Unsupervised Classification of Speaker Profiles as a Point Anomaly Detection Task

[edit]

Cedric Fayet, Arnaud Delhay, Damien Lolive, Pierre-François Marteau ;
Proceedings of the First International Workshop on Learning with Imbalanced Domains: Theory and Applications, PMLR 74:152-163, 2017.

Abstract

This paper presents an evaluation of three different anomaly detector methods over different feature sets. The three anomaly detectors are based respectively on Gaussian Mixture Model (GMM), One-Class SVM and isolation Forest. The considered feature sets are built from personality evaluation and audio signal. Personality evaluations are extracted from the BFI-10 Questionnaire, which allows to manually evaluate five personality traits (Openness, Conscientiousness, Extroversion, Agreeableness, Neuroticism). From the audio signal, we automatically extract a prosodic feature set, which performs well in affective computing. The different combinations of models and feature sets are evaluated on the SSPNET-Personality corpus which has already been used in several experiments, including a previous work on separating two types of personality profiles in a supervised way. In this work, we propose an evaluation of the three anomaly detectors with consideration to the features used. Results show that, regardless of the feature set, GMM based method is the most efficient one (it obtains 0.96 ROC-AUC score with the best feature set). The prosodic feature set seems to be a good compromise between performance (0.91 ROC-AUC score with GMM based method) and ease of extraction.

Related Material