The compliant nature of soft wearable robots makes them ideal for complex multiple degrees of freedom (DoF) joints, but also introduce additional structural nonlinearities. Intuitive control of these wearable robots requires robust sensing to overcome the inherent nonlinearities. This paper presents a joint kinematics estimator for a bio-inspired multi-DoF shoulder exosuit capable of compensating the encountered nonlinearities. To overcome the nonlinearities and hysteresis inherent to the soft and compliant nature of the suit, we developed a deep learning-based method to map the sensor data to the joint space. The experimental results show that the new learning-based framework outperforms recent state-of-the-art methods by a large margin while achieving 12ms inference time using only a GPU-based edge-computing device. The effectiveness of our combined exosuit and learning framework is demonstrated through real-time teleoperation with a simulated NAO humanoid robot.