Speech recognition concept confidence measurement
Systems and methods for determining a confidence score associated with a decoding output of a speech recognition engine. In one embodiment, a method of determining the confidence score comprises arranging time frame and acoustic score data into an array, determining a phoneme sequence in the array that yields the highest sum of acoustic scores under certain constraints, e.g., minimum number of time frames and order of phonemes in a phoneme string. A relative score is derived by applying a functional relationship between the acoustic score and different sums comprising acoustic scores from the array. The confidence score, in some embodiments, depends at least in part on the relative score and a measure of ambiguity associated with similar sounding phrases being included in different concepts of a specified grammar.
1. A method of determining a confidence score for decoding of a speech input by a speech recognition engine, in which the engine decodes the speech input using a grammar comprising a plurality of phonemes, the method comprising:
receiving an ordered string of phonemes, wherein the phonemes of the string are identified by a speech recognition engine as being part of a speech input, wherein each phoneme is associated with a time frame, and wherein the speech input spans a time period comprising a plurality of time frames;
receiving a phoneme acoustic score map, wherein the map comprises an acoustic score for each phoneme of a grammar at each of the plurality of time frames;
obtaining a first sum comprising the addition of the highest acoustic scores in each of the time frames;
obtaining a second sum comprising the addition of the lowest scores in each of the time frames;
determining a confidence score for a best path for the ordered string of phonemes, wherein the confidence score is determined as a weighted average based at least in part on a functional relationship between the best path score, the second sum, and the first sum.
2. The method of claim 1 , wherein the best path comprises a sequence of time frames ordered in the same manner as the order of the phonemes in the ordered string of phonemes.
3. The method of claim 1 , wherein the best path is a particular path from among a plurality of permissible paths having the highest sum of acoustic scores of the phonemes in the particular path.
4. The method of claim 1 , wherein the grammar comprises a plurality of phrases, each phrase comprising a string of phonemes.
5. A method of determining a confidence score for decoding of a speech input by a speech recognition engine in which the engine decodes the speech input using a grammar comprising a plurality of phonemes the method comprising:
receiving an ordered string of phonemes wherein the phonemes of the string are identified by a speech recognition engine as being part of a speech input wherein each phoneme is associated with a time frame, and wherein the speech input spans a time period comprising a plurality of time frames;
receiving a phoneme acoustic score map, wherein the map comprises an acoustic score for each phoneme of a grammar at each of the plurality of time frames, wherein the grammar comprises a plurality of phrases each phrase comprising a string of phonemes;
determining a confidence score for a best path for the ordered string of phonemes; and
determining a confidence score for each of the phrases of the grammar, wherein the phrases are grouped into concepts, at least one of the concepts comprising the ordered string of phonemes.
6. The method of claim 5 , further comprising obtaining a first sum comprising the addition of the highest acoustic scores in each of the time frames.
7. The method of claim 6 , further comprising obtaining a second sum comprising the addition of the lowest scores in each of the time frames.
8. The method of claim 7 , wherein the confidence score is determined as a weighted average based at least in part on a functional relationship between the best path score, the second sum, and the first sum.
9. The method of claim 5 , further comprising comparing the confidence score for each of the phrases of the concepts not comprising the ordered string of phonemes against the confidence score of the ordered string of phonemes.
10. The method of claim 9 , wherein the confidence score of the ordered string of phonemes is reduced if the confidence score of one of the phrases of the concepts not comprising the ordered string is greater or equal to the confidence score of the ordered string.
11. The method of claim 5 , wherein the best path comprises a sequence of time frames ordered in the same manner as the order of the phonemes in the ordered string of phonemes.
12. The method of claim 5 , wherein the best path is a particular path from among a plurality of permissible paths having the highest sum of acoustic scores of the phonemes in the particular path.