OPTICOM OPERA - V 3.5 User Manual page 148

Loudness
Measurement
VAD Parameters,
Front End Clipping
(FEC) and Hold
Over Time (HOT)
C H A P T E R 6 : T E L E P H O N Y
T E S T I N G
The loudness is a more psycho-acoustic view of the signal levels. It expresses
not only how much energy is contained in a signal, but also how loud it is
perceived by the listener. Of course PESQ can not know the characteristics of
the telephone used, since it operates on the electrical level only, but it assumes
a fixed relation between the binary signal level in the input files, and the
loudness. It blocks the out of band energy, filters the input signals and calculates
the overall energy. It is then assumed that this overall energy – still in dB
corresponds to 79dB
SPL
for the assessment of equipment exhibiting noise substitution, PESQ provides
the loudness of the entire reference signal, the test signal and the silent intervals
of the test signal separately. The actual loudness calculation is performed by the
PESQ perceptual model as defined in P.862, taking into account the above
relation between dB
ov
Most VoIP systems use VAD (Voice Activity Detection) to save bandwidth. If the
VAD indicates active speech, then the encoder transmits packets containing
speech. If the VAD decides on silence at the input, the encoder simply informs
the decoder of the characteristics of the noise at the encoders input and this
noise is then substituted by the decoder. This requires significantly less
bandwidth, than transmitting the entire speech signal. The two most common
problems of VADs are because they must meet severe realtime constraints.
Once the VADs decide on active speech, they cannot tell the encoder a few
milliseconds later that in fact their decision was wrong and that the encoder had
to transmit the signals differently. It is imply too late and the signal is already
transmitted. This false or slow detection of active speech and silence is
characterised by two parameters, Front End Clipping (FEC) and Hold Over
Time (HOT). The Hold Over Time is frequently also called Hang Over Time.
Both parameters are expressed in ms and specify the time between the actual
start of the active speech sequence until the VAD decides on speech (FEC),
respectively the time after active speech ended and the VAD decided on silence
(HOT). Opera gives the average HOT and FEC as well as HOT and FEC on a
per utterance basis.
Measuring FEC and HOT in real networks is far not a trivial problem. Currently
there are several methods available to assess these parameters. Some of these
will be explained in the following. All methods have in common that they are
not perfect and that they may fail under some circumstances. Especially if other
network effects, like varying delay due to jitter buffer adjustments or packet loss
occur at the same time as FEC or HOT. Under such circumstances it is almost
impossible to determine the correct result, since the situation is ambiguous.
Detection of FEC and HOT is much easier if special test signals are used in stead
of real voice, however the situation for the device under test is also less realistic.
Of course, artificial signals with algorithms that were designed to work for real
voice can always be used, but not vice versa.
Noise Burst and low Level Sine Tone
The first method uses a noise burst, together with a low level sine tone ("dyer
tone"). At the beginning of the test sequence both signals are overlaid. The
noise is switched off after a well defined time and the sine tone continues. Since
the duration of the noise burst and the frequency of the sine tone are known, it
can be assumed that the difference between the duration of the received noise
pulse minus the duration of the original pulse is the duration of the Front End
B A N D
at the ear reference point according to P.830. To allow
and dB .
SPL
142
V O I C E Q U A L I T Y
–
ov