In the act of music listening, many people break down musical pieces into chunks such as verses and choruses. Recent work on music segmentation has shown that highly agreed segment boundaries are also considered strong and are described by using multiple cues. However, these studies could not pinpoint the effects of data collection methods and of musicianship on boundary perception. Our study investigated the differences between segmentation tasks performed by musicians in real-time and non real-time listening contexts. Further, we assessed the effect of musical training on the perception of boundaries in real-time listening. We collected perceived boundaries by 18 musicians and 18 non-musicians in 9 musical examples. Musicians also completed a non real-time segmentation task for 6 of the examples. We observed high significant correlations between participant groups and between task groups at a time-scale of 10 seconds after comparing segmentation data at different resolutions. Further, musicians located significantly more boundaries in the non real-time task than in the real-time task for 5 out of 6 examples. We found a clear effect of the task but no effects of musical training upon perceived segmentation.