Language recognition using a similarity measure
First Claim
Patent Images
1. A comparison apparatus comprising:
- a receiver operable to receive an input signal;
a recognition processor operable to compare said input signal with stored label models to generate a recognised sequence of labels in said input signal and confidence data representative of the confidence that the recognised sequence of labels is representative of said input signal; and
a similarity measure calculator operable to compare said recognised sequence of labels received from said recognition processor with a stored sequence of annotation labels using a combination of i) predetermined confusion data which defines confusability between different labels, and ii) said confidence data received from the recognition processor and representative of the confidence that said received recognized sequence of labels is representative of the input signal, to provide a measure of the similarity between the recognised sequence of labels and the stored sequence of annotation labels,wherein said similarity measure calculator comprisesan aligner operable to align labels of the recognised sequence of labels with labels of the stored sequence of annotation labels to form a number of aligned pairs of labels;
a comparator operable to compare the labels of each aligned pair of labels using said combination of said predetermined confusion data and said confidence data, to generate a comparison score representative of the similarity between the aligned pair of labels; and
a combiner operable to combine the comparison scores for all the aligned pairs of labels to provide said similarity measure,wherein said comparator comprisesa first sub-comparator operable to compare, for each aligned pair, the recognised sequence label in the aligned pair with each of a plurality of labels taken from a set of predetermined labels using said confusion data and said confidence data to provide a corresponding plurality of intermediate comparison scores representative of the similarity between said recognised sequence label and the respective labels from the set;
a second sub-comparator operable to compare, for each aligned pair, the stored sequence label in the aligned pair with each of said plurality of labels from the set using said confusion data and said confidence data to provide a further corresponding plurality of intermediate comparison scores representative of the similarity between said stored sequence label and the respective labels from the set; and
a calculator operable to calculate said comparison score for the aligned pair by combining said pluralities of intermediate comparison scores.
1 Assignment
0 Petitions
Accused Products
Abstract
A dynamic programming technique is provided for matching two sequences of phonemes both of which may be generated from text or speech. The scoring of the dynamic programming matching technique uses phoneme confusion scores, phoneme insertion scores and phoneme deletion scores which are obtained in advance in a training session and, if appropriate, confidence data generated by a recognition system if the sequences are generated from speech.
-
Citations
46 Claims
-
1. A comparison apparatus comprising:
-
a receiver operable to receive an input signal; a recognition processor operable to compare said input signal with stored label models to generate a recognised sequence of labels in said input signal and confidence data representative of the confidence that the recognised sequence of labels is representative of said input signal; and a similarity measure calculator operable to compare said recognised sequence of labels received from said recognition processor with a stored sequence of annotation labels using a combination of i) predetermined confusion data which defines confusability between different labels, and ii) said confidence data received from the recognition processor and representative of the confidence that said received recognized sequence of labels is representative of the input signal, to provide a measure of the similarity between the recognised sequence of labels and the stored sequence of annotation labels, wherein said similarity measure calculator comprises an aligner operable to align labels of the recognised sequence of labels with labels of the stored sequence of annotation labels to form a number of aligned pairs of labels; a comparator operable to compare the labels of each aligned pair of labels using said combination of said predetermined confusion data and said confidence data, to generate a comparison score representative of the similarity between the aligned pair of labels; and a combiner operable to combine the comparison scores for all the aligned pairs of labels to provide said similarity measure, wherein said comparator comprises a first sub-comparator operable to compare, for each aligned pair, the recognised sequence label in the aligned pair with each of a plurality of labels taken from a set of predetermined labels using said confusion data and said confidence data to provide a corresponding plurality of intermediate comparison scores representative of the similarity between said recognised sequence label and the respective labels from the set; a second sub-comparator operable to compare, for each aligned pair, the stored sequence label in the aligned pair with each of said plurality of labels from the set using said confusion data and said confidence data to provide a further corresponding plurality of intermediate comparison scores representative of the similarity between said stored sequence label and the respective labels from the set; and a calculator operable to calculate said comparison score for the aligned pair by combining said pluralities of intermediate comparison scores. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18)
-
-
19. A comparison apparatus comprising:
-
a receiver operable to receive an input signal; a recognition processor operable to compare said input signal with stored label models to generate a recognised sequence of labels in said input signal and confidence data representative of the confidence that the recognised sequence of labels is representative of said input signal; and a similarity measure calculator operable to compare said recognised sequence of labels received from said recognition processor with a stored sequence of annotation labels using a combination of i) predetermined confusion data which defines confusability between different labels, and ii) said confidence data received from the recognition processor and representative of the confidence that said received recognized sequence of labels is representative of the input signal, to provide a measure of the similarity between the recognised sequence of labels and the stored sequence of annotation labels, wherein said comparator is operable to compare said recognised sequence of labels with a plurality of stored sequences of labels using said confusion data and said confidence data to provide a respective measure of the similarity between the recognised sequence of labels and said plurality of stored sequences of labels, wherein said comparator comprises a normaliser operable to normalise each of said similarity measures, wherein said normaliser is operable to normalise each similarity measure by dividing each similarity measure by a respective normalisation score which varies in dependence upon the length of the corresponding stored sequence of annotation labels wherein said comparator comprises a first sub-comparator operable to compare, for each aligned pair, the recognised sequence label in the aligned pair with each of a plurality of labels taken from a set of predetermined labels using said confusion data and said confidence data to provide a corresponding plurality of intermediate comparison scores representative of the similarity between said recognised sequence label and the respective labels from the set; a second sub-comparator operable to compare, for each aligned pair, the stored sequence label in the aligned pair with each of said plurality of labels from the set using said confusion data and said confidence data to provide a further corresponding plurality of intermediate comparison scores representative of the similarity between said stored sequence label and the respective labels from the set; and a calculator operable to calculate said comparison score for the aligned pair by combining said pluralities of intermediate comparison scores; and wherein said respective normalisation scores vary with the corresponding intermediate comparison scores calculated by said second sub-comparator. - View Dependent Claims (20, 21, 22)
