Digital Signal Processing Reference
In-Depth Information
1.
INTRODUCTION
Adaptive noise cancellation through Subband Adaptive Filtering (SAF)
has shown good performance when the two input noises are correlated [1].
However, many real-life noise fields in mobile and vehicular applications are
correlated only at lower frequencies because they are approximately diffuse
[2].
Diffuse noise fields are mathematically characterized through the spatial
coherence function commonly used to specify the correlation of two noise
signals x and y recorded at two input microphones in a noise field. The
spatial coherence function is defined based on the cross- and auto-spectral
densities as [3]:
For 3-dimensional diffuse noise fields, the spatial coherence, obtained by
averaging Eq. (1) over spherical coordinates, drops with frequency,
following a
of the form [2,3]:
where
c
is the sound velocity in air) and
d
is the distance
between the two input microphones. Considering Eq. (2), it is obvious that
SAF (and any adaptive noise cancellation method based on correlation
cancellation) can only eliminate the noise in the lower frequency regions
where there is a high correlation (coherence) between the two microphone
signals.
To compensate for the inability of SAF to eliminate noise in diffuse noise
fields, we have proposed a hybrid system integrating the SAF system and
Wiener filtering (called SAFWF here), and have examined its performance in
an isolated non-reverberant sound room [4]. Based on the promising results
obtained, we further evaluate the performance of the SAFWF system in real-
life mobile and vehicular noisy environments.
This chapter is organized as follows. The employed speech enhancement
system is described in the next section. Section 3 describes the noise and
speech materials used. System evaluations are reported in Section 4, and
finally conclusions of this work are presented in Section 5.
