Reliably and efficiently predicting room acoustic parameters in rectangular rooms with an uneven distribution of materials and sound absorption is a common task in room acoustic design. Readily available statistical formulas are often not applicable to typical classrooms, cellular offices, or hospital wards. Furthermore, many of these methods only predict the reverberation time, each with their own, often ambiguous limitations of applicability which is impractical for an efficient room acoustic design process. On the other hand, predicting room acoustic parameters by means of ray-tracing or other numerical methods might not be economically feasible for a large number of spaces in many development projects. In this paper, we present a deep learning framework and results based on a sequence processing recurrent neural network to predict several room acoustic parameters. The deep learning models yield prediction speed improvements over ray-tracing or numerical methods, and also significantly improve accuracy over traditional statistical estimation formulas for rectangular rooms. We present the implementation of the method in the soundy.ai application.