Path link passing speech recognition with vocabulary node being capable of simultaneously processing plural path links

• US 6,230,128 B1
• Filed: 11/21/1995
• Issued: 05/08/2001
• Est. Priority Date: 03/31/1993
• Status: Expired due to Term

• Current Assignee: Cisco Technology, Inc. (Cisco Systems, Inc.)
• Original Assignee: British Telecommunications PLC (BT Group PLC)
• Inventors: Smyth, Samuel Gavin
• Primary Examiner(s): Tsang, Fan
• Assistant Examiner(s): Opsasnick, Michael N.
• Application Number:
  US08/530,157
• Time in Patent Office: 1,995 Days
• Field of Search: 395/2.4, 395/2.41, 395/2.45, 395/2.46, 395/2.47, 395/2.51, 395/2.64, 395/2.65, 395/2.66
• US Class Current: 704/236
• CPC Class Codes: G10L 15/08 Speech classification or search

Abstract

A path link passing speech recognition system and method recognizes input connected speech. The recognition system has a plurality of vocabulary nodes associated with word representation models, at least one of the vocabulary nodes of the network being able to process more than one path link simultaneously, so allowing for more than one recognition result.

Claims

1. A speech recognition system comprising:

means for deriving a recognition feature vector from an input speech signal for each of predetermined time frames;

means for modelling predetermined possible input speech comprising a plurality of vocabulary nodes each of which has an associated word representation model and links between said vocabulary nodes;

processing means for comparing the recognition feature vectors with the modelled input speech and for generating a path link for each node and time frame, said path links indicating the most likely prior sequence of vocabulary nodes for each vocabulary node and time frame, each path link comprising a field for storing an accumulated recognition score and a field for storing a reference to the most likely previous path link in the sequence; and

means for indicating recognition of the input speech signal in dependence upon the comparison;

the processing means being capable of processing more than one path link for at least one vocabulary node, other than the final node, in a single time frame.

2. A speech recognition system as in claim 1 wherein at least one of the vocabulary nodes is associated with more than one identical word representation model.

3. A speech recognition system as in claim 2 wherein the word representation models are Hidden Makov Models.

4. A speech recognition system as in claim 1 wherein all the vocabulary nodes have signatures assigned to them.

5. A speech recognition system as in claim 1 wherein only the vocabulary nodes occurring before a decision point have signatures assigned to them.

6. A speech recognition system as in claim 4 wherein the path links include an accumulated signature.

7. A speech recognition system as in claim 4 wherein at least some of the nodes are constrained only to propagate path links having certain predetermined signatures.

8. A speech recognition system as in claim 4 wherein the recognition indicating means includes means for comparing the said recognition score and signature of the path links to determine the path with the best match to the input connected speech and those with the next best alternative matches.

9. A method of speech recognition comprising:

deriving a recognition feature vector from an input speech signal for each of predetermined time frames,

modelling predetermined possible input speech;

comparing the feature data with the modelled input speech by generating a network containing a plurality of vocabulary nodes associated with word representation models and generating a path link for each node and time frame, the path link indicating the most likely prior sequence of vocabulary nodes for each vocabulary node and time frame, each path link comprising a field for storing an accumulated recognition score and a field for storing a reference to the most likely previous path link in the sequence; and

indicating recognition of the speech independence upon the comparison wherein more than one path link is processed in a single time frame for at least one vocabulary node other than the final node.

10. A method as in claim 9 wherein the at least one of the vocabulary nodes is associated with more than one identical word representation model.

11. A method as in claim 10 wherein at least one of the vocabulary nodes is associated with a number of identical word representation models equal to the number of desired recognition results.

12. A method as in claim 10 wherein the said recognition scores of the path links are compared at each decision point of the network, only the n top scoring path links being propagated to the next nodes(s).

13. A method as in claim 10 wherein signatures are assigned to all the vocabulary nodes.

14. A method as in claim 10 wherein signatures are only assigned to vocabulary nodes occurring before a decision point in the network.

15. A method as in claim 12 wherein signatures are assigned to all the vocabulary nodes and the signatures of the path links are also compared, only path links having different signatures being propagated to the next node(s).

16. A method as in claim 13 wherein at least some nodes are constrained only to pass path links having certain predetermined signatures in their signature fields.

17. A method as in claim 9 wherein the input speech signal deemed to be recognised is determined by tracing the path links back through the network.

18. A method as in claim 13 wherein the input speech signal deemed to be recognised is determined by the accumulated signature of each path link.

19. A method as in claim 10 wherein the best scoring path link is processed by the first word representation model of a vocabulary node, the next best by the second and so on until either parallel models or incoming path links run out.