-
-
23. An apparatus for searching a database comprising a plurality of information entries to identify information to be retrieved therefrom, each of said plurality of information entries having an associated annotation comprising a sequence of labels, the apparatus comprising:
-
a label comparison apparatus for comparing a query sequence of labels obtained from an input query with the labels of each annotation to provide a set of comparison results, wherein the label comparison apparatus comprises a receiver operable to receive an input signal, a recognition processor operable to compare said input signal with stored label models to generate a recognised sequence of labels in said input signal and confidence data representative of the confidence that the recognised sequence of labels is representative of said input signal, a similarity measure calculator operable to compare said recognised sequence of labels received from said recognition processor with a stored sequence of annotation labels using a combination of i) predetermined confusion data which defines confusability between different labels, and ii) said confidence data received from the recognition processor and representative of the confidence that said received recognized sequence of labels is representative of the input signal, to provide a measure of the similarity between the recognised sequence of labels and the stored sequence of annotation labels; and an information identifier operable to identify said information to be retrieved from said database using said comparison results, wherein said label comparison apparatus has a plurality of different comparison modes of operation and wherein the apparatus further comprises a determiner operable to determine if the sequence of labels of a current annotation was generated from an audio signal or from text, and for outputting a determination result; and a selector operable to select, for the current annotation, the mode of operation of said label comparison apparatus in dependence upon said determination result.
-
-
24. A label comparison method comprising:
-
receiving an input signal; a recognition processing step of comparing said input signal with stored label models to generate a recognised sequence of labels in said input signal and confidence data representative of the confidence that the recognised sequence of labels is representative of said input signal; and calculating a measure of the similarity between the recognised sequence of labels generated in said recognition processing step and a stored sequence of annotation labels by comparing said recognised sequence of labels with the stored sequence of annotation labels using a combination of i) predetermined confusion data which defines confusability between different labels, and ii) said confidence data generated in said recognition processing step and representative of the confidence that the recognised sequence of labels is representative of said input signal, wherein said calculating step comprises aligning labels of the recognised sequence of labels with labels of the stored sequence of annotation labels to form a number of aligned pairs of labels; comparing the labels of each aligned pair of labels using said predetermined confusion data and said confidence data to generate a comparison score representative of the similarity between the aligned pair of labels; and combining the comparison scores for all the aligned pairs of labels to provide said similarity measure, wherein said comparing step comprises a first sub-comparing step of comparing, for each aligned pair, the recognised sequence label in the aligned pair with each of a plurality of labels taken from a set of predetermined labels using said confusion data and said confidence data to provide a corresponding plurality of intermediate comparison scores representative of the similarity between said recognised sequence label and the respective labels from the set; a second sub-comparing step of comparing, for each aligned pair, the stored sequence label in the aligned pair with each of said plurality of labels from the set using said confusion data and said confidence data to provide a further corresponding plurality of intermediate comparison scores representative of the similarity between said stored sequence label and the respective labels from the set; and a step of calculating said comparison score for the aligned pair by combining said pluralities of intermediate comparison scores. - View Dependent Claims (25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41)
-
-
42. A label comparison method comprising:
-
receiving an input signal; a recognition processing step of comparing said input signal with stored label models to generate a recognised sequence of labels in said input signal and confidence data representative of the confidence that the recognised sequence of labels is representative of said input signal; and calculating a measure of the similarity between the recognised sequence of labels generated in said recognition processing step and a stored sequence of annotation labels by comparing said recognised sequence of labels with the stored sequence of annotation labels using a combination of i) predetermined confusion data which defines confusability between different labels, and ii) said confidence data generated in said recognition processing step and representative of the confidence that the recognised sequence of labels is representative of said input signal, wherein said comparing step compares said recognised sequence of labels with a plurality of stored sequences of labels using said confusion data and said confidence data to provide a respective measure of the similarity between the recognised sequence of labels and said plurality of stored sequences of labels, wherein said comparing step comprises a normalising step for normalising each of said similarity measures, wherein said normalising step normalises each similarity measure by dividing each similarity measure by a respective normalisation score which varies in dependence upon the length of the corresponding stored sequence of annotation labels, wherein said comparing step comprises a first sub-comparing step of comparing, for each aligned pair, the recognised sequence label in the aligned pair with each of a plurality of labels taken from a set of predetermined labels using said confusion data and said confidence data to provide a corresponding plurality of intermediate comparison scores representative of the similarity between said recognised sequence label and the respective labels from the set; a second sub-comparing step of comparing, for each aligned pair, the stored sequence label in the aligned pair with each of said plurality of labels from the set using said confusion data and said confidence data to provide a further corresponding plurality of intermediate comparison scores representative of the similarity between said stored sequence label and the respective labels from the set; and a step of calculating said comparison score for the aligned pair by combining said pluralities of intermediate comparison scores; and wherein said respective normalisation scores vary with the corresponding intermediate comparison scores calculated by said second sub-comparing step.
