Can Visual Context Improve Automatic Speech Recognition for an Embodied Agent?