Digital Signal Processing Reference
In-Depth Information
4.1
Constrained Switched Adaptive Array-Processing
(CSA-BF)
The proposed CSA-BF array processing algorithm consists of four parts: a
constraint section (CS), a speech adaptive beamformer (SA-BF), a noise
adaptive beamformer (NA-BF) and a switch. Figure 2-4 shows the detailed
structure of CSA-BF, for a 5-microphone array. The CS is designed to
identify potential speech and noise locations. If a speech source is detected,
the switch will activate SA-BF to adjust the beam pattern and enhance the
desired speech. At the same time, NA-BF is disabled to avoid speech leakage.
If however, a noise source is detected, the switch will activate NA-BF to
adjust the beam pattern for noise and switch off SA-BF processing to avoid
the speech beam pattern from being altered by the noise. The combination of
SA-BF and NA-BF processing results in a framework that achieves noise
cancellation for interference in both time and spatial orientation. Next, we
consider each processing stage of the proposed CSA-BF scheme.
4.1.1
Constraint Section
Many source localization methods have been considered in the past with
effective performance for large microphone arrays in conference rooms or
large auditoriums. Their ability to perform well in changing noisy car
conditions has not been documented to the same degree, but is expected to be
poor. Here, we propose three practical constraints that can be used to separate
speech and noise sources with high accuracy.
Criterion 1 (Maximum averaged energy):
Since speech coming from the
driver's direction will have on average the highest intensity of all sources
present, therefore, we calculate the averaged signal TEO energy [18]
frame by frame, and if this energy is greater than some threshold (Please
refer to [19] for the details, we take the current signal frame as speech
candidate.
Criterion 2 (LMS adaptive filter):
In order to separate the front-seat driver
and passenger, we choose the adaptive LMS filter method and incorporate
the geometric structure of the microphone array to locate the source.
Criterion 3 (Bump noise detector)
This final criterion is set to avoid
instability in the filtering process which is affected by impulsive noise
with high-energy content, such as road impulse/bump noise.
