Interpersonal coordination is assumed to be a complex phenomenon due to the large number of factors involved and the non-trivial way in which they interact. Temporal collective patterns, including coordination, may emerge during natural conversations in ways that cannot be predicted from the behavior of the participants alone. The characterization of these complex patterns is still challenging but is necessary in order to understand the emergence of coordination at the collective level. We aim to characterize the structural complexity and long-term dynamics of the acoustic activity at the group level. To do so, we performed different analyses: Burstiness, Allan factor analysis, Recurrence quantification analysis, and BPM estimation using a sliding window technique on the time series of a corpus of small group conversations and their randomized and isochronous versions. Together, these analyses showed that the collective acoustic rhythm has a structural complexity that reorganizes over time and that, despite their intrinsic differences, conversations can exhibit two distinct levels of complexity. By uncovering these regularities, our results complement previous knowledge on the complexity of interpersonal coordination on short timescales. Further studies in this direction should evaluate the validity of this approach and findings in different experimental settings