produces a cluster ID identifying the nearest cluster.

23. The system of claim 21, further comprising stored probability data including respective probability data for each of the phoneme probability clusters, the constraint satisfier using the cluster identifying data to access the respective probability data of the identified nearest cluster.

24. The system of claim 23 in which the respective probability data for each cluster includes a respective set of probability values for the cluster, each cluster'"'"'s respective set of probability values including a respective probability value for each of the phoneme types.

25. The system of claim 23 in which the classifier is further for providing respective probability data for a set of sample phonemes;

the system further comprising a trainer for storing the respective probability data for each of the phoneme probability clusters based on the respective probability data for the sample phonemes.

26. The system of claim 25 in which the trainer is further for obtaining the phoneme probability clusters and the respective probability data for each of the clusters by clustering the probability data for the sample phonemes.