Speech Quality in modern
Network Terminal Configurations
Network-Terminal
H. W. Gierlich
HEAD acoustics GmbH
Outline
• Overview of configurations and their impact
q
y
on quality
• Impact of
•
•
•
•
•
•
Delay
Coding
Noise and noise cancellation
Insufficient echo loss/improper echo cancellation
Cascading
g of signal
g
processing
p
g
Missing signaling information between endpoints
and network elements
• Summary
Network - Terminal
© HEAD acoustics GmbH 2
1
Terminal - Network
- typical elements found in modern connections
Network - Terminal
© HEAD acoustics GmbH 3
Delay aggregation (I)
tMT1
tMNA
tIPB
(10-50ms)
(≈ 80ms)
(10-400ms)
Network - Terminal
tPSTN1
tLDC
(1-10ms) (≈ 240ms, Satellite)
(≈ 20 ms, fiber)
© HEAD acoustics GmbH 4
2
Delay aggregation (II)
tPSTN2
tIPC
tPSTN3
tCT1
(1-10ms)
(10-400ms)
(1-10ms)
(13ms)
Network - Terminal
© HEAD acoustics GmbH 5
Delay aggregation (III)
toneway = tMT1 + tMNA + tIPB + tPSTN1 + tLDC + tPSTN2 + tIPC + tPSTN3
Best case
= (10 + 80 + 10 + 1 + 20 + 1 + 10 + 1 + 13)ms = 146ms
Worst case
= (50 + 80 + 400 + 10 + 240 + 10 + 400 + 10 + 13)ms = 1113 ms
Network - Terminal
© HEAD acoustics GmbH 6
3
E-model: Influence of delay
If no terminal impairment except TCLw:
146 ms/55 dB TCLw
R_
1113 ms/55 dB TCLw
87,9
MOS (4.5)
R_
4,2
146 ms/35 dB TCLw
R_
82,1
MOS (4.5)
26,6
MOS (4.5)
1113 ms/35 dB TCLw
R_
4,1
1,5
4,3
MOS (4.5)
Network - Terminal
1
© HEAD acoustics GmbH 7
ITU-T G.114: Effect of transmission time
100
Users
y satisfied
very
90
E-model rating R
Users
satisfied
80
Some users
dissatisfied
70
Many users
dissatisfied
60
Nearly all
users
dissatisfied
50
0
100
200
Network - Terminal
300
400
Mouth-to-ear-delay/ms
500
G.114_F01
© HEAD acoustics GmbH 8
4
Terminal - Network
- Coding and transcoding
AMR/EFR
G. 729
G. 726
G. 723
Network - Terminal
© HEAD acoustics GmbH 9
E-Model: The effect of codec tandeming
Type
Ie (G.113)
Coder 1
EFR (ACELP)
5
Coder 2
G. 729
G
(CS-ACELP)
10
Coder 3
G. 723 (ACELP, 5.3
kbit/s)
19
Coder 4
G. 726 (ADPCM,
32 kbit/s)
7
=> Ie(tot) = 5+10+19+7 = 46
R_
47,2
MOS (4.5)
2,43
Network - Terminal
© HEAD acoustics GmbH 10
5
Objective Testing based on ITU-T P.862 and others
Instrumental measures based on hearing models Perceptual models:
Modeling the Results of Auditory Tests by Comparison of Reference Speech
Signal with Processed Speech Signal
Results of
Listening Tests!
Typical Processing Steps (Schematic):
Speech
Signal
Test Object
processed
signal
Adaptation
reference
signal
Hearing Model
Comparison,
Reference,
Reduction,
Signal Value
Hearing
g Model
PSQM
PESQ
PSQM99
PAMS
TOSQA
P.861
P.862
KPN
BT
T-Systems
Network - Terminal
SQ index
PACE
ITU--TVQI
ITU
Ascom
Ericsson
P.OLQA
© HEAD acoustics GmbH 11
Example: Tandeming of Codecs
Vehicle
GSM
BT
Preamp.
A/D
MOSTHands-Free Algorithm
BUS
BT
chip
CVSD & EFR
CVSD & FR
Microphone & CVSD & FR
Microphone (speech & noise) & CVSD & FR
Network - Terminal
Mobile
Phone
GSM
Network
(.wav)
(.wav)
(.wav)(.wav)
(.wav)
© HEAD acoustics GmbH 13
6
Echo loss in IP – DECT configurations
•
•
•
Assuming… G.711 speech coder, no compression
Assumimg… no transcoding
Assuming no VAD, sufficient bandwidth
DECT
ISDN
Echo canceller “w/o
w/o echo
echo“,,
echo canceller with low level
echos… Echoes, clipping!
ISDN
DECT terminals, delay… .
Audible echoes!
GW
IAD
Coding, packetization,
buffer,… Delay! …Degrades
conversational quality!
DECT
EC in IAD!
IAD!... DECT delay
delay,
DECT distortions!
Packet loss and jitter under
network load, PLC, jitter
buffer… Artifacts, delay!
Network - Terminal
© HEAD acoustics GmbH 14
Echo Cancellation in IAD
Echo Performance
Si l d DECT echo
Simulated
h
Echo loss 36.7 dB
Spectrum avg. FFT Size:8192 Overlap:0,0% Rectangle
L/dB[V]
-20
-30
-40
EC
-50
td
-60
-70
t
-80
-90
-100
-110
100
200
500 f/Hz 1000
2000
NLP
5000
Echo loss 31 dB
td 30 ms
+ Network delay 100, 150, … ms !!!
