Biometric person authentication is a secure and user-friendly way of identifying persons in a variety of everyday applications. In order to achieve high recognition rates, we propose an audio-visual person recognition system based on voice, lip motion and still image. The combination of these three data sources (called sensor fusion) may be performed in several ways. In this paper we present a method for sensor normalization based on statistical sensor properties. We call to this procedure sensor calibration. The final decision fusion simplifies to a multiplication or addition of the normalized outputs of each sensor. This approach is evaluated on a large database of 170 people with a total of 6315 recordings.