ISCA Archive SpeechProsody 2014
ISCA Archive SpeechProsody 2014

Processing Prosodic Boundaries in Natural and Filtered Speech

Grace Kuo

The prosody of an utterance can carry information that is critically important to understand the meaning of a sentence. In addition, previous studies have shown that listeners are able to detect major prosodic boundaries in their native language in stimuli whose segmental information has been removed, such as low-pass filtered [1][2] and hummed speech [2][3][5]. The present boundary strength rating study is conducted on native and non-native speakers to Taiwanese and Swedish, in an attempt to observe native and non-native speakers’ accuracy in judging the upcoming boundary size in natural and filtered speech. 36 Taiwanese and American English speakers were recruited for the rating task whose stimuli consisted of Taiwanese and Swedish utterances from three prosodic boundary types (word boundary, phrase/tone sandhi group boundary, and Intonation Phrase boundary). In Experiment 1, participants rated the upcoming boundary strength on a slider for filtered speech stimuli. In Experiment 2, they rated the boundary strength for natural speech stimuli. The results show that both non-native speakers could accurately predict the upcoming prosodic boundary type in both natural and filtered speech. The acoustic analyses of duration, f0 range, f0 median, spectral tilt, and harmonics-to-noise ratio reveal that non-native speakers use these prosodic cues to make their judgment, however, they put different emphasis on different cues when they were presented with stimuli of different qualities (natural vs. filtered) and lengths.