15th Annual Conference of the International Speech Communication Association

September 14-18, 2014

Foreign Accent Recognition Based on Temporal Information Contained in Lowpass-Filtered Speech

Marie-José Kolly, Adrian Leemann, Volker Dellwo

Universität Zürich, Switzerland

Can the foreign accent of a speaker be recognized based on suprasegmental temporal information? For a perception experiment we created stimuli based on German sentences read by six French and six English speakers. These foreign-accented sentences were manipulated by (1) applying a lowpass filter with a cutoff frequency of 300 Hz and (2) applying the same lowpass filter and monotonizing F0. In a between-subject 2AFC perception experiment we tested the accent recognition ability of 15 Swiss German listeners per signal manipulation condition. The results showed that speakers' native language could be recognized above chance in both conditions. However, listeners obtained significantly lower recognition scores in the monotonized condition. Furthermore, higher recognition scores were obtained for French-accented speech in the monotonized condition, a result that is discussed in light of research on speech rhythm. We further report an effect for speaker within each accent group. The results suggest that suprasegmental temporal information allows for foreign accent recognition to some degree.

Full Paper

Bibliographic reference.  Kolly, Marie-José / Leemann, Adrian / Dellwo, Volker (2014): "Foreign accent recognition based on temporal information contained in lowpass-filtered speech", In INTERSPEECH-2014, 2175-2179.