Dual Particle-Number RBPF for Speech Enhancement

In this paper, we propose a new single channel dual particle-number Rao-Blackwellized particle filter (RBPF). Additive noise i.e. white and color noises corrupt speech signal and degrade its intelligibility and quality. Quality measurement scores are ITU-T P.862.1 (PESQ), also computation cost in implementation are important. Particular emphasis is placed on the removal of colored noise, such as industrial noise. At first describe some of the similar method such as Kalman filter and particle filter. The simulation results show that the proposed method provides a significant gain in ITU-T P.862.1 score. Taking measure to reduce computational complexity by separating silent-speech and assign different particle number to each type of frames.