12-07-2012, 11:51 AM
Dual Kalman Filtering for Speech Enhancement
[attachment=27482]
INTRODUCTION
The background noise is a dominant source of errors in speech recognition systems. Noise
reduction for speech signals has therefore application in entry procedures of those systems.
The Kalman filter is known in signal processing for its efficient structure. There are many
studies of using of Kalman filtering for noise reduction in speech signals, [1],[2], [3].
Speech signals are modeled as stationary AR process. I suggest modeling and filtering
noisy speech signals in the subband domain [1]. Since the power spectral densities (PSD’s)
of subband speech signals are flatter than their fullband signals, low-order AR models are
satisfactory and only lower-order Kalman filters will be required.
IMPLEMENTATION AND RESULTS
We used 4-channel filter bank with FIR filters and first order Kalman filter in each subband.
We implemented system for speech enhancement on PC in program Matlab 6.2.
We tested system by real signals like some czech sentences and words corrupted with
white noise generated by Matlab or real colored noise of running car and bus.
CONCLUSIONS
Simulations in Matlab showed that the Dual Kalman filtering reaches better results for
speech enhancement of speech signals corrupted by white and colored noises than the
system with NLMS algorithm estamitation of AR coefficients. The computation amount
is higher, because the first order Kalman filter is more complicated than the first order
NLMS algorithm but not so much in the fist order system using filter bank. The number
of unchanging system’s constants is lower. The constants have to be set before start of
the system and must be found by testing of the system. Dual Kalman Filtering makes
the system more simple for settings because there are only two unchanging constants.
Results are also better than the system with NLMS algorithm. The speech enhancement
is approximatelly about 1dB higher for all kind of noises.
[attachment=27482]
INTRODUCTION
The background noise is a dominant source of errors in speech recognition systems. Noise
reduction for speech signals has therefore application in entry procedures of those systems.
The Kalman filter is known in signal processing for its efficient structure. There are many
studies of using of Kalman filtering for noise reduction in speech signals, [1],[2], [3].
Speech signals are modeled as stationary AR process. I suggest modeling and filtering
noisy speech signals in the subband domain [1]. Since the power spectral densities (PSD’s)
of subband speech signals are flatter than their fullband signals, low-order AR models are
satisfactory and only lower-order Kalman filters will be required.
IMPLEMENTATION AND RESULTS
We used 4-channel filter bank with FIR filters and first order Kalman filter in each subband.
We implemented system for speech enhancement on PC in program Matlab 6.2.
We tested system by real signals like some czech sentences and words corrupted with
white noise generated by Matlab or real colored noise of running car and bus.
CONCLUSIONS
Simulations in Matlab showed that the Dual Kalman filtering reaches better results for
speech enhancement of speech signals corrupted by white and colored noises than the
system with NLMS algorithm estamitation of AR coefficients. The computation amount
is higher, because the first order Kalman filter is more complicated than the first order
NLMS algorithm but not so much in the fist order system using filter bank. The number
of unchanging system’s constants is lower. The constants have to be set before start of
the system and must be found by testing of the system. Dual Kalman Filtering makes
the system more simple for settings because there are only two unchanging constants.
Results are also better than the system with NLMS algorithm. The speech enhancement
is approximatelly about 1dB higher for all kind of noises.