A robust and fast method for estimating and tracking the instantaneous fundamental frequency of audio signals

Extended Kalman filter frequency tracker converges to an optimal estimation with a very low computational complexity only if an exact knowledge about the initial state is available and the model is accurate otherwise it may diverge. On the other hand, particle filter performance is not so sensitive to the initial state estimation and results in a more robust estimation even if no prior knowledge about the initial state is available but at a cost of much more computational complexity. In this paper, we propose a hybrid algorithm that takes advantages of both extended Kalman and particle filter in order to achieve a robust and fast algorithm for estimating the fundamental frequency of audio signals. Simulation results reveal that this hybrid algorithm is robust, fast and accurate enough for real-time applications.