A framework is provided to offer composite voice applications and services. A composite application and service begins from the user dialing in via phone and ends by the user hanging up the phone. The composite interactive voice services architecture includes a session initiation protocol session service unit is in the loop of session signaling all the time starting from the time the user first dials in, during the user roaming across various voice servers, and until the end of the composite service when user hangs up the phone. This unit accepts a command and login instruction of the next interactive voice service from the previous interactive voice service. The unit has knowledge of DTMF sequences required for the user to login to next interactive voice service. The session service unit automatically accomplishes a roaming process such that composite applications and services can be achieved across various voice servers.