This project aims at testing different methods from the literature for formant normalization, given specific classification tasks for conversational speech. Formants are the main acoustic correlates relevant to distinguishing vowel sounds. Formant values are largely dependent on the pitch of the speakers, which in general is highest for children, followed by female speakers and in general lowest for men, as pitch hight correlates with the biological vocal tract length. In the linguistic literature, several approaches have been tested with the purpose of sociolinguistic vowel space analyses. The aim of this project now is to test them for classification approaches, and to evaluate them in terms of both performance and robustness.

TEAMS are WELCOME!

Your Profile

• interest in speech phenomena and speech signal processing
• good experience in Python
• background in machine learning is appreciated

Your tasks

• literature research, esp. on suitable algorithms
• data preprocessing
• implementation of feature extraction algorithms
• evaluation of classification results

Groups are welcome!

Contact:

Barbara Schuppler (b.schuppler@tugraz.at)