In
order
to
solve
the
problem
that
current
bearing
fault
diagnosis
model
based
on
voiceprint
signal
is
not
enough
extract
time
features,
a
3DCNN
proposed
in
this
paper.
First,
Mel-spectrogram
used
of
bearing.
Then,
diagnose
make
better
use
timing
information
model.
Finally,
paper
has
improved
precision
and
recall
rate
by
6.25%
7.03%
respectively
compared
with
classical
algorithm.
The
good
accuracy
important
for
engineering
practice.
Automation in Construction,
Год журнала:
2024,
Номер
165, С. 105485 - 105485
Опубликована: Май 31, 2024
This
paper
presents
a
model
for
sound
classification
in
construction
that
leverages
unique
combination
of
Mel
spectrograms
and
Mel-Frequency
Cepstral
Coefficient
(MFCC)
values.
combines
deep
neural
networks
like
Convolution
Neural
Networks
(CNN)
Long
short-term
memory
(LSTM)
to
create
CNN-LSTM
MFCCs-LSTM
architectures,
enabling
the
extraction
spectral
temporal
features
from
audio
data.
The
data,
generated
activities
real-time
closed
environment
is
used
evaluate
proposed
resulted
an
overall
Precision,
Recall,
F1-score
91%,
89%,
respectively.
performance
surpasses
other
established
models,
including
Deep
(DNN),
CNN,
Recurrent
(RNN),
as
well
these
models
CNN-DNN,
CNN-RNN,
CNN-LSTM.
These
results
underscore
potential
combining
MFCC
values
provide
more
informative
representation
thereby
enhancing
noisy
environments.
IEEE Access,
Год журнала:
2023,
Номер
11, С. 60078 - 60108
Опубликована: Янв. 1, 2023
The
concept
of
Acoustic
Source
Identification
(ASI),
which
refers
to
the
process
identifying
noise
sources
has
attracted
increasing
attention
in
recent
years.
ASI
technology
can
be
used
for
surveillance,
monitoring,
and
maintenance
applications
a
wide
range
sectors,
such
as
defence,
manufacturing,
healthcare,
agriculture.
signature
analysis
pattern
recognition
remain
core
technologies
source
identification.
Manual
identification
acoustic
signatures,
however,
become
increasingly
challenging
dataset
sizes
grow.
As
result,
use
Artificial
Intelligence
(AI)
techniques
relevant
useful.
In
this
paper,
we
provide
comprehensive
review
AI-based
techniques.
We
analyze
strengths
weaknesses
processes
associated
methods
proposed
by
researchers
literature.
Additionally,
did
detailed
survey
machinery,
underwater
applications,
environment/event
recognition,
other
fields.
also
highlight
research
directions.
Scientific Journal of Astana IT University,
Год журнала:
2025,
Номер
unknown, С. 172 - 185
Опубликована: Март 30, 2025
Road
accidents
continue
to
pose
a
serious
threat
public
safety,
underscoring
the
need
for
innovative,
automated
emergency
response
systems.
This
study
presents
development
of
mobile
application
that
detects
road
by
analyzing
audio
signals
in
real
time
and
immediately
sends
SMS
alerts
with
GPS
coordinates
services
user-specified
contacts.
The
system
comprises
two
parts:
user-facing
Android
server-side
component
data
processing.
To
build
train
detection
models,
we
leverage
MIVIA
Audio
Events
dataset
applied
preprocessing
techniques
including
amplitude
normalization,
background
noise
filtering,
augmentation.
Feature
extraction
involved
zero-crossing
rate,
spectral
centroid,
flux,
energy
entropy,
short-time
Fourier
transform
(STFT),
Mel-frequency
cepstral
coefficients
(MFCCs).
Two
classification
approaches
were
investigated:
traditional
machine
learning
models
(Support
Vector
Machine,
Random
Forest,
Gradient
Boosting)
deep
model
based
on
convolutional
neural
networks
(CNNs)
using
Mel
spectrogram
inputs.
Experimental
results
demonstrate
CNN
achieved
highest
performance
91.2%
accuracy,
89.5%
recall,
an
F1-score
90.3%,
outperforming
best
classical
(Random
Forest),
which
85.1%
accuracy.
also
reduced
average
accident
alert
from
5–7
minutes
1–2
minutes,
representing
60–80%
improvement
speed.
These
confirm
system’s
reliability
practical
benefit,
particularly
regions
like
Kazakhstan,
where
timely
medical
intervention
is
critical.
Limitations
include
reliance
smartphone
availability,
internet
access,
environmental
sound
conditions.
Future
work
will
explore
real-world
testing,
integration
accelerometer
gyroscope
data,
deployment
edge
computing
faster
on-device
Overall,
proposed
solution
cost-effective,
scalable
approach
improving
safety
saving
lives
through
rapid,
detection.