Mobile assistive robots can enhance elderly people's perception of safety and prevent loss of confidence at home. Therefore, multi-modal systems that allow robots to operate in complex environments represent an enticing milestone for self-care and independent living applications. We present a humanoid robot that assists a person in daily activities and detects situations of danger such as fall events. Our system integrates multiple sensor modalities to enhance the perception of the robot through visual active tracking, sound source localization, and automatic speech recognition. Robot motor control is triggered from the interplay of audio-visual cues conveyed by onboard sensors. We propose a multi-modal controller to modulate sensor-driven behaviour of the humanoid robot Nao and present preliminary results in a home-like environment for a fall detection scenario.