System and method for improving robustness of speech recognition using vocal tract length normalization codebooks
View Patent ↗Disclosed are systems, methods, and computer readable media for performing speech recognition. The method embodiment comprises selecting a codebook from a plurality of codebooks with a minimal acoustic distance to a received speech sample, the plurality of codebooks generated by a process of (a) computing a vocal tract length for a each of a plurality of speakers, (b) for each of the plurality of speakers, clustering speech vectors, and (c) creating a codebook for each speaker, the codebook containing entries for the respective speaker's vocal tract length, speech vectors, and an optional vector weight for each speech vector, (2) applying the respective vocal tract length associated with the selected codebook to normalize the received speech sample for use in speech recognition, and (3) recognizing the received speech sample based on the respective vocal tract length associated with the selected codebook.
1. A method of performing speech recognition, the method comprising:
(1) selecting a codebook from a plurality of codebooks with a minimal acoustic distance to a received speech sample, the plurality of codebooks generated by a process of:
(a) computing a vocal tract length for a each of a plurality of speakers;
(b) for each of the plurality of speakers, clustering speech vectors; and
(c) creating a codebook for each speaker, the codebook containing entries for the respective speaker's vocal tract length, speech vectors, and an optional vector weight for each speech vector;
(2) applying the respective vocal tract length associated with the selected codebook to normalize the received speech sample for use in speech recognition; and
(3) recognizing the received speech sample based on the respective vocal tract length associated with the selected codebook.
2. The method of claim 1 , further comprising during run-time the steps of:
selecting an additional codebook with an additional minimal acoustic distance to the received speech sample for each additional speech vector;
applying the respective vocal tract length associated with the additional codebook to normalize the received speech sample for each additional speech vector for use in speech recognition; and
recognizing the received speech sample based on the respective additional vocal tract length associated with the additional codebook.
3. The method of claim 1 , wherein the method is performed frame by frame.
4. A system for performing speech recognition, the system comprising:
(1) a module configured to select a codebook from a plurality of codebooks with a minimal acoustic distance to a received speech sample, the plurality of codebooks generated by a process of:
(a) computing a vocal tract length for a each of a plurality of speakers;
(b) for each of the plurality of speakers, clustering speech vectors; and
(c) creating a codebook for each speaker, the codebook containing entries for the respective speaker's vocal tract length, speech vectors, and an optional vector weight for each speech vector;
(2) a module configured to apply the respective vocal tract length associated with the selected codebook to normalize the received speech sample for use in speech recognition; and
(3) a module configured to recognize the received speech sample based on the respective vocal tract length associated with the selected codebook.
5. The system of claim 4 , the system further comprising:
a module configured to select an additional codebook with an additional minimal acoustic distance to the received speech sample for each additional speech vector;
a module configured to apply the respective vocal tract length associated with the additional codebook to normalize the received speech sample for each additional speech vector for use in speech recognition; and
a module configured to recognize the received speech sample based on the respective additional vocal tract length associated with the additional codebook.
6. The system of claim 4 , wherein the system performs frame by frame.
7. A non-transitory computer readable medium storing a computer program having instructions for performing speech recognition, the instructions comprising:
(1) selecting a codebook from a plurality of codebooks with a minimal acoustic distance to a received speech sample, the plurality of codebooks generated by a process of:
(a) computing a vocal tract length for a each of a plurality of speakers;
(b) for each of the plurality of speakers, clustering speech vectors; and
(c) creating a codebook for each speaker, the codebook containing entries for the respective speaker's vocal tract length, speech vectors, and an optional vector weight for each speech vector;
(2) applying the respective vocal tract length associated with the selected codebook to normalize the received speech sample for use in speech recognition; and
(3) recognizing the received speech sample based on the respective vocal tract length associated with the selected codebook.
8. The non-transitory computer readable medium of claim 7 , the instructions further comprising during run-time the steps of:
selecting an additional codebook with an additional minimal acoustic distance to the received speech sample for each additional speech vector;
applying the respective vocal tract length associated with the additional codebook to normalize the received speech sample for each additional speech vector for use in speech recognition; and
recognizing the received speech sample based on the respective additional vocal tract length associated with the additional codebook.
9. The non-transitory computer readable medium of claim 7 , wherein the instructions are performed frame by frame.