System and method for text-to-speech performance evaluation
AssigneeBayerische Motoren Werke Aktiengesellschaft
InventorsLu Chen, Carsten Isert, Sam Lai, Wenhui Lei
Life of this patent
Estimating the term…
Abstract
A system and method for text-to-speech performance evaluation are provided. The method (100) for text-to-speech performance evaluation includes providing a plurality of speech samples and scores associated with the respective speech samples (110); extracting acoustic features that influence the associated scores of the respective speech samples from the respective speech samples (120); training a machine learning model by the acoustic features and corresponding scores (130); and evaluating a text-to-speech engine by the trained machine learning model (140).
Claims17
A method for text-to-speech performance evaluation, comprising:
- providing a plurality of speech samples and scores associated with respective speech samples;
- extracting acoustic features that influence the associated scores of the respective speech samples from the respective speech samples;
- training a machine learning model by the acoustic features and corresponding scores;
- and
- evaluating a text-to-speech engine by the trained machine learning model;
- wherein the machine learning model is a pairwise support vector machine (SVM) model, and training the machine learning model by the acoustic features and corresponding scores comprises:
- combining a vector representing the acoustic features of a first speech sample and a vector representing the acoustic features of a second speech sample, wherein the first speech sample and the second speech sample are samples with the same textual content but from different speech sources;
- and
- training the SVM model with the combined vector and a label representing the relative size of the associated score of the first speech sample and the second speech sample.
A system for text-to-speech performance evaluation, comprising:
- a sample store containing a plurality of speech samples and scores associated with respective speech samples;
- an acoustic feature extracting section configured to extract acoustic features that influence the associated scores of the respective speech samples from the respective speech samples;
- a model training section configured to train a machine learning model by the acoustic features and corresponding scores;
- and
- an evaluating section configured to evaluate a text-to-speech engine by the trained machine learning model;
- wherein the machine learning model is a pairwise support vector machine (SVM) model, and to train the machine learning model by the acoustic features and corresponding scores, the model training section is configured to:
- combine a vector representing the acoustic features of a first speech sample and a vector representing the acoustic features of a second speech sample, wherein the first speech sample and the second speech sample are samples with the same textual content but from different speech sources;
- and
- train the SVM model with the combined vector and a label representing the relative size of the associated score of the first speech sample and the second speech sample.
A vehicle equipped with a system for text-to-speech performance evaluation, the system comprising:
- a sample store containing a plurality of speech samples and scores associated with respective speech samples;
- an acoustic feature extracting section configured to extract acoustic features that influence the associated scores of the respective speech samples from the respective speech samples;
- a model training section configured to train a machine learning model by the acoustic features and corresponding scores;
- and
- an evaluating section configured to evaluate a text-to-speech engine by the trained machine learning model;
- wherein the machine learning model is a pairwise support vector machine (SVM) model, and to train the machine learning model by the acoustic features and corresponding scores, the model training section is configured to:
- combine a vector representing the acoustic features of a first speech sample and a vector representing the acoustic features of a second speech sample, wherein the first speech sample and the second speech sample are samples with the same textual content but from different speech sources;
- and
- train the SVM model with the combined vector and a label representing the relative size of the associated score of the first speech sample and the second speech sample.
Dependencies are inferred from the claim text.