Syllables, segments and speech tempo perception
Leendert Plug (University of Leeds)
Speech tempo is a fundamental parameter in speech analysis. Tempo fluctuations are of interest to researchers studying speech production, speech perception and verbal interaction. Inter-speaker variation in habitual tempo is relevant for forensic speaker profiling, and speech recognition and synthesis systems require tempo models to process input variation and produce natural-sounding output. Speech tempo also informs fluency measures and language learner assessments, and various speech, language and other disorders are associated with atypical tempo patterns. The research I will report on in this talk is motivated by our limited grasp on this fundamental parameter -- in particular, our limited understanding of how speech tempo is perceived by ordinary listeners, and how commonly used methods of measuring tempo reflect this perception. I will present results of initial experiments probing the impact of phonological complexity on speech tempo perception -- complexity which results in tempo measurements in syllables per second diverging from measurements in segments per second. I will describe ongoing experiments into the relationship between rhythm and tempo perception, and outline plans for investigation the extent to which segment deletions are taken into account in tempo estimation: in essence, when listeners hear [sport] meaning 'support', do they estimate its tempo on the observation that the speaker is articulating a one-syllable form, or on the understanding that the speaker is communicating a two-syllable one?