Semantically conditioned voice activity detection
A method includes recognizing words comprised by a first utterance; interpreting the recognized words according to a grammar comprised by a domain; from the interpreting of the recognized words, determining a timeout period for the first utterance based on the domain of the first utterance; detecting end of voice activity in the first utterance; executing an instruction following an amount of time after detecting end of voice activity of the first utterance in response to the amount of time exceeding the timeout period, the executed instruction based at least in part on interpreting the recognized words.
1 . A method comprising:
recognizing words within an utterance of a user by a speech recognition engine to generate one or more transcription hypotheses;
interpreting, by a natural language understanding processor, the recognized words by comparing the transcription hypotheses to a plurality of semantic grammars respectively associated with a plurality of domains, each semantic grammar producing a probability score representing a likelihood that the utterance corresponds to the respective domain;
from the interpreting of the recognized words, determining by the processor, a weighted-average timeout period for the utterance as a weighted average of domain-specific timeout values, each domain-specific timeout value being multiplied by its corresponding probability score and divided by a sum of the probability scores;
detecting an end of voice activity in the first utterance using a voice activity detector coupled to a microphone sensor; and
executing, by the processor, an instruction following an amount of time after detecting the end of voice activity, the instruction being executed in response to the amount of time exceeding the determined weighted-average timeout period, and being based at least in part on the interpreted words.
2 . The method of claim 1 , wherein interpreting includes determining a probability of correctness, the method further comprising interpreting the recognized words according to a second grammar comprised by a second domain, wherein the timeout period is selected based on which of multiple interpretations has the highest probability of correctness.
3 . The method of claim 1 , wherein interpreting includes determining a probability of correctness, the method further comprising interpreting the recognized words according to a second grammar comprised by a second domain; and
computing a weighted average of multiple probabilities of correctness, wherein the timeout period is selected based on the weighted average.
4 . The method of claim 1 , wherein the timeout period is specified as a parameter of the domain.
5 . The method of claim 4 , wherein the timeout period is specified as a multiple of a general timeout period.
6 . The method of claim 1 further comprising, computing a user speech rate, wherein the timeout period is based at least in part on the user speech rate.
7 . The method of claim 1 further comprising, computing whether the recognized words can be a prefix to another interpretation, wherein the timeout period is based at least in part on whether the recognized words can be a prefix to another interpretation.
8 . The method of claim 1 further comprising:
entering a modal dialog having a modal timeout period different from a default timeout period,
wherein the timeout period is based, at least in part, on the default timeout period while in a default mode, and wherein the timeout period is based, at least in part, on the modal timeout period while in the modal dialog.