The iterated extended kalman particle filter for speech enhancement

Particle filters have been proposed as a new form of state-space filtering for speech enhancement applications. A crucial issue in particle filtering is the selection of the importance proposal distribution. In this paper, the iterated extended Kalman filter (IEKF) is used to generate the proposal distribution. The proposal distribution integrates the latest measurements into state transition density, so it can match the posteriori density well. We apply time-varying autoregressive (TVAR) models with stochastically evolving parameters to the problem of speech modeling and enhancement, which is superior to conventional AR models. The experimental results indicate that the new particle filter superiors to the standard particle filter and the other filters such as the extended Kalman particle filter (PF-EKF) in low SNR.