IEEE Consumer Electronics Magazine - January/February 2023 - 82

Security, Trust and Privacy Solutions for Intelligent Internet of Vehicular Things
verification. The maximum accuracy of speech
emotion recognition is found to be 98.5%
observed for two case studies of anger emotion
in airport noise and traffic noise.
The results convey that the modulation
domain enhancement algorithm, when employed
as preprocessing stage, dramatically enhances
recognition of an emotion recognition system in
the presence of varying noise environmental conditions
and inputs SNRs for driver's safety in IoVT
applications.
PREPROCESSING METHODOLOGY:
ANALYSIS-MODIFICATION-
SYNTHESIS (AMS)
To present the application of the modulation
and phase spectra in automatic speech recognition
(ASR) system, we employ the AMS method
for speech enhancement as preprocessing. Consider
additive noise scenario as in
xðnÞ¼ sðnÞþNðnÞ
(1)
where x(n) is the discrete-time noisy voice signal,
s(n) is the discrete-time clean voice, and discretetime
background noise is indicated by N(n). The
discrete time index is represented as " n. " In our
whole discussion, when referring to magnitude or
phase spectrum in an acoustic AMS framework,
the discrete-time short-time Fourier transform
(STFT) with small window typically 20-40 ms is
employed until and unless stated
Xðn; kÞ¼
X1
l¼1
where l refers to an acoustic frame number, xðlÞ
is a discrete acoustic sample, WðnÞ is an acoustic
analysis window function, k refers to discrete
frequency, and M refers to frame duration in
samples. Q denotes frame shift. The maximum
frame shift of 40% (about 8-ms duration) can be
applied. Modified Hamming window is applied at
both acoustic processing and modulation processing,
since it has reported improved performance
as to other windows. Typically, an Mis
chosen between 20 and 40 ms and is applied in
speech processing. We apply spectral subtraction
on a modulation domain spectrum, as
shown in Figure 2. Thus, by applying STFT in (1),
Xðn; kÞ, Sðn; kÞ; and Nðn; kÞ are referred as a
82
xðlÞWðn QlÞ e
j2pkl
M
(2)
Figure 2. Preprocessing flowchart of a proposed
optimized modulation spectral subtraction for IoVT.
discrete-time STFT spectrum of noisy speech,
pure voice, and background noise, respectively.
AMS-BASED MODULATION ANALYSIS
FOR IoVT
The modulation domain enhancement is
achieved by extending acousticAMSprocedure discussed
in the section " Preprocessing Methodology:
Analysis-Modification-Synthesis (AMS). " In this
processing, the discrete-time STFT is computed
with the respective acoustic frequency using time
series of magnitudes jXRðn; kÞj at that frequency.
In modulation domain processing, the Hamming
window with 128- and 16-ms frame shift is used.
Algorithm 1 shows the flowchart of the proposed
OMSS method. m is termed as an index of
the discrete modulation frequency. In this section,
we discuss the concurrent effect of various noise
estimation methods with the proposed method.
One way is to employ a Voice Activity Detector
(VAD), but this will result in increase in the computational
complexity. It is observed that at a
large frame duration of modulation frame shift, no
IEEE Consumer Electronics Magazine
IEEE Consumer Electronics Magazine - January/February 2023

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - January/February 2023
