Multimedia conferencing is the real-time exchange of multimedia content between multiple parties. It is the basis of several interactive multiuser applications, such as distance learning and multimedia multiplayer online games. The cloud-based provisioning of the conferencing services on which these applications rely on can have several benefits, including the easy provisioning of new applications, efficient use of resources, and elastic scalability. This paper proposes a holistic cloud-based architecture for conferencing service provisioning, which covers both the infrastructure and platform layers of the cloud. The proposed infrastructure layer offers conferencing substrates-as-a-service (e.g., dial-in signaling, video mixing, and audio mixing), instead of virtual machines or containers. The platform layer abstracts the details of the conferencing concepts and offers a high-level interface to simplify conference service provisioning for a wide range of service and application providers (experts versus non-experts). It also enables the on-the-fly scaling of the running conferences while guaranteeing the required quality of service, enables substrates composition to create new conferencing services, and eases the reuse of conferencing services in building new applications. The presented architecture is supported by a proof-of-concept prototype and performance measurements. The latter provides the analysis of resource allocation efficiency and response time, as well as the scalability of the system under suboptimal and over-provisioned conditions. It also provides recommendations for service providers regarding the best alternatives for provisioning their service.