The Shannon energy of the heart sound record with S3 was shown in Figure 2(c).The threshold value was initially set to 70% of maximal value. If the threshold is chosenupper than 70%, some S2 with lower peak would be excluded and missed. But forthreshold lower than 70%, noise, murmur, or additional heart sound would easily bepicked up and affect the results of S1 and S2 recognition. Nevertheless, higher thresholdis rather better because it would be easier to re-pick the missing S2 than dealingwith noises. S1 and S2 were recognized by three steps in the following [24]:
a. If two peaks higher than the threshold were detected within 50 ms, the one withlower energy was eliminated.
b. For every interval between peaks, an interval with shorter length than the previousinterval was denoted as a systolic period, while the other one was a diastolic period.The uncertain intervals were annotated.
c. For those uncertain intervals, a secondary threshold was set to find S1 or S2 whichprobably not have been recognized.These steps were applied to ensure that S1 and S2 could be picked out correctly.Systolic and diastolic periods were then recognized. S3 and S4 with larger amplitudes could also be detected during segmentation.