Back to results
US 10,540,990 B2Granted
Processing of speech signals
AssigneeInternational Business Machines Corporation
InventorsMasayuki Suzuki, Takashi Fukuda, Toru Nagano
Life of this patent
Estimating the term…
Abstract
A method for processing a speech signal. The method comprises obtaining a logmel feature of a speech signal. The method further includes one or more processors processing the logmel feature so that the logmel feature is normalized under a constraint that a power level of the logmel feature is kept as originally obtained. The method further includes inputting the processed logmel feature into a speech-to-text system to generate corresponding text data.
Claims17
A method, the method comprising:
- obtaining a logmel feature of a speech signal, wherein the speech signal includes utterances by a first speaker and utterances by a second speaker, wherein a power level of the logmel feature corresponds to a power level of the speech signal, and wherein the power level of the logmel feature changes over frames of the logmel feature based on different distances at which the first speaker and the second speaker were positioned from a microphone recording the speech signal;
- one or more processors converting the logmel feature into lossless Mel Frequency Cepstral Coefficients (MFCCs), including a coefficient corresponding to the power level of the logmel feature;
- one or more processors processing the lossless MFCCs so that statistics are normalized for each coefficient of the lossless MFCCs except for the coefficient corresponding to the power level of the logmel feature, resulting in partially normalized lossless MFCCs that include:
- (i) the coefficients with the normalized statistics, and (ii) the non-normalized coefficient corresponding to the power level of the logmel feature;
- one or more processors converting the partially normalized lossless MFCCs into a partially normalized logmel feature, wherein a power level of the partially normalized logmel feature changes over different frames of the partially normalized logmel feature based on the different distances at which the first speaker and the second speaker were positioned from the microphone recording the speech signal;
- and
- inputting the partially normalized logmel feature into a speech-to-text system to generate corresponding text data.
4 more elements
A system, the system comprising:
- one or more processors;
- one or more computer readable storage media;
- and
- program instructions stored on the computer readable storage media for execution by at least one of the one or more processors, the program instructions comprising:
- program instructions to obtain a logmel feature of a speech signal, wherein the speech signal includes utterances by a first speaker and utterances by a second speaker, wherein a power level of the logmel feature corresponds to a power level of the speech signal, and wherein the power level of the logmel feature changes over frames of the logmel feature based on different distances at which the first speaker and the second speaker were positioned from a microphone recording the speech signal;
- program instructions to convert the logmel feature into lossless Mel Frequency Cepstral Coefficients (MFCCs), including a coefficient corresponding to the power level of the logmel feature;
- program instructions to process the lossless MFCCs so that statistics are normalized for each coefficient of the lossless MFCCs except for the coefficient corresponding to the power level of the logmel feature, resulting in partially normalized lossless MFCCs that include:
- (i) the coefficients with the normalized statistics, and (ii) the non-normalized coefficient corresponding to the power level of the logmel feature;
- program instructions to convert the partially normalized lossless MFCCs into a partially normalized logmel feature, wherein a power level of the partially normalized logmel feature changes over different frames of the partially normalized logmel feature based on the different distances at which the first speaker and the second speaker were positioned from the microphone recording the speech signal;
- and
- program instructions to input the partially normalized logmel feature into a speech-to-text system to generate corresponding text data.
8 more elements
A computer program product, the computer program product comprising:
- one or more computer readable storage media and program instructions stored on the one or more computer readable storage media, the program instructions comprising:
- program instructions to obtain a logmel feature of a speech signal, wherein the speech signal includes utterances by a first speaker and utterances by a second speaker, wherein a power level of the logmel feature corresponds to a power level of the speech signal, and wherein the power level of the logmel feature changes over frames of the logmel feature based on different distances at which the first speaker and the second speaker were positioned from a microphone recording the speech signal;
- program instructions to convert the logmel feature into lossless Mel Frequency Cepstral Coefficients (MFCCs), including a coefficient corresponding to the power level of the logmel feature;
- program instructions to process the lossless MFCCs so that statistics are normalized for each coefficient of the lossless MFCCs except for the coefficient corresponding to the power level of the logmel feature, resulting in partially normalized lossless MFCCs that include:
- (i) the coefficients with the normalized statistics, and (ii) the non-normalized coefficient corresponding to the power level of the logmel feature;
- program instructions to convert the partially normalized lossless MFCCs into a partially normalized logmel feature, wherein a power level of the partially normalized logmel feature changes over different frames of the partially normalized logmel feature based on the different distances at which the first speaker and the second speaker were positioned from the microphone recording the speech signal;
- and
- program instructions to input the partially normalized logmel feature into a speech-to-text system to generate corresponding text data.
5 more elements
Dependencies are inferred from the claim text.