Context based voice activity detection sensitivity
A speech dialog system is described that adjusts a voice activity detection threshold during a speech dialog prompt to reflect a context-based probability of user barge in speech occurring. For example, the context-based probability may be based on the location of one or more transition relevance places in the speech dialog prompt.
1. A method for a speech dialog system implemented as a plurality of computer processes functioning in a computer processor, the method comprising:
in one or more computer processes:
adjusting a voice activity detection threshold of the speech dialog system during a speech dialog prompt to reflect a context-based probability of user barge in speech occurring;
adjusting the voice activity detection threshold based upon a heuristic that includes a distribution of legitimate barge-ins and false barge-ins;
receiving speech from a user during the speech dialog prompt; and
processing the speech from the user using the heuristic to determine whether the speech is a legitimate barge-in,
wherein the legitimate barge-ins are known to recur in same places in the speech dialog prompt and false barge-ins are known to be distributed over a duration of the speech dialog prompt.
2. A method according to claim 1 , wherein the voice activity detection threshold is for a limited sub-set recognition vocabulary.
3. A method according to claim 1 , wherein the context-based probability is based on one or more transition relevance places (TRPs) in the speech dialog prompt.
4. A method according to claim 1 , wherein the context-based probability is based on syntax of the speech dialog prompt.
5. A method according to claim 1 , wherein the context-based probability is based on pragmatics of the speech dialog prompt.
6. A method according to claim 1 , wherein the context-based probability is a built-in design value.
7. A method according to claim 1 , wherein the context-based probability is a variable function responsive to operating history of the speech dialog system.
8. A speech recognition engine, comprising:
a memory; and
a processor coupled to the memory, the processor and the memory configured to:
adjust a voice activity detection threshold of the speech dialog system during a speech dialog prompt to reflect a context-based probability of user barge in speech occurring;
adjust the voice activity detection threshold based upon a heuristic that includes a distribution of legitimate barge-ins and false barge-ins;
receive speech from a user during the speech dialog prompt; and
process the speech from the user using the heuristic to determine whether the speech is a legitimate barge-in,
wherein the legitimate barge-ins are known to recur in same places in the speech dialog prompt and false barge-ins are known to be distributed over a duration of the speech dialog prompt.
9. The system according to claim 8 , wherein the voice activity detection threshold is for a limited sub-set recognition vocabulary.
10. The system according to claim 8 , wherein the context-based probability is based on one or more transition relevance places (TRPs) in the speech dialog prompt.
11. The system according to claim 8 , wherein the context-based probability is based on syntax of the speech dialog prompt.
12. The system according to claim 8 , wherein the context-based probability is based on pragmatics of the speech dialog prompt.
13. The system according to claim 8 , wherein the context-based probability is a variable function responsive to operating history of the speech dialog system.
14. An article, comprising:
a non-transitory computer storage medium having stored instructions that enable a machine to:
adjust a voice activity detection threshold of the speech dialog system during a speech dialog prompt to reflect a context-based probability of user barge in speech occurring;
adjust the voice activity detection threshold based upon a heuristic that includes a distribution of legitimate barge-ins and false barge-ins;
receive speech from a user during the speech dialog prompt; and
process the speech from the user using the heuristic to determine whether the speech is a legitimate barge-in,
wherein the legitimate barge-ins are known to recur in same places in the speech dialog prompt and false barge-ins are known to be distributed over a duration of the speech dialog prompt.
15. The article according to claim 14 , wherein the voice activity detection threshold is for a limited sub-set recognition vocabulary.
16. The article according to claim 14 , wherein the context-based probability is based on one or more transition relevance places (TRPs) in the speech dialog prompt.
17. The article according to claim 14 , wherein the context-based probability is based on syntax of the speech dialog prompt.
18. The article according to claim 14 , wherein the context-based probability is based on pragmatics of the speech dialog prompt.
19. The method according to claim 1 , wherein the false barge-ins are generally evenly distributed.
20. The method according to claim 1 , further including adjusting the voice activity detection threshold based upon a likelihood of successful recognition of the barge-in.
21. A method for a speech dialog system implemented as a plurality of computer processes functioning in a computer processor, the method comprising:
in one or more computer processes:
adjusting a voice activity detection threshold of the speech dialog system during a speech dialog prompt based upon a heuristic that includes a distribution of legitimate barge-ins and false barge-ins,
receiving speech from a user during the speech dialog prompt; and
processing the speech from the user using the heuristic to determine whether the speech is a legitimate barge-in,
wherein the legitimate barge-ins are known to recur in same places in the speech dialog prompt and false barge-ins are known to be distributed over a duration of the speech dialog prompt.
22. The method according to claim 1 , further including adjusting the voice activity detection threshold to minimize a sum of false rejection and false acceptance barge-ins.
23. The method according to claim 22 , further including weighting the false rejections and/or false acceptance barge-ins prior to computing the sum.
24. The method according to claim 1 , wherein the heuristic includes computing a median frequency of barge-in events per time segment and weighting scores based upon the median frequency computation.
25. The method according to claim 1 , wherein the heuristic includes scoring barge-in events based upon a success of an in-grammar utterance.