Speaker Classification by Means of Orthographic and Broad Phonetic Transcriptions of Speech

摘要

In this study we investigate whether a classification algorithm originally designed for authorship verification can be used to classify speakers according to their gender, age, regional background and level of education by investigating the lexical content and the pronunciation of their speech. Contrary to other speaker classification techniques, our algorithm does not base its decisions on direct measurements of the speech signal; rather it learns characteristic speech features of speaker classes by analysing the orthographic and broad phonetic transcription of speech from members of these classes. The resulting class profiles are subsequently used to verify whether unknown speakers belong to these classes.