Apparatuses, computer program products, and computer-implemented methods for adapting speech recognition confidence scores based on expected response
Embodiments of the disclosure provide for adapting speech recognition confidence scores to one or more users based on expect response(s). Some embodiments, receive input speech and generate, based on the input speech and using a recognition algorithm, a recognition hypothesis. The recognition hypothesis may comprise at least one pairing comprising a word and an initial confidence score associated with the word. Some embodiments, determine whether the recognition hypothesis matches an expected response based at least in part on comparing the word to a corresponding word in an expected response. Some embodiments, in response to determining that the recognition hypothesis matches the expected response, updates statistical data associated with the word.
1 . A computer-implemented method for adapting speech recognition to one or more users, the computer-implemented method comprising:
receiving input speech;
generating, based on the input speech and using a recognition algorithm, a recognition hypothesis wherein the recognition hypothesis comprises at least one pairing comprising a word and an initial confidence score associated with the word;
determining whether the recognition hypothesis matches an expected response based at least in part on comparing the word to a corresponding word in the expected response;
in response to determining that the recognition hypothesis matches the expected response, updating statistical data associated with the word; and
based on the updated statistical data, updating a confidence score adjustment value associated with the word according to a condition, wherein the condition represents that an average adjusted confidence score for the word based on a set of historical recognition hypothesis satisfies a rejection threshold for the word by a predefined value.
2 . The computer-implemented method of claim 1 , wherein the statistical data associated with the word comprises one or more of a sum measure, count measure, or sum of squares measure.
3 . The computer-implemented method of claim 1 , wherein the statistical data for the word is determined based at least in part on each confidence score associated with the word in the recognition hypothesis and/or each initial confidence score associated with the word in at least one historical recognition hypothesis.
4 . The computer-implemented method of claim 1 , further comprising:
storing the updated statistical data in data repository.
5 . The computer-implemented method of claim 1 further comprising:
generating an adjusted confidence score for the word by applying the confidence score adjustment value for the word to the initial confidence score associated with the word; and
determining if the adjusted confidence score satisfies the rejection threshold for the word.
6 . The computer-implemented method of claim 5 , wherein the recognition hypothesis is rejected in response to determining that at least one word in the recognition hypothesis is associated with the adjusted confidence score that fails to satisfy the rejection threshold for the at least one word.
7 . The computer-implemented method of claim 5 , wherein the recognition hypothesis is accepted in response to determining that each word in the recognition hypothesis is associated with the adjusted confidence score that satisfies each word's respective rejection threshold.
8 . The computer-implemented method of claim 5 further comprising:
generating the updated confidence score adjustment value for the word based at least in part on the updated statistical data; and
applying the updated confidence score adjustment value to a subsequent recognition hypothesis that includes another instance of the same word and an associated initial confidence score by adjusting the initial confidence score for the word in the subsequent recognition hypothesis based at least in part on the updated confidence score adjustment value.
9 . The computer-implemented method of claim 5 , wherein the updated confidence score adjustment value for the word is generated based at least in part on a condition, wherein the condition represents that average adjusted confidence score for the word based on the recognition hypothesis and historical recognition hypotheses correspond to N standard deviations above the rejection threshold.
10 . The computer-implemented method of claim 5 , wherein the word is accepted in response to determining that the word is associated with an adjusted confidence score that satisfies the rejection threshold for the word.
11 . An apparatus for adapting speech recognition to one or more users, the apparatus comprising at least one processor and at least one non-transitory memory comprising program code stored thereon, wherein the at least one non-transitory memory and the program code are configured to:
receive input speech;
generate, based on the input speech and using a recognition algorithm, a recognition hypothesis wherein the recognition hypothesis comprises at least one pairing comprising a word and an initial confidence score associated with the word;
determine whether the recognition hypothesis matches an expected response based at least in part on comparing the word to a corresponding word in the expected response;
in response to determining that the recognition hypothesis matches the expected response, update statistical data associated with the word; and
based on the updated statistical data, update a confidence score adjustment value associated with the word according to a condition, wherein the condition represents that an average adjusted confidence score for the word based on a set of historical recognition hypothesis satisfies a rejection threshold for the word by a predefined value.
12 . The apparatus of claim 11 , wherein the statistical data associated with a particular word comprises one or more of a sum measure, count measure, or sum of squares measure.
13 . The apparatus of claim 11 , wherein the statistical data for a particular word is determined based at least in part on each confidence score associated with the word in the recognition hypothesis and/or each initial confidence score associated with the word in at least one historical recognition hypothesis.
14 . The apparatus of claim 11 , wherein the at least one non-transitory memory and the program code are configured to:
store the updated statistical data in a data repository.
15 . The apparatus of claim 11 wherein the at least one non-transitory memory and the program code are configured to:
generate an adjusted confidence score for the word by applying the confidence score adjustment value for the word to an initial confidence score associated with the word; and
determine if the adjusted confidence score satisfies a rejection threshold.
16 . The apparatus of claim 15 wherein the at least one non-transitory memory and the program code are configured to:
generate the updated confidence score adjustment value for the word based at least in part on the updated statistical data; and
apply the updated confidence score adjustment value to a subsequent recognition hypothesis that includes another instance of the same word and an associated initial confidence score by adjusting the initial confidence score for the word in the subsequent recognition hypothesis based at least in part on the updated confidence score adjustment value.
17 . The apparatus of claim 11 , wherein the updated confidence score adjustment value for the word in each pairing is generated based at least in part on a condition, wherein the condition represents that the average adjusted confidence score for the word based on the recognition hypothesis and historical recognition hypotheses correspond to N standard deviations above the rejection threshold.
18 . A computer program product for adapting speech recognition to one or more users, the computer program product comprising at least one non-transitory computer-readable storage medium having computer-readable program code portions stored therein, the computer-readable program code portions comprising an executable portion configured to:
receive input speech;
generate, based on the input speech and using a recognition algorithm, a recognition hypothesis wherein the recognition hypothesis comprises at least one pairing comprising a word and an initial confidence score associated with the word;
determine whether the recognition hypothesis matches an expected response based at least in part on comparing the word to a corresponding word in the expected response;
in response to determining that the recognition hypothesis matches the expected response, update statistical data associated with the word; and
based on the updated statistical data update a confidence score adjustment value associated with the word according to a condition, wherein the condition represents that an average adjusted confidence score for the word based on a set of historical recognition hypothesis satisfies a rejection threshold for the word by a predefined value.