The Speech detachment and contribute estimation loud conditions are thought to be a "chicken-and-egg" issue. Pitch data is an one of the essential prompt for discourse partition. Also, discourse detachment makes pitch estimation simpler when foundation commotion is expelled. In this paper, we propose a directed learning design to take care of these two issues iteratively. The proposed calculation depends on the profound stacking system (DSN), which gives a technique to stacking straightforward preparing modules to manufacture profound structures. Every module is a classifier whose objective is the perfect paired veil (IBM), and the information vector incorporates otherworldly components, pitch based elements and the yield from the past module. Amid the testing stage, we gauge the pitch utilizing the partition results and refresh the pitch-based components to the following module. At the point when inserted into the DSN, pitch estimation and discourse detachment each run a few times. To examination technique demonstrate that the proposed framework result in both a top notch assessed twofold cover and precise pitch estimation and outflanks late frameworks in its speculation capacity.