University of Twente Student Theses
Prominence detection in spoken Dutch using prosodic features and machine learning
Kok, M. (2023) Prominence detection in spoken Dutch using prosodic features and machine learning.
|
PDF
299kB |
| Abstract: | Research into various aspects of speech is increasingly making use of more advanced machine learning. These advancements are beneficial to the development of better, more realistic conversational agents. If the use of prosodic features can be extended further, agents could use these to understand implicit contextual clues in human speech. We use several types of machine learning models to test and compare how well word prominence can be classified. The machine learning models tested are a support vector machine, a random forest classifier, and a multi-layer perceptron. The openSMILE Python library is used to extract prosodic features as described by the GeMAPS feature set. We also use two data preprocessing methods, standardization and feature selection, and tested how these affect the results. The support vector machine with both data preprocessing methods performed best with an F_1-score of 0.698. Conversely, the multi-layer perceptron performed the worst with F_1-scores ranging from 0.542 on unprocessed data to 0.692 on standardized data. |
| Item Type: | Essay (Bachelor) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 54 computer science |
| Programme: | Computer Science BSc (56964) |
| Link to this item: | https://purl.utwente.nl/essays/94364 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page