-
A Transcript for a Conference Paper
Presentation
Slide 1 (Title):
Time: 35 seconds
(0:00-0:35)
Say:
Good morning! My major is electric
engineering. My research field is signal and
information
processing.
The
topic
I
concern
is
about
speech
enhancement.
Today
I will
give
a
presentation
with
the
title
“
A
Two-stage
Beamforming
Approach
for
Noise
Reduction
and
Dereverberation
”.
This work is done by Habets and Benesty
from University of Quebec, Montreal, Canada.
Slide 2 (Introduction)
Time: 55 seconds (0:35-1:30)
Say:
First, I
will give a brief introduction of microphone
arrays and make you have a preliminary
understanding of the research problem.
Simply, when you place several microphones
according
to certain geometric shapes,
you get a microphone array. Here is a liner array
and a circle array is
also
general
. Let’s take this picture as an
example. There is a noise source at the location
of the
star.
When
the
girl
speaks,
her
sound
will
get
captured
by
all
microphones.
In
the
same
time,
received signals are
polluted by the undesired noise
and we
can’t have the clean speech.
Slide 3 (Introduction)
Time: 55 seconds
(1:30-2:25)
Say:
So,
we
need
do
something
to
solve
this
problem.
In
this
slide,
the
background
and
significance of speech denosing and
dereverberation are introduced. In many
applications, such
as
speech
recognition
and
teleconferencing,
we
need
distant
or
hand-free
audio
acquisition.
However,
in
many
cases,
we
just
receive
a
noise
corrupted
or
reverberant
version
of
desired
speech
signals.
To
achieve
high-quality
human-to-human
or
human-to-machine
speech
communication,
we
need
to
develop
efficient
noise reduction
and
dereverberation
algorithms.
Microphone
arrays can be very useful in these situation.
Slide 4 (Body)
Time: 50 seconds
(2:25-3:05)
Say:
First, we can use beamforming for the
microphone arrays during the processing of
received
signals. What is the
beamforming? Here are its definition and working
principle. Beamforming is
a signal
processing technique that applies microphone
arrays for directional signal transmission
or reception. By operating on received
multichannel signals, beamforming allows us to
recover
signals
from
a
particular
direction
and
suppress
noise
signals
from
undesired
directions.
This
technique
is
so called “beamforming”.
Slide 5 (Body)
Time: 45 seconds (3:05-3:55)
Say:
In
this
paper,
we
use
beamforming
to
achieve
noise
reduction
and
dereverberation.
Noise
reduction
is
important
since
noise
is
everywhere
around
us.
Some
common
noise
includes
machine
noise,
vehicle
noise,
music
noise,
babble
noise,
and
so
on.
On
the
other
hand,
the
reverberation is created
when a sound is produced in an enclosed space
causing a large number
of echoes to
build up and then slowly decay as the sound is
absorbed by the walls and air.
Slide 6 (Body)
Time: 60 seconds (3:55-4:55)
Say:
To achieve
both noise reduction and dereverberation, the two-
stage approach is proposed in
this
paper and before the noise reduction stage, a
dereverberation stage is needed. Here is the
principle diagram. These y represent
the reserved signals by the microphone array and
we have N
microphone.
These
Q
and
H
represent
weighting
coefficients
of
two
different
beamforming
stages. The Z
represents the final signal after noise reduction
and
dereverberation. In the next
few slides, the details about how the
algorithm work are given.
Slide 7
(Body)
Time: 45 seconds
(4:55-5:45)
Say:
The
first
stage
is
dereverberation
stage.
In
this
slide,
the
computational
process
of
dereverberation stage is presented. All
channel inputs are
weighted
.
The weighted channel inputs
are sent to
the next stage for noise reduction. So the key is
to find proper weights so that the
reverberation
components
are
minimized.
This
is
implemented
by
complex
mathematics
computation. The final weights are
independent on signals.
Slide 8 (Body)
Time: 45 seconds
(5:45-6:25)
Say:
On
the
basis,
further
analysis
of
the
dereverberation
stage
is
needed.
The
first
stage
comprises
a
signal-independent
beamformer
that
generates
a
reference
signal
that
contains
a
dereverberated version of the desired
speech and residual interference. In general, the
desired
speech
component
at
the
output
of
the
beamformer
contains
less
reverberation
compared
to
reverberant speech signal
received at the microphones.