Audio analysis
As such it is quite easily described as an integral across analysis window w in Equation
(6.9).
S
x
(t, v) =
x
(u)w(u − t)e
−j2πvu
du
2
.
(6.9)
The instantaneous frequency energy at time t is then given by:
E
x
(t) =
1
2
S
x
(t, v)dv.
(6.10)
The Wigner–Ville distribution (WVD) is a bilinear distribution that is qualitatively dif-
ferent from the STFD. At any particular time, the signal contributions from the past are
multiplied by the signal contributions from the future in order to compute and correlate
between these left and right parts of the signal. It has the ability to give a clearer picture
of the instantaneous frequency and group delay than the spectrogram:
WV
x
(t, v) =
x
t
+
τ
2
x
t
−
τ
2
e
−j2πvτ
d
τ.
(6.11)
Similarly, the instantaneous frequency energy at time t is then given by:
E
x
(t) =
1
2
|WV
x
(t, v)| dv.
(6.12)
Although the WVD is perfectly localised on linear signals due to the forward-backward
nature of the analysis. However, if there are several frequency components existing
simultaneously, noise will be found in the WVD distribution due to the phenomenon of
cross-term interference between those components. It also provides equal weighting to
both past and future components.
The pseudo-Wigner–Ville distribution (PWVD) is an advance on the WVD since it
emphasises the signal properties near the time of interest compared to far away times.
A window function h
(τ) is used that peaks around τ = 0 to weight the Wigner–Ville
distribution toward an emphasis of the signal around time t. The PWVD is defined as:
PWV
x
(t, v) =
h
(τ)x
t
+
τ
2
x
t
−
τ
2
e
−j2πvτ
d
τ.
(6.13)
Again, the instantaneous frequency energy calculation at time t is relatively simple:
E
x
(t) =
1
2
|PWV
x
(t, v)| dv.
(6.14)
Finally, a fourth method exists as a refinement to PWVD. This is the reassigned smoothed
PWVD (refer to [8] for its application). It uses a centre-of-gravity reassignment method
with both time and frequency domain smoothing.
Do'stlaringiz bilan baham: |