Information Technology Reference
In-Depth Information
objective Metrics of Interactive VoIP.
We define two objective metrics
for characterizing the quality of multi-party VoIP [10]. These metrics can be
tailored to a two-party VoIP in a straightforward fashion.
1. The conversational efficiency (CE) measures the extension in time
to accomplish a VoIP conversation when there are communication
delays (Figure 2.2). Since a conversation over a network may be
charged according to its duration, the same conversation will cost
more for a network with lower CE.
Speaking Time ListeningTime
TotalTimeofCal
+
CE
=
(2.1)
l
Note that CE is identically perceived by all participants (see Table 2.1).
2. Conversational symmetry (CS). When a participant experiences
highly asymmetric response times in a conversation, he or she tends
to perceive degradation in the naturalness of the conversation because
it does not resemble a face-to-face conversation with uniform delays.
One possible effect is, when A perceives B to be responding slowly,
then A tends to respond slowly as well. To capture the asymmetry of
MSs perceived by A, we define
CS
A
to be the ratio of the maximum
and the minimum MSs experienced by A in a past window when the
conversation switches from
i
to
j
(
i
→
j
), and
j
is the speaker:
i
→
j
max
min
MS
MS
(
)
.
j
A
(( ))
Amp
=
S
=
2-partyVoIP
(2.2)
Ap
()
2
i
→
j
j
A
TaBLe 2.1
Research Issues in Interactive VOIP Systems
Section II: Evaluations of conversational quality
−Statistical model of subjective tests
−Trade-offs on objective metrics and JND
−Testbed for evaluating algorithms and systems
−Classifiers for learning and generalization
Section III: Cross-layer designs of speech codecs
−Codec design for dynamic encapsulation
−Rate-adaptive generation of enhancement layers
Section IV: Packet-stream-layer control algorithms
−Dynamic POS and LC control using classifiers
−Distributed equalization of MSs
Section V: Network-layer control algorithms
−Overlay networks and transport protocols
−Admissions control for multi-party VoIP
