Tempo tracking by using a parallel particle filter on the GPU

Recently, using graphics processing unit (GPU) for accelerating applications is becoming very popular and particle filters are no exception. Tempo tracking is one of the basic problems in music processing. In this paper, we present an implementation of the bar pointer model with a particle filter on CUDA. We describe the algorithms used to implement the parallel particle filter. Then, in order to demonstrate the implementation, we create a simulated observation data and run the filter on it. We also give the running times of the application for different number of particles.