Evaluating Acoustic, Textual and Grammar Features for Alcohol Classification

Abstract:

This paper evaluates the detection of alcohol intoxication in speech, using different classification approaches. It compares the classification based on acoustic, textual and grammar features as well as the combination of these three. Furthermore, it examines the influence of experiment and category dependency. The paper concludes that neither classification using textual features nor classifica- tion based on grammar features can challenge the baseline. Nevertheless, experi- ment and category dependency have a significant influence on the result.


Year: 2016
In session: Sprechercharakteristik und Stimmanalyse
Pages: 214 to 220