Information Technology Reference
In-Depth Information
CS
A
is approximately one in a face-to-face conversation, but increases as the
round-trip delay increases. In the two-party case, the minimum MS is always
experienced by the speaker and the maximum experienced by the listener.
Since there are multiple listeners in the multi-party case and the majority of
the clients are passive listeners, it is important to identify the asymmetry for
the passive listeners alone. Hence, we choose to eliminate the current speaker
when evaluating the minimum MS in the multi-party case. Note that
CS
and
CE
are counteracting: as
CS
improves,
CE
degrades.
In VoIP, a user does not have an absolute perception of MEDs because he or
she does not know when the other person will start talking or who will speak
next (in multi-party VoIP). However, by perceiving the indirect effects of MED,
such as MS and CE, the participants can deduce the existence of MED.
Table 2.1 summarizes the four research issues in the design of interactive
VoIP systems. These are related to the classification and generalization of
subjective test results and the design of network control and coding algo-
rithms. For each of these issues, we present some existing work and our
approaches in Sections 2.2-2.5. Lastly, Section 2.6 concludes the chapter.
2.2 Evaluating Conversational Quality
In this section, we first survey existing metrics for measuring conversational
quality. We then present results on evaluating subjective quality of VoIP
systems and methods for learning the mapping from objective metrics to
control algorithms that optimize subjective quality.
2.2.1 Previous Work
Effects of MED on Conversational Quality.
Subjective tests by Brady [9] and
Richards [6] in the 1970s have led to the conclusion that MED affects the user
perception of conversational quality, and that longer MEDs increase the dis-
satisfaction rate. However, their conclusions are limited when used for eval-
uating VoIP systems, since only a few constant delays were experimented.
Subjective tests by Kiatawaki and Itoh [7] at NTT show that one-way delays are
detectable, with a detectability threshold of 100-700 ms for a trained crew and
of 350-1,100 ms for untrained subjects. The International Telecommunication
Union (ITU) G.114 [11] prescribes that a one-way delay of less than 150 ms is
desirable in voice communication, and that a delay of more than 400 ms is
unacceptable. Without specifying the trade-offs with LOSQ, MED alone is not
adequate for evaluating VoIP.
objective Measures on Conversational Quality.
The ITU has several rec-
ommendations for the objective and subjective evaluations of the end-to-end
quality of a voice transmission system. Table 2.2 shows the naming standard
