Understanding Children’s Speech Finding the Underlying Representation of Speech
Keywords
No Thumbnail Available
Authors
Issue Date
2020-03-04
Language
en
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The use of Speech Recognition increases in our daily lives. Both in Games and Education, Speech Recognition is used, as well as in the medical world. However, Speech Recognition systems for children perform worse than those for adults. In this thesis a new method is tested, where an underlying representation of speech is used to match the child’s speech to a target speech signal. To test this method, the Cosine Similarity, RMSE, DTW and two combination of machine learning systems and DTW, are used to match the LPCs of the Child Speech Signals and Target Speech Signals. The results show us that the accuracy of this method is below the 1% and also takes a long time. It is concluded that this method is not sufficient to be used for Speech recognitions. Besides, this method is too slow, it can’t be used in live interaction.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen