mpeg surround
DESCRIPTION
Student: Shang-Yu Yeh Adviser : H.-M. Hang 2007/12/17. MPEG Surround. Outline. MPEG Surround Introduction T/F Resolution and DataMode Adaptive Parameter Smoothing Experimental Results. MPEG Surround Introduction. - PowerPoint PPT PresentationTRANSCRIPT
MPEG SURROUND
Student: Shang-Yu Yeh Adviser : H.-M. Hang
2007/12/17
Outline MPEG Surround Introduction T/F Resolution and DataMode Adaptive Parameter Smoothing Experimental Results
MPEG Surround Introduction Low-bitrate parametric coding technology
for multi-channel audio signal Backward compatibility to stereo
equipment Exploits inter-channel differences in level,
phase and coherence equivalent to the spatial cues to capture the spatial image of a multi-channel audio signal
Spatial cues: CLD, ICC, CPC
Spatial Audio CodingEncoder:
Decoder:
MPEG Surround EncoderArchitecture:
T/F TransformT/F Transform
T/F Transform
Downmix
SpatialParameterEstimation
AudioEncoder
CompressedAudio
Bitstream
Spatial Parameters
1s
2s
1x2x
Nx
F/T Transform
F/T Transform
MPEG Surround Encoder
T/F Resolution and DataMode Parameter extraction
reduce the accuracy of samples
71 hybrid subbands
parameterbands
parametersets
T/F Resolution and DataMode
New Architecture:
T/F TransformT/F Transform
T/F Transform
Downmix
SpatialParameterEstimation
AudioEncoder
CompressedAudio
Bitstream
Spatial Parameters
1s
2s
1x2x
Nx
F/T Transform
F/T Transform
MPEG Surround Encoder
TimeResolutionDecision
DataMode & Freq
ResolutionDecision
Adaptive Parameter
Smooth
T/F Resolution “Resolution” refers to the number of
parameters 0 1 2 3 0
Time(ps)
Frequency(pb)
pbStride
Time Resolution Decision Choose the number of parameter sets
for a frame Can be fixed or variable
0 1 2 3
0 1 2 3
DataMode Decision Exploit the correlation between
subsequent sets 4 modes: default, keep, interpolation
and lossless 0 1 2 3
parameter set
0 1 2
data set
Freq Resolution Decision Choose a stride that make the decoding
error minimum Exploit correlation between bands pbStride can be: 1, 2, 5, 28 0 1 2
data set
stride
stride
stride
Freq Resolution Decision Pairing: grouping of 2 parameter
subsets to share some data (QP, pbStride…)
Pairing of non-neighboring sets will not have gain 0 1 2
data set
pair
Combination of 2 modules
Adaptive Parameter Smoothing For artifacts from coarse quantization
and low up-date rate of spatial cues Especially in the case of stationary and
tonal signals First order IIR filtering of the parameter
bands
Adaptive Parameter Smoothing
1lW
lWlkonjW
BASdelta
A
B
)( 11 llkonjdelta
ll WWSWW
Smooth strengh:
Adaptive Parameter Smoothing Configure for each parameter set Error estimation:
Parameter normalization:
set
set
stddevmeanbpbp
)()( '
parametersresolutionhighestthefbpCPCICCCLDpbforparametersnormalizedpbp
bandfullfbbandparameterpb
fbppbpError
bandfull
boxIdx pbbandfullps
___:)'(,,:___:)'(
)28(_:_:
)'()'(
_
_
Experimental Results Add DataMode and Freq Resolution
Decision Modules without any loss Bitrate reduction(%):
ps1 ps2 ps4
Input01 4.22 2.52 2.36Input02 25.12 21.09 20.07Input03 6.67 3.89 3.54Input04 3.88 2.52 2.28Input05 4.63 2.77 2.57
(pb=10)
Experimental Results DataMode Decision:
Bitrate reduction(%):1 2 3 4 5
0%
20%
40%
60%
80%
100%
ps1_pb10
losslessinterpolationkeepdefault
1 2 3 4 50%
20%
40%
60%
80%
100%
ps1_pb10_dm7_111
losslessinterpolationkeepdefault
Theoretical Experimental
Input01 59.03 44.70Input02 75.74 55.68Input03 66.85 48.05Input04 59.52 44.11Input05 63.37 47.07
(ps=1, pb=10)1 2 3 4 5
0%10%20%30%40%50%60%70%80%90%
100%
exp_ps1theo_ps1
Experimental Results Freq Resolution Decision:
Bitrate reduction(%):Theoretical Experimental
Input01 30.08 16.29Input02 20.83 8.58Input03 14.11 6.80Input04 32.35 17.29Input05 28.11 15.35
1 2 3 4 50%
20%
40%
60%
80%
100%ps1_pb10
stride28stride5stride2stride1
1 2 3 4 50%
20%
40%
60%
80%
100%ps1_pb10_dm7_111
stride28stride5stride2stride1
(ps=1, pb=10) 1 2 3 4 50%
10%20%30%40%50%60%70%80%90%
100%
exp_ps1theo_ps1
Experimental Results Overall Bitrate reduction(%):
ps1 ps2 ps4
Input01 53.71 40.20 28.99Input02 59.49 56.66 55.97Input03 51.58 41.24 30.34Input04 53.77 42.47 29.96Input05 55.19 46.59 35.34
(pb=10)
Experimental ResultsInput01 (LS) Input02 (L) Input02(R)
Original
No smooth
Smooth with normalized error
Additional bit rate 0.005 kB/s 0.00498 kB/s 0.00498 kB/s