Agents that Listen: High-Throughput Reinforcement Learning with Multiple Sensory Systems