Adapting an utterance cut-off period based on parse prefix detection
A processing system detects a period of non-voice activity and compares its duration to a cutoff period. The system adapts the cutoff period based on parsing previously-recognized speech to determine, according to a model, such as a machine-learned model, the probability that the speech recognized so far is a prefix to a longer complete utterance. The cutoff period is longer when a parse of previously recognized speech has a high probability of being a prefix of a longer utterance.
1. A method of adjusting a cut-off period for detecting an end of a sentence, the method comprising:
receiving a plurality of spoken words;
detecting periods of voice activity and periods of no voice activity within the plurality of spoken words;
parsing the plurality of spoken words according to a first grammar and a second grammar to determine whether the plurality of spoken words are a complete parse according to the first grammar and a prefix to another complete parse according to the second grammar;
accessing an NVAD cut-off period calculated from prior speech; and
scaling the NVAD cut-off period in accordance with a prefix hypothesis numerical score.
2. The method of claim 1 , further comprising producing the prefix hypothesis numerical score indicating a likelihood that the received plurality of spoken words is both: (a) the complete parse and (b) the prefix to the other complete parse.
3. The method of claim 2 , wherein producing the prefix hypothesis numerical score comprises analyzing natural grammar rules determining that the received plurality of spoken words is a complete parse prefix.
4. The method of claim 3 , wherein analyzing natural grammar rules comprises aggregating multiple grammar rules.
5. The method of claim 1 , wherein scaling the NVAD cut-off period in accordance with the prefix hypothesis numerical score comprises scaling the NVAD cut-off period based on the plurality of spoken words being the prefix to the other complete parse.
6. The method of claim 1 , wherein detecting periods of voice activity and periods of no voice activity within the plurality of spoken words comprises detecting inter-word pause lengths.
7. The method of claim 1 , further comprising computing an average phoneme rate from the periods of voice activity and the periods of no voice activity within the plurality of spoken words; and
wherein scaling the NVAD cut-off period comprises scaling the NVAD cut-off period based on the average phoneme rate and in accordance with the prefix hypothesis numerical score.
8. A system comprising:
a processor; and
system memory coupled to the processor and storing instructions configured to cause the processor to:
receive a plurality of spoken words;
detect periods of voice activity and periods of no voice activity within the plurality of spoken words;
parse the plurality of spoken words according to a first grammar and a second grammar to determine whether the plurality of spoken words are a complete parse according to the first grammar and a prefix to another complete parse according to the second grammar;
access an NVAD cut-off period calculated from prior speech; and
scale the NVAD cut-off period in accordance with a prefix hypothesis numerical score.
9. The system of claim 8 , further comprising instructions configured to produce the prefix hypothesis numerical score indicating a likelihood that the received plurality of spoken words is both: (a) the complete parse and (b) the prefix of the other complete parse.
10. The system of claim 9 , wherein instructions configured to produce the prefix hypothesis numerical score comprise instructions configured to analyze natural grammar rules to determine that the received plurality of spoken words is a complete parse prefix.
11. The system of claim 10 , wherein instructions configured to analyze natural grammar rules comprise instructions configured to aggregate multiple grammar rules.
12. The system of claim 8 , wherein instructions configured to scale the NVAD cut-off period in accordance with the prefix hypothesis numerical score comprise instructions configured to scale the NVAD cut-off period based on the plurality of spoken words being the prefix to the other complete parse.
13. The system of claim 8 , wherein instructions configured to detect periods of voice activity and periods of no voice activity within the plurality of spoken words comprise instructions configured to detect inter-word pause lengths.
14. The system of claim 8 , further comprising instructions configured to compute an average phoneme rate from the periods of voice activity and the periods of no voice activity within the plurality of spoken words; and
wherein instructions configured to scale the NVAD cut-off period comprise instructions configured to scale the NVAD cut-off period based on the average phoneme rate and in accordance with the prefix hypothesis numerical score.