Speech elicits brain activity that is time-locked to its amplitude envelope, a signal dominated by modulations at the syllable rate. The resulting Cerebro-acoustic coherence (CACoh) is thought to be an essential feature of a speech parsing mechanism crucial to comprehension and has been examined especially in the context of degraded or noisy speech, which showed evidence for a functional meaning of CACoh. The functional role of CACoh – and its variation - in the context of noise-free speech is underexplored and not well understood. Here, we set out to examine magnetoencephalography (MEG) data of participants (n=53) while listening to clear speech stories. We examined CACoh using Gaussian copula mutual information to establish the extent of tracking at time constants consistent with the phrasal, lexical and syllable rates of the stimuli. At the syllable rate (3-7Hz), bilateral superior temporal gyrus significantly tracked speech information. Notably, individual differences in tracking were positively linked to recognition accuracy in a completely independent words-in-noise (WiN) task. Intriguingly, left STG CACoh was also positively associated with higher fluid intelligence. Further examining potential sources of task-related top-down communication to left and right STG , using Granger causality, we found that a specific pattern of top-down connectivity from frontal areas was indeed associated to WiN performance. Subsequent mediation analysis confirmed that the relationship of CACoh and WiN was at least partially mediated by this top-down connectivity pattern, while in contrast, a mediation of CACoh by WiN through bottom-up connectivity was absent. Thus, results indicate that individual variability of speech tracking in the auditory system - even during clear speech - reflects the potential robustness of the speech system to noise, and that top-down influences from brain areas associated with higher-order linguistic processing and articulatory processes contribute to speech-brain entrainment.