Traffic speed forecasting becomes a thriving research area in modern transportation systems. The intensification of travel flow volumes due to fast urbanization, vehicle path planning, demands on efficient transport planning policies, commercial objectives, and many other factors contribute to fuel this revival dynamics. Moreover, predicting vehicle speed is of paramount importance in congestion management to help transport authorities as well as network users to handle congestion over road infrastructures or to provide a global overview of daily passenger flow. In this work, we propose a novel approach to forecast the future traffic speed of the road segments (links) based on traffic flow data without the need for previous traffic speed as input. To do this, we first pre-process floating car data of several million vehicles for multiples network links spread all over the Greater Paris area from 2016 to 2017. A convolutional attention-based recurrent neural network is used to capture the local-temporal features of traffic data to unveil the underlying pattern between the traffic flow and speed sequences for all links over the network. While the convolutional layer captures the local dependency, the attention layer learns patterns from weights of near-term traffic flow. It extracts the inherent interdependency of traffic speed due to many factors such as incidents, rush hour, land use, to cite a few, in non-free-flow conditions. The efficiency of the proposed model is evaluated using several metrics in traffic speed forecasting excluding additional data such as historical traffic speed and network graph contrary to cutting-edge work in the field. This is a substantial property since it allows avoiding the cumbersomeness in data mixing and facilitating resource availability. The proposed model is also evaluated on several roads located in the Greater Paris area separately on weekdays and weekends.