Parameter Selection

It is important to correctly identify and select the quality-affecting parameters. If an important parameter is neglected, the accuracy of the tool will be degraded and it will not behave as expected. However, the number of parameters to be taken into account should be minimized. This is because this number determine the minimum number of samples (see Sec. 4.3) in the quality database to train and test the NN, knowing that this database is built by subjective quality tests, which are very expensive and hard to carry out. Hence some mechanisms may be used to merge two or more parameters in one. For example, the case of delay variation (jitter) can be mapped to be included in the loss parameter by using dejittering buffer. We can classify the quality-affecting parameters into three categories: $\bullet$

• Network parameters: Parameters concerning the packet network that carrying media-type signals. The most known parameters are packet loss rate, arrival jitter, packetization interval, loss distribution, and end-to-end delay. Furthