Listeners are able to tell apart read-aloud and spontaneously produced speech. Prosody appears to be important for this perceptual distinction. In this paper, the importance of the distribution and realization of prosodic boundaries is investigated. Recordings were made of five male speakers, spontaneously producing so-called instruction monologues. Transcripts of these monologues were read aloud by the same speakers. A perception experiment was carried out to obtain classification scores for isolated utterances selected from the spontaneous and read material. Auditory prosodic transcriptions were made of the entire spontaneous and read monologues, assessing the distribution and realization of underlying prosodic boundaries in both speech types. The underlying prosodic structure was assessed by means of an automatic text-to-speech system. Observed differences in the production of prosodic boundaries in the spontaneous and read material are related to the perceptual classification scores by means of a multiple regression analysis. Results show a significant correlation, suggesting that differences in the distribution and realization of prosodic boundaries contribute significantly to the perceptual difference between spontaneous and read speech.
