Pitch and Intonation Contribution to Speakers' Traits Classification

摘要

The article describes the system we submitted for the three sub-challenges of INTERSPEECH 2012 Speaker Trait Challenge for the classification of the five personality traits of OCEAN, likability and intelligibility. The system was based on a two-class SVM-classifier using leave-one-speaker-out cross,validation to optimize SVM complexity parameter and to select the feature set for trait classification. Pitch, intonation and spectrum contributions to speakers' traits classification were studied. One feature set was specially designed for intonation modeling. Variations from the official feature set based on pitch and spectrum were assessed in traits classification. Preliminary results on the Test set have shown a significant improvement of 4.7% on the likability trait, an improvement of 1.1% on the intelligibility trait and no improvement on the OCEAN traits compared to the official baseline unweighted accuracy results.