Figures for Chapter 7 Advanced signal processing

1
Figures for Chapter 7Advanced signal processing

• Dillon (2001)
• Hearing Aids

2
Fixed directional arrays
-
Subtractive array

?
T
Output
Figure 7.1 (a) Block diagram of a subtractive
directional microphone comprised of either a
single microphone with two ports, or two separate
microphones with one port each. The negative sign
next to one of the inputs of the summer indicates
that the two signals are subtracted. (b) A
delay-and-add directional microphone array with
four ports.
Source Dillon (2001) Hearing Aids
3
Frontal sensitivity and port spacing
Figure 7.2 Frontal sensitivity of a two-port (or
two-microphone) subtractive directional
microphone relative to the sensitivity of an
equivalent single-port microphone. The parameter
shown is the port spacing. The internal delay
needed to produce a cardioid polar response has
been assumed.
Source Dillon (2001) Hearing Aids
4
Figure 7.3 End-fire and broadside microphone
arrays.
Source Dillon (2001) Hearing Aids
5
Adaptive directional microphone
Front
?
-

?
T
Output
Figure 7.4 A simple adaptive directional
microphone with steerable nulls.
Source Dillon (2001) Hearing Aids
6
Widrow LMS noise reduction
Speech Noise

?
Delay
-
Noise
Figure 7.5 The Widrow Least Mean Squares
adaptive noise reduction scheme, based on a
reference microphone that picks up only the
noise. The fixed delay compensates for the delay
inherent in the adaptive filter.
Source Dillon (2001) Hearing Aids
7
Griffiths Jim adaptive noise reduction
Right


?
?
Delay
-


?
-
Left
Figure 7.6 A Griffiths-Jim adaptive noise
canceller, whereby the two microphone outputs are
added in the top chain but subtracted in the
bottom chain.
Source Dillon (2001) Hearing Aids
8
Microphone array benefit
Figure 7.7 Improvement in speech reception
threshold for an adaptive array relative to a
single microphone. The experiment used frontal
speech and a single noise masker at 45 degrees
from the front in three simulated environments
that differed in the amount of reverberant sound
relative to the direct sound. From Hoffman et al
(1994).
Source Dillon (2001) Hearing Aids
9
Blind source separation


?
R1
S1
Y1

G1
R4
G2
R3

Y2
?
S2
R2


Figure 7.8 Blind source separation of two
sources, S1 and S2, occurs when the two adaptive
filters, G1 and G2, adapt to the response shapes
that compensate for the room transmission
characteristics, R1, R2, R3 and R4, from each
source to each microphone. Note that everything
to the right of the dotted line is in the hearing
aid, whereas the blocks to the left are the
transfer functions of the transmission paths
within the room. When properly adapted, the
response of G1 R3/R1 and G2 R4/R2. The
output Y1 then does not contain any components of
S2. The blocks G1 and G2 can alternatively be
feed-forward blocks rather than feed-back blocks.
Source Dillon (2001) Hearing Aids
10
Wiener filter noise reduction
Input
Noise
Averager
-
Avg speech Spectrum

?
F.T.
Switch

Avg speech plus noise spectrum
Speech plus noise
Averager
Speech/non-speech detector
Figure 7.9 A Wiener Filter incorporating a
Fourier Transform (F.T) to calculate the spectrum
of the combined speech and noise. A
speech/non-speech detector classifies the
spectrum as noise or speech plus noise, and thus
enables the average spectral power of the speech
to be estimated.
Source Dillon (2001) Hearing Aids
11
Spectral Subtraction
Phase

Magnitude
?
F.T.
I.F.T.
-
Avg noise Spectrum
Switch
Averager
Speech/non-speech detector
Figure 7.10 A Spectral Subtraction noise
reduction system incorporating a Fourier
Transform to calculate the power spectrum, a
speech/non-speech detector to enable the average
spectral power of the noise to be estimated, and
an Inverse Fourier Transform to turn the
corrected spectrum back into a waveform.
Source Dillon (2001) Hearing Aids
12
Feedback management
Figure 7.11 The gain-frequency response of a
(hypothetical) four-channel hearing aid, where
feedback oscillationhas been avoided by
decreasing the gain of the band from 2 kHz to 4
kHz (solid line) from the original response
(dotted line).
Source Dillon (2001) Hearing Aids
13
Feedback-loop response
Figure 7.12 Gain-frequency and phase-frequency
response of the complete feedback loop for an ITE
hearing aid. Redrawn from Hellgren et al.,
(1999).
Source Dillon (2001) Hearing Aids
14

?
Figure 7.13 Internal feedback path added to
cancel the effects of the external, unintentional
leakage path.
Source Dillon (2001) Hearing Aids
15
Transposition
Figure 7.14 Input and output spectra for a
frequency transposition scheme in which the
output frequency equals half the input frequency.
The amplifier also provides some high frequency
pre-emphasis. The arrows show the reduction in
frequency of each formant.
Intensity
1000
250
4000
Frequency
Intensity
1000
250
4000
Frequency
Source Dillon (2001) Hearing Aids
16
Spectral enhancement
Figure 7.15 Spectrograms of the syllable /ata/
(a) unprocessed and (b) spectrally enhanced,
showing more pronounced formants (Fisher, Dillon
Storey, in preparation).
Time (seconds)
Source Dillon (2001) Hearing Aids