Using word confidence score, insertion and substitution thresholds for selected words in speech recognition
A method and system for improving the accuracy of a speech recognition system using word confidence score (WCS) processing is introduced. Parameters in a decoder are selected to minimize a weighted total error rate, such that deletion errors are weighted more heavily than substitution and insertion errors. The occurrence distribution in WCS is different depending on whether the word was correctly identified and based on the type of error. This is used to determine thresholds in WCS for insertion and substitution errors. By processing the hypothetical word (HYP) (output of the decoder), a mHYP (modified HYP) is determined. In some circumstances, depending on the WCS's value in relation to insertion and substitution threshold values, mHYP is set equal to: null, a substituted HYP, or HYP.
1. A method for recognizing speech in acoustic data, comprising:
receiving the acoustic data at a decoder,
generating a plurality of hypothetical words (HYP) in the decoder from the acoustic data;
deriving a word confidence score (WCS) for each HYP at a post-decoder;
for each HYP, at the post-decoder:
determining whether the HYP is on a selected word list;
if the HYP is not on the selected word list, setting a modified hypothetical word (mHYP) for the HYP equal to the HYP;
if the HYP is on the selected word list, determining mHYP by comparing the WCS score for the HYP to at least one WCS threshold value associated with HYP on the selected word list; and
outputting the determined mHYP for each HYP.
2. The method of claim 1 , wherein each HYP on the selected word list has associated therewith an insertion threshold value and substitution threshold value.
3. The method of claim 2 , wherein, for each HYP on the selected word list, determining the mHYP includes setting the mHYP is equal to the HYP when the WCS is greater than both the associated insertion and substitution threshold values.
4. The method of claim 2 , wherein, for each HYP on the selected word list, determining the mHYP includes setting the mHYP to a null when the WCS is less than the associated insertion threshold value.
5. The method of claim 2 , wherein, for each HYP on the selected word list, determining the mHYP includes setting the mHYP to be a substituted HYP when the WCS is less than the associated substitution threshold value, and greater than the associated insertion threshold value.
6. A method for recognizing speech in acoustic data, comprising:
inputting the acoustic data to a decoder to generate a plurality of hypothetical words (HYP) therefrom;
at a post-decoder, deriving a word confidence score (WCS) for the HYP generated by the decoder;
at the post-decoder, determining a modified hypothetical word (mHYP) for the HYP using associated insertion and substitution threshold values associated with the HYP in a selected word list, determining the mHYP including:
setting the mHYP is equal to the HYP when the WCS is greater than both the associated insertion and substitution threshold values;
setting the mHYP to a null when the WCS is less than the associated insertion threshold value; and
setting the mHYP to be a substituted HYP when the WCS is less than the associated substitution threshold value, and greater than the associated insertion threshold value; and
outputting the mHYP determined for the HYP.
7. The method of claim 6 , further comprising:
determining whether the HYP is on the selected word list prior to determining the mHYP; and
if mHYP is not present on the selected word list, setting mHYP equal to the HYP.