A NEW ITERATIVE SPEECH ENHANCEMENT SCHEME BASED ON KALMAN FILTERING


  A NEW ITERATIVE SPEECH ENHANCEMENT SCHEME.doc (Size: 174 KB / Downloads: 12)

ABSTRACT

A new iterative speech enhancement scheme that can be seen as an approximation to the Expectation-Maximization (EM) algorithm is proposed. The algorithm employs a Kalman filter that models the excitation source as a spectrally white process with a rapidly time-varying variance, which calls for a high temporal resolution estimation of this variance.

. INTRODUCTION

Single channel noise reduction of speech signals using itera-tive estimation methods has been an active research area for the last two decades. Most of the known iterative speech enhancement schemes are based on, or can be interpreted as, the Expectation-Maximization (EM) algorithm or a certain approximation to it. Pro-posals of the EM algorithms for speech enhancement can be found in [1] [2] [3] [4] [5]. Some other iterative speech enhancement tech-niques can be seen as approximations to the EM algorithm, see e.g. [6] [7] [8] [9]. A paradigm of these EM based approaches is to iterate between an expectation step comprising Wiener or Kal-man filtering given the current estimate of signal model parameters, and a maximization step comprising the estimation of the parame-ters given the filtered signal.

INITIALIZATION AND SEQUENTIAL
APPROXIMATION

The Weighted Power Spectral Subtraction procedure combines the signal power spectrum estimated in the previous frame and the one estimated by the Power Spectral Subtraction method in the cur-rent frame, so that the iteration of the current frame is started with the result of the previous iteration as well as the new information in the current frame. The weight of the previous frame is set much lar-ger than the weight of the current frame because the signal spectrum envelope varies slowly between neighboring frames.

CONCLUSION

In this paper, a new iterative Kalman filtering based speech en-hancement scheme is presented. It is an approximation to the EM al- gorithm embracing the maximum likelihood principle. A high tem-poral resolution signal model is used to model voiced speech and the rapidly varying variance of the excitation source is estimated by a prediction-error Kalman filter.