Second Language Studies: Acquisition, Learning, Education and Technology
We introduce a set of speaker dependent features derived from the positions of vowels in Mel-Frequency Cepstral Coefficient (MFCC) space relative to a reference vowel. The MFCCs for a particular speaker are transformed using simple operations into features that can be used to classify vowels from a common reference point. Classification performance of vowels using Gaussian Mixture Models (GMMs) is significantly improved, regardless of which vowel is used as the target among /A/, /i/, /u/, or /´/. We discuss how this technique can be applied to assess pronunciation with respect to vowel structure rather than agreement with absolute position in MFCC space.
Bibliographic reference. Peabody, Mitchell / Seneff, Stephanie (2010): "A simple feature normalization scheme for non-native vowel assessment", In L2WS-2010, paper O2-2.