Simulating the Pronunciation of a Second Language in WaveNet

Keywords
No Thumbnail Available
Issue Date
2017-07-19
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
In 2016 WaveNet, a generative raw audio DNN model was introduced and showed unprecedented results in the area of speech synthesis. In this study the learning of the pronunciation of a second language is simulated. Using two conditions, bilingual from the onset of training, and monolingual from onset followed by bilingualism later on in training, a human being acquiring a second language in two different situations is simulated. Using formant analysis on the output of the network, data was acquired, and compared to average formants of both English and Dutch males and females. Against our expectations, the model generated output closer to the Dutch average, when it was trained on both languages from onset.
Description
Citation
Faculty
Faculteit der Sociale Wetenschappen