Cardiac auscultation is a non-invasive procedure that is mainly used in primary care, and involves diagnosis through analysis of the two heart sounds emanating from the cardiac cycle. None of the existing methods of computer aided auscultation can run in real time, and identify both S1 and S2 heart sounds, and operate without prior learning, and operate without making crude assumptions about the signal. This paper proposes a novel approach that is able to perform cardiac auscultation in real time and supports all of these features. Existing approaches try to identify and employ unique features of S1 and S2, which could be different for different patients, equipment, placement of stethoscope, background noise, etc. The proposed approach leverages the fact that distinct, dominant groups (e.g. Heart sounds S1 and S2) will naturally emerge if the the time-frequency content of each sound is thresholded, correlated with the other sounds, and finally clustered. The time-frequency information is derived from the continuous wavelet transform of the heart sound, and K-means is used for clustering. The system was tested on a dataset of 230 recordings with over 5000 heart sound pairs, and test results show a predictive rate for both heart sounds of above 86% -- on par with existing approaches. The system has been demonstrated working in real time, and an example application that uses this capability was developed.