System and method for frequency domain audio post-processing based on perceptual masking
In an embodiment, a method of frequency domain post-processing is disclosed. The method includes applying adaptive modification gain factor to each frequency coefficient, and determining gain factors based on Local Masking Magnitude and Local Masked Magnitude.
1. A method of post-processing of a frequency domain audio signal implemented by an audio post-processor, the method comprising:
applying adaptive modification gain factor to each frequency coefficient of the frequency domain audio signal by using the audio post-processor; and
determining gain factors based on Local Masking Magnitude and Local Masked Magnitude;
wherein the Local Masking Magnitude and Local Masked Magnitude are estimated according to perceptual masking effects,
wherein the Local Masking Magnitude is estimated by taking a weighted sum around a specific frequency at i:
M
0
(
i
)
=
∑
k
w
0
i
(
k
)
·
F
0
(
i
+
k
)
,
where M 0 (i) is the Local Masking Magnitude, w 0 i (k) is a first frequency dependent weighting window, F0(i) are frequency coefficients of the frequency domain audio signal before the post-processing is applied, and k is an index value,
wherein Local Masked Magnitude M 1 (i) is estimated by taking a weighted sum the specific frequency at i:
M
1
(
i
)
=
∑
k
w
1
i
(
k
)
·
F
0
(
i
+
k
)
,
where M 1 (i) is the Local Masked Magnitude, w 1 i (k) is a second frequency dependent weighting window, and wherein weighting window w 1 i (k) is flatter and longer in the frequency domain than w 0 i (k), and
wherein an initial gain factor for each frequency is
Gain
0
(
i
)
=
M
0
(
i
)
α
·
M
1
(
i
)
+
(
1
-
α
)
·
M
av
,
where i is a frequency index, M av is an Average Magnitude, and 0≦α≦1.
2. The method of claim 1 , wherein the audio post-processor performs post-processing in a Modified Discrete Cosine Transform (MDCT) domain or a Fast Fourier Transform (FFT) domain.
3. A method of post-processing of a frequency domain audio signal implemented by using an audio post-processor, the method comprising:
applying adaptive modification gain factor to each frequency coefficient of the frequency domain audio signal by using the audio post-processor; and
determining gain factors based on Local Masking Magnitude, Local Masked Magnitude, and Average Magnitude;
wherein the Local Masking Magnitude is estimated by taking a weighted sum around a specific frequency at i:
M
0
(
i
)
=
∑
k
w
0
i
(
k
)
·
F
0
(
i
+
k
)
,
where M 0 (i) is the Local Masking Magnitude, w 0 i (k) is a first frequency dependent the weighting window, F0(i) are frequency coefficients of the frequency domain audio signal before the post-processing is applied, and k is an index value,
wherein Local Masked Magnitude M 1 (i) is estimated by taking a weighted sum the specific frequency at i:
M
1
(
i
)
=
∑
k
w
1
i
(
k
)
·
F
0
(
i
+
k
)
,
where M 1 (i) is the Local Masked Magnitude, w 1 i (k) is a second frequency dependent weighting window, and wherein weighting window w 1 i (k) is flatter and longer in the frequency domain than w 0 i (k),
wherein the Average Magnitude is calculated on a whole spectrum band of the frequency domain audio signal, and
wherein an initial gain factor for each frequency is
Gain
0
(
i
)
=
M
0
(
i
)
α
·
M
1
(
i
)
+
(
1
-
α
)
·
M
av
,
where i is a frequency index, M av is the Average Magnitude, and 0≦α≦1.
4. The method of claim 3 , wherein the Average Magnitude is calculated by:
M
av
=
∑
k
F
0
(
k
)
/
N
F
,
wherein M av is the Average Magnitude, NF is a total number of the frequency coefficients, and k is an index value.
5. The method of claim 3 , wherein:
M
av
=
∑
k
F
0
(
k
)
/
N
F
;
and
NF is a total number of the frequency coefficients.
6. The method of claim 5 , wherein the first frequency dependent weighting window is asymmetric and the second frequency dependent weighting window is asymmetric.
7. The method of claim 5 , wherein gain factors are normalized according to:
Gain 1 ( i )=Gain 0 ( i )·Norm ,
wherein normalization factor Norm is defined as,
Norm
=
∑
i
F
0
(
i
)
2
∑
i
Gain
0
(
i
)
·
F
0
(
i
)
2
.
8. The method of claim 7 , wherein the normalized gain factors can be controlled by parameter β such that:
Gain 2 ( i )=β·Gain 1 ( i )+(1−β)
wherein (0 ≦β≦1), β is a parameter that controls strong post-processing or weak post-processing.
9. The method of claim 8 , wherein β is replaced by a smoothed controlling parameter β , such that:
β 0.75 β +0.25β.
10. The method of claim 3 , wherein determined gain factors are multiplied with the frequency coefficients to produce post-processed frequency coefficients.
11. The method of claim 3 , further comprising receiving the frequency domain audio signal from a voice over internet protocol (VOIP) network.
12. The method of claim 3 , further comprising receiving the frequency domain audio signal from a mobile telephone network.
13. The method of claim 3 , further comprising converting the post-processed frequency domain signal into a time domain audio signal.
14. A system for receiving a frequency domain audio signal, the system comprising a post-processor configured to:
apply an adaptive modification gain factor to each frequency coefficient of the frequency domain audio signal; and
determine gain factors based on Local Masking Magnitude and Local Masked Magnitude and Average Magnitude,
wherein the post-processor estimates the Local Masking Magnitude by taking a weighted sum around a specific frequency at i:
M
0
(
i
)
=
∑
k
w
0
i
(
k
)
·
F
0
(
i
+
k
)
,
where M 0 (i) is the Local Masking Magnitude, w 0 i (k) is a first frequency dependent weighting window, F0(i) are frequency coefficients of the frequency domain audio signal before the post-processing is applied, and k is an index value,
wherein the post-processor estimates the Local Masked Magnitude M 1 (i) by taking a weighted sum the specific frequency at i:
M
1
(
i
)
=
∑
k
w
1
i
(
k
)
·
F
0
(
i
+
k
)
,
where M 1 (i) is the Local Masked Magnitude, w 1 i (k) is a second frequency dependent weighting window, and wherein weighting window w 1 i (k) is flatter and longer in the frequency domain than w 0 i (k),
wherein the post-processor calculates the Average Magnitude on a whole spectrum band of the frequency domain audio signal, and
wherein the post-processor calculates an initial gain factor Gain 0 (i) for each frequency according to:
Gain
0
(
i
)
=
M
0
(
i
)
α
·
M
1
(
i
)
+
(
1
-
α
)
·
M
av
,
where i is a frequency index, M av (i) is the Average Magnitude, and 0≦α≦1.
15. The system of claim 14 , wherein:
M
av
=
∑
k
F
0
(
k
)
/
N
F
;
and
NF is a total number of the frequency coefficients.
16. The system of claim 14 , wherein the system is configured to operate over a voice over internet protocol (VOW) system or a cellular telephone network.
17. The system of claim 14 , further comprising an audio decoder configured to receive audio parameters and produce the audio signal based on the received audio parameters.
18. The system of claim 14 , wherein the receiver is further configured to convert an output of the post-processor to an output audio signal.
19. The system of claim 18 , wherein the output audio signal is configured to be coupled to a loudspeaker.