This paper describes the design and experimentation of a Kalman Filter used to improve position tracking of a 3-D gesture-based musical controller known as the Radiodrum. The Singer dynamic model for target tracking is used to describe the evolution of a Radiodrum’s stick position in time. The autocorrelation time constant of a gesture’s acceleration and the variance of the gesture acceleration are used to tune the model to various performance modes. Multiple Kalman Filters tuned to each gesture type are run in parallel and an Interacting Multiple Model (IMM) is implemented to decide on the best combination of filter outputs to track the current gesture. Our goal is to accurately track Radiodrum gestures through noisy measurement signals.