Sincronia de sistemas multimídia distribuídos utilizando técnicas de aprendizagem de máquina

Abstract

The exchange of multimedia data streams in a distributed scenario represents a challenge mainly because of two problems generally present in this context: the synchronization between several streams and the management of heterogeneous hardware present in the system. The difficulty in maintaining the time limits necessary for the reproduction of time-dependent multimedia data stream increases with disturbances introduced by the communication channel, which is directly related to the quality of service. On the Internet, streaming real-time multimedia data requires efficient synchronization and storage strategies, which can add a significant amount of latency and interfere with the end-user quality of experience (QoE). Among the applications that require low latency to ensure an acceptable QoE, the composition and reproduction of songs through the Internet represent the biggest challenge. Towards this direction, we propose an effective method for low latency synchronization of several streams in a distributed global multimedia network, such as the Internet. Throughout the experiments we made to verify our proposed systems, we identified good accuracy levels on the synchronization of audio and video streams. We specifically emphasize the accuracy levels of 72% for the synchronization of video streams and up to 93% for the synchronization of raw audio sequences. In order to achieve these results, we employed statistical analysis of superior order and neural networks, both convolutional and recurrent, to compute a time relationship between multimedia flows.