A method and networking architecture for implementing service voice dynamic loading on an intelligent network is provided, wherein a voice file and a configuration file are scanned by a voice equipment to perform the voice dynamic loading based on voice requirement of the configuration file at any time. Service voice dynamic loading can be implemented automatically through setting loading conditions and the configuration file. A service voice can be managed by a Service Management System (SMS), rather than only by a switching system. During the service voice loading, the operation will not be affected. The hardware operation is not necessary for loading of the service voice. Furthermore, the voice processing can be performed in a "multiple-to-multiple" mode.