SLaTE 2015 - Workshop on Speech and Language Technology in Education
Within the framework of computer assisted foreign language learning for the French/German pair, we evaluate different HMM phone models for detecting accurate phone boundaries. The optimal parameters are determined by minimizing on the non-native speech corpus the number of phones whose boundaries are shifted by more than 20 ms compared to the manual boundaries. We observe that the best performance was obtained by combining a French native HMM model with an automatically selected German native HMM model.
Bibliographic reference. Fohr, Dominique / Mella, Odile (2015): "Detection of phone boundaries for non-native speech using French-German models", In SLaTE-2015, 181-182.