Network - Terminal
© HEAD acoustics GmbH 15
7
DECT Echoes
td
DECT
td
td
linear
non-linear
Spectrum avg. FFT Size:8192 Overlap:0,0% Rectangle
100
200
500 f/Hz 1000
2000
L/dB[Pa]
-20
Spectrum avg. FFT Size:8192 Overlap:0,0% Rectangle
L/dB[Pa]
-20
Spectrum avg. FFT Size:8192 Overlap:0,0% Rectangle
L/dB[Pa]
-20
-30
-30
-30
-40
-40
-40
-50
-50
-50
-60
-60
-60
-70
-70
-70
-80
-80
-90
-90
-90
-100
-100
-100
5000
Echo loss 17.8 dB
100
200
500 f/Hz 1000
2000
-80
5000
100
Echo loss 31.0 dB
200
500 f/Hz 1000
2000
5000
Echo loss 36.3 dB
Network - Terminal
© HEAD acoustics GmbH 16
Tandeming of signal processing:
Car hands-free and network
Car audio
playback
system
Hands-free
Terminal
gain EQU
Microphone array
signal processing
gain
gain
EC
NR
Mobile
network
ES
EC
ES
NR
NLP, NR,
AGC,/gain …
Network - Terminal
NR
Audio playback
system in the
echo path of
( both ) ECs
(…both…)
• Information
exchange
between
gateway and
terminal not
possible
p
Tandemed NLP,
NR, AGC…
NLP, NR,
AGC,/gain …
© HEAD acoustics GmbH 17
8
Tandeming of noise cancellation
Audio
playback
system
Hands-free
Terminal
gain EQU
Gateway
gain
gain
• Tandeming of
noise
cancellation:
EC
EC
• Typical result
ES
NR
• Optimized
settings
ES
NR
NR
Microphone array
signal processing
AGC/
gain
AGC/
gain
Network - Terminal
© HEAD acoustics GmbH 18
Background noise transmission
Mobile
Terminal
Mobile
Network
Base
Station
gain
gain
EC
EC
CN
≈
ES
• Influence on
double talk
performance
• Both EC & ES
active
Network 1
ES
NR
NR
AGC/
gain
(.wav)
Network - Terminal
AGC/
gain
(.wav)
Network 2
© HEAD acoustics GmbH 19
9
Objective Testing based on ETSI EG 202 396-3
MOS for
• speech quality (S-MOS)
• noise transmission quality (N-MOS)
• overall quality (G-MOS)
Subjective MOS
Objective MOS
Noise
S
N
G
S
N
G
Car 1
4.7
4.7
4.6
4.8
4.8
4.8
Cafeteria
4.2
2.2
3.1
4.0
2.1
3.1
Office
4.0
2.8
3.5
4.2
2.6
3.5
Car 2
2.0
2.3
1.9
2.3
2.3
1.8
Road
1.4
1.3
1.0
1.3
1.0
1.0
Network - Terminal
© HEAD acoustics GmbH 20
Double Talk – Performance Limits
Characterization of HFT acc. to ITU-T P.340
MOS
≥ 4,0
4,0 – 3,5
3,5 – 3,0
3,0 – 2,5
2,5 – 2,0
≤ 2,0
TELRDT
[dB]
AHsdt [dB]
≥ 37
≥ 33
≥ 27
≥ 21
≥ 13
< 13
≤3
≤6
≤9
≤ 12
≤ 15
> 15
AHrdt [dB]
≤3
≤5
≤8
≤ 10
≤ 12
> 12
Type 2
Type 1
Type 3
TELRDT [dB]
≥ 37
Type 2a
≥ 33
AHsdt [dB]
≤3
≤6
≤9
≤ 12
> 12
AHrdt [dB]
≤3
≤5
≤8
≤ 10
> 10
Network - Terminal
Type 2b
≥ 27
Type 2c
≥ 21
< 21
© HEAD acoustics GmbH 23
10
Terminal test Setup
Network - Terminal
© HEAD acoustics GmbH 24
Background Noise Test
HMS II.3/6
3 PEQ & poweramps
≈
≈
≈
PEQ V
τ
τ
τ
≈
τ
PEQ V
τ
PEQ V
HAE-BGN
b
Subwoofer
Playback
≈
Radiotester/
BackgroundnoiseBackgroundnoisesimulation
HAE--BGN
HAE
HAE--Car
HAE
Network - Terminal
MFE VI
© HEAD acoustics GmbH 25
11
IP-Test Setup
HMS II.3,
HHP II.1,
HER IV.2,
HEL IV.2
IV 2
Network simulation & Monitoring
MFE VIX
(Network simulation & monitor)
SIP
PC
MFE VIII
(or equivalent IP gateway)
Bal. Out R(II)
( )
Bal. In L(I)
()
Mic In R(II)
USB
MFE VI.1
Bal. Out L(I)
Network - Terminal
© HEAD acoustics GmbH 26
Summary
• Potential problems in a connection may occur due to
tandemed signal processing in terminals and networks
• Quality evaluation needs to involve all components of a
connection and their interaction!
• Currently no signalling is provided between terminals
and networks for information exchange about signal
processing active in the different devices
• Related work:
• G.mdcspne started in SG16 (establish the best arbitration among
Network Signal Processing Equipments and terminal signal
processing)
• G.115.x sub series (Logic and protocols for the control of signal
processing network elements and functions) may potentially be
enhanced to support this capability
Network - Terminal
© HEAD acoustics GmbH 27
12