Fusion plays an important role to retrieve the single output data from the input data set and in turn improves the quality of the results. Single channel based separation of speech process is to separate the utter speech from the composite speech recorded with a single mic. This research work proposed a weight fusion scheme to combine the voiced matrix and unvoiced speech matrix obtained using feature based single channel speech separation systems. The voiced segment matrix obtained using periodicity features and unvoiced segment matrix obtained using onset/offset feature are combined using the proposed weight fusion principle. The proposed system is evaluated using the standard speech and noise database. The experimental outcome results of the weight fusion system reveal that there is a better improvement than the other speech separation systems.