There are many techniques for modelling properties of speech duration patterns, including models of rhythm as oscillation, partial models of rhythm types as departures from isochrony, models of tempo acceleration and deceleration, and models of duration hierarchies and their relation to hierarchies in word and phrase structure. Except for oscillator modelling, many approaches use data extraction from speech annotations, often with mainly manual methods. We employ computational data-mining for phonetic research, as opposed to phonological research on the one hand or speech technological research on the other, and explore the potential of the computational annotation data-mining paradigm for improving efficiency and scope of analysis. We show consistent variation in syllable duration patterns in selected speech varieties in English, Chinese and Polish, chosen for their known different prosodic typological properties. Results include a possible limen of 50ms for relevant timing patterns. For data-mining we use the Time Group Analysis (TGA) methodology, directly in the TGA online tool and integrated into the Annotation Pro+TGA desktop software.