Speech recognition using discriminant features
View Patent ↗Methods and arrangements for representing the speech waveform in terms of a set of abstract, linguistic distinctions in order to derive a set of discriminative features for use in a speech recognizer. By combining the distinctive feature representation with an original waveform representation, it is possible to achieve a reduction in word error rate of 33% on an automatic speech recognition task.
1. A method of facilitating speech recognition, said method comprising the steps of:
obtaining speech input data;
building a model for each feature of an original set of linguistic features, wherein the model reflects whether or not each feature is present;
ranking the linguistic features;
rebuilding the model for each of a preselected number N of the ranked linguistic features; and
compiling a confusion matrix for each feature of the original set of features subsequent to said step of building a model for each feature of an original set of features, wherein said compiling a confusion matrix comprises:
computing a score for each feature based on the likelihood of its presence in a frame of the speech input data, and
calculating mutual information between truth and labels for each feature;
wherein the ranking comprises ranking the mutual information calculated in compiling the confusion matrix.
2. The method according to claim 1 , wherein said step of building a model for each of a preselected number N of the ranked features comprises building a model for the top N ranked features.
3. The method according to claim 1 , wherein said step of computing a score for each feature comprises computing a score as a log-likelihood ratio.
4. The method according to claim 1 , wherein said step of compiling a confusion matrix further comprises comparing each score of each feature with a threshold.
5. The method according to claim 1 , wherein said step of building a model for each feature of an original set of features comprises:
partitioning the speech input data in parallel, once for ezich feature; and
producing an observation vector.
6. The method according to claim 5 , wherein said step of building a model for each feature of an original set of features comprises:
partitioning data in parallel from the observation vector, once for each feature; and
producing final observations.
7. The method according to claim 1 , wherein said step of building a model for each of a preselected number N of the ranked features comprises:
partitioning the speech input data in parallel, once for each feature; and
producing an observation vector.
8. The method according to claim 7 , wherein said step of building a model for each of a preselected number N of the ranked features comprises:
partitioning data in parallel from the observation vector, once for each feature; and
producing final observations.
9. An apparatus for facilitating speech recognition, said method tomprising the steps of:
an input medium which obtains speech input data;
a first model builder which builds a model for each feature of an orininal set of linguistic features, wherein the model reflects whether or not each feature is present;
a ranking arrangement which ranks the linguistic features;
a second model builder which rebuilds the model for each of a preselected number N of the ranked linguistic features; and
a matrix compiler which compiles a confusion matrix for each feature of the original set of features subsequent to said step of building a model for each feature of an original set of features, wherein said matrix compiler is adapted to:
compute a score for each feature based on the likelihood of its presence in a frame of the speech input data, and
calculate mutual information between truth and labels for each feature;
wherein said ranking arrangement is adapted to rank the mutual information calculated in compiling the confusion matrix.
10. The apparatus according to claim 9 , wherein said second model builder is adapted to build a model for the top N ranked features.
11. The apparatus according to claim 9 , wherein said matrix compiler is adapted to compute a score as a log-likelihood ratio.
12. The apparatus according to claim 9 , wherein said matrix compiler is adapted to compare each score of each feature with a threshold.
13. The apparatus according to claim 9 , wherein said first model builder is adapted to:
partition the speech input data in parallel, once for each feature; and
produce an observation vector.
14. The apparatus according to claim 13 , wherein said first model builder is adapted to:
partition data in parallel from the observation vector, once for each feature; and
produce final observations.
15. The apparatus according to claim 9 , wherein said second model builder is adapted to:
partition the speech input data in parallel, once for each feature; and
produce an observation vector.
16. The apparatus according to claim 15 , wherein said second model builder is adapted to:
partition data in parallel from the observation vector, once for each feature; and
produce final observations.
17. A program storage device readable by computer, tangibly embodying a program of instructions executable by the computer to perform method steps for speech recognition, said method comprising the steps of:
obtaining speech input data;
building a model for each feature of an original set of linguistic features, wherein the model reflects whether or not each feature is present;
ranking the linguistic features;
rebuilding the model for each of a preselected number N of the ranked linguistic features; and
compiling a confusion matrix for each feature of the original set of features subsequent to said step of building a model for each feature of an original set of features, wherein said compiling a confusion matrix comprises:
computing a score for each feature based on the likelihood of its presence in a frame of the speech input data, and
calculating mutual information between truth and labels for each feature;
wherein the ranking comprises ranking the mutual information calculated in compiling the confusion matrix.