Monte Carlo smoothing with application to speech enhancement