1 / 13

Evaluation of English Intonation based on Combination of Multiple Evaluation Scores

Evaluation of English Intonation based on Combination of Multiple Evaluation Scores. Akinori Ito, Tomoaki Konno, Masashi Ito and Shozo Makino. 報告者:郝柏翰. Outline. Introduction Evaluation of intonation based on word importance factor Combination of score using multiple regression

lassie
Download Presentation

Evaluation of English Intonation based on Combination of Multiple Evaluation Scores

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Evaluation of English Intonation based on Combination of MultipleEvaluation Scores Akinori Ito, Tomoaki Konno, Masashi Ito and Shozo Makino 報告者:郝柏翰

  2. Outline • Introduction • Evaluation of intonation based on word importance factor • Combination of score using multiple regression • Experiments • Conclusion

  3. Evaluation of intonation based on word importance factor (1/6) • Overview • This system first extracts features for intonation evaluation from English utterance by Japanese learner. • The extracted features are then Compared with those from teacher utterances word by word • the system calculate intonation scores for all teacher utterances. Then the best score among the all scores is chosen as the final score • Feature extraction • Using the four dimensional feature vectors • F0, log power and their first derivatives

  4. Evaluation of intonation based on word importance factor (2/6) • Calculation of intonation score • Letbe thedistance between the learner’s utterance. • Then the intonationscore of the -th word is calculated as follows. • Then the intonation score of the entire utterance is calculated by averaging for all words. • Finally, the best score among all t is chosen as the final score of the learner’s utterance.

  5. Evaluation of intonation based on word importance factor (3/6) • Introduction of word importance factor • native speakers appear to evaluate a learner’s prosody by focusing on several keywords. • Let be a word importance factor of the -th word of the -th sample uttered by a learner. • This factor is estimated by the least squares method. • Then the error Q is defined as follows

  6. Evaluation of intonation based on word importance factor (4/6) • On determining , it is difficult to determine individual for each of distinct words in the sentences because the number of words in the training sentences are not sufficient. • a decision tree is generated using the questions so that all words in a cluster share a word importance factor and the estimation error Q become minimum by clustering using the decision tree.

  7. Evaluation of intonation based on word importance factor (5/6) • Let be a mapping from the -th word in the -thutterance to the cluster to which the word belongs, and be the coefficients that minimize the error Q. Then the intonation score of an utterance is calculated as follows • where K is the number of words in a sentence to be evaluated and is a mapping from the -th word to a cluster determined by the decision tree.

  8. Evaluation of intonation based on word importance factor (6/6) • Combining rhythm feature • In addition to the features for intonation, features for rhythm evaluation are also introduced into the intonation evaluation. • Word duration ratio and DP distance are used as features of rhythm evaluation. Using a rhythm score calculated from the rhythm features. • the intonation score is calculated as follows:

  9. Combination of score using multiple regression (1/2) • Problems of the conventional framework • only one teacher is considered for evaluating one utterance. • concerned with combination of the intonation and rhythm features.

  10. Combination of score using multiple regression (2/2) • Let be an intonation score of the learner’s -th utterance calculated using the teacher utterance . • Minimum: = • Maximum: = • Median: • Regression:

  11. Experiments (1/2) • Experimental conditions • Effect of score combination

  12. Experiments (2/2) • Effect of number of classes • we changed the maximum number of word classes for training of decision trees. • Figure 4 shows the correlation coefficients with respect to the maximum number of word classes. • The optimum number of classes was around 20.

  13. Conclusion • Our method is based on the evaluation method by Suzuki et al. that uses a decision tree for estimating word importance factor. • We extended Suzuki’s method so that multiple regression trees were trained and the scores were combined using the multiple regression. • One of remaining problems is how to feedback the evaluation results to the learner. • More research is needed to design the best way of showing a learner these kinds of information.

More Related