Auditory selective attention plays a central role in the human capacity to reliably process complex sounds in multi-source environments. The ability to track the attentional state of individuals in such environments is of interest to neuroscientists and engineers due to its importance in the study of attention-related disorders and its potential application in the hearing aid and advertising industries. The underlying neural basis of auditory attention is not well established, however evidence exists to suggest that cortical activity entrainment to the temporal envelope of speech is modulated by attention. Leveraging this finding, we introduce a probabilistic approach based on Hidden Markov Model Regression (HMMR) to decode the attentional state of the listener with respect to a given speech stimulus. Our method is novel in that it uses only the target stream to detect the attended segments, while existing methods require knowledge about the target and distractor. This is a particular advantage in real-world applications where the number of sources are often time-variant and unknown to the decoder. We use synthetic data to evaluate robustness and tracking capability, and real electrophysiological data to demonstrate how the proposed method achieves accuracies commensurate to BCI (Brain Computer Interface) systems deployed in the field.