A Methodology for Deriving VoIP Equipment Impairment Factors for a Mixed NB/WB Context
1. Wireless Access Research
Adil Raja August 2007
A Methodology for Deriving VoIP
Equipment Impairment Factors for a
Mixed NB/WB Context
Adil Raja
13th
August, 2007
2. Wireless Access Research
Adil Raja August 2007
VoIP- The Shift to WB Paradigm
• The NB and WB would have to coexist for
some time.
i.e. operate either in cascade or in parallel.
• This work proposes an extension to E-Model
for the parallel case.
3. Wireless Access Research
Adil Raja August 2007
The E-Model
• An Instrumental Model initially designed for
NB handset telephony.
• R= R0-Is-Id-Ie,eff+A
• R0=93.2 – the basic SNR.
• Is - simultaneous impairments e.g. OLR, non-
optimum sidetone.
• Id - e.g e2e delay and echo.
• Ie,eff Impairments due to low-bitrate codecs and
packet loss.
• A – Advantage factor.
5. Wireless Access Research
Adil Raja August 2007
Moller’s WB extension to E-Model
• Consider Two experiments
• Each experiment is composed of two sets of speech
samples.
• MOS evaluation is done based on ACR paradigm.
• One set is evaluated in the NB/WB context.
• The other set is evaluated in the NB context, by
downsampling the WB coded speech samples.
6. Wireless Access Research
Adil Raja August 2007
Example
Speech quality evaluation on narrowband
conditions only
1
2
3
4
5
1
SubjectiveMOS
C1 C2
C3 C4
C5 C6
C7 C8
C9 C10
C11 C12
C13 C14
C15 C16
C17 C18
C19 C20
C21 Reference
Speech quality evaluation for narrowband and
wideband conditions
1
2
3
4
5
1 2
SubjectiveMOS
C1 C2
C3 C4
C5 C6
C7 C8
C9 C10
C11 C12
C13 C14
C15 C16
C17 C18
C19 C20
C21 Reference
7. Wireless Access Research
Adil Raja August 2007
Example
Relationship between MOS for narrowband conditions
and MOS for wideband conditions
1
2
3
4
5
1 2 3 4 5
Subjective NMOS
SubjectiveMMOS
STUDY GROUP 12 – DELAYED
CONTRIBUTION 46
Discussion on unified objective
methodologies for the comparison
of voice quality of narrowband and
wideband scenarios
8. Wireless Access Research
Adil Raja August 2007
The Extension
• Rnb=a+(eRnb/wb/b
-1)
• a=169.38
• b=176.32
• The R-scale is stretched by 29%
• A linear extension with somewhat similar
results exists.
9. Wireless Access Research
Adil Raja August 2007
R-scale extension with PESQ
• Two sets of speech samples are created.
• Both contain stimuli coded by NB and WB
codecs. (124 samples each).
• The first (NB/WB) set is evaluated using
PESQ-WB.
• The NB coded stimuli were upsampled to
16 kHz.
10. Wireless Access Research
Adil Raja August 2007
R-scale extension with PESQ
• How to evaluate the 2nd
part of the test?
• For the 2nd
(NB) part of test three options
were considered for evaluating the
samples.
• In Option 1 All the files (ref|deg of NB|
WB) were downsampled to 8 kHz.
• This qualifies as a NB test.
11. Wireless Access Research
Adil Raja August 2007
R-scale extension … Option 1
RNB=a+b*RNB/WB
Max RNB=100.48
Suggests no extension.
13. Wireless Access Research
Adil Raja August 2007
R-scale extension … Option 2
All Evaluation is done by
PESQ-WB.
Contracts the scale
instead.
14. Wireless Access Research
Adil Raja August 2007
R-scale extension … Option 3
• Same as Option 2 but the Ref samples for
NB stimuli are downsampled/upsampled.
• This would clearly tantamount to repeating
option 1 but with PESQ-WB
15. Wireless Access Research
Adil Raja August 2007
An Alternative
Using Supp 23 Exp 1&3.
Mapping between PESQ-
WB & -NB evaluted NB
coded samples.
Max RNB=107.1786
16. Wireless Access Research
Adil Raja August 2007
The Codecs
• G.729
• AMR-NB – 7.4 & 12.2 kbps
• G.723.1 – 6.3 kbps
• G.722.1 24/32 kbps
• G.722.2 – all 9 modes
17. Wireless Access Research
Adil Raja August 2007
Other metrics
• MLR - [0,2.5,. . . , 15, 20, . . . , 40]%
• CLP - 10, 50, 60, 70 and 80%.
• PI – 10-60 ms.
• A total of 2,820 conditions were simulated
• A given condition was applied to a file for 30 times; to
negate the effect of loss location.
• Thus a total of 3,38,400 speech files were created.
• Evaluation was done on beowulf using PESQ-WB