-
-
43. A label comparison method comprising:
-
receiving an input signal; a recognition processing step of comparing said input signal with stored label models to generate a recognised sequence of labels in said input signal and confidence data representative of the confidence that the recognised sequence of labels is representative of said input signal; and calculating a measure of the similarity between the recognised sequence of labels generated in said recognition processing step and a stored sequence of annotation labels by comparing said recognised sequence of labels with the stored sequence of annotation labels using a combination of i) predetermined confusion data which defines confusability between different labels, and ii) said confidence data generated in said recognition processing step and representative of the confidence that the recognised sequence of labels is representative of said input signal wherein said comparing step compares said recognised sequence of labels with a plurality of stored sequences of labels using said confusion data and said confidence data to provide a respective measure of the similarity between the recognised sequence of labels and said plurality of stored sequences of labels wherein said comparing step comprises a normalising step for normalising each of said similarity measures wherein said aligning step comprises a dynamic programming step for aligning said recognised and stored sequences of labels using a dynamic programming technique and wherein said normalising step calculates the respective normalisation scores during the progressive calculation of said possible alignments by said dynamic programming step wherein said normalising step calculates, for each possible aligned pair of labels where P(ai|pr) represents the probability of confusing set label pr as stored sequence label ai and P(pr) represents the probability of set label Pr occurring in a sequence of labels. - View Dependent Claims (44)
-
-
45. A method of searching a database comprising a plurality of information entries to identify information to be retrieved therefrom, each of said plurality of information entries having an associated annotation comprising a sequence of labels, the method comprising:
-
a label comparison method for comparing a query sequence of labels obtained from an input query with the labels of each annotation to provide a set of comparison results, wherein the label comparison method comprises the steps of receiving an input signal, a recognition processing step of comparing said input signal with stored label models to generate a recognised sequence of labels in said input signal and confidence data representative of the confidence that the recognised sequence of labels is representative of said input signal, and calculating a measure of the similarity between the recognised sequence of labels generated in said recognition processing step and a stored sequence of annotation labels by comparing said recognised sequence of labels with the stored sequence of annotation labels using a combination of i) predetermined confusion data which defines confusability between different labels, and ii) said confidence data generated in said recognition processing step and representative of the confidence that the recognised sequence of labels is representative of said input signal; and a step of identifying said information to be retrieved from said database using said comparison results, wherein said label comparison method has a plurality of different comparison modes of operation and in that the method further comprises the steps of determining if the sequence of labels of a current annotation was generated from an audio signal or from text, and outputting a determination result; and selecting, for the current annotation, the mode of operation of said label comparison method in dependence upon said determination result.
-
-
46. A method of searching a database comprising a plurality of information entries to identify information to be retrieved therefrom, each of said plurality of information entries having an associated annotation comprising a sequence of labels, the method comprising:
-
a label comparison method for comparing a query sequence of labels obtained from an input query with the labels of each annotation to provide a set of comparison results, wherein the label comparison method comprises the steps of receiving an input signal, a recognition processing step of comparing said input signal with stored label models to generate a recognised sequence of labels in said input signal and confidence data representative of the confidence that the recognised sequence of labels is representative of said input signal, and calculating a measure of the similarity between the recognised sequence of labels generated in said recognition processing step and a stored sequence of annotation labels by comparing said recognised sequence of labels with the stored sequence of annotation labels using a combination of i) predetermined confusion data which defines confusability between different labels, and ii) said confidence data generated in said recognition processing step and representative of the confidence that the recognised sequence of labels is representative of said input signal; and a step of identifying said information to be retrieved from said database using said comparison results, wherein one or more of said information entries is the associated annotation.
-
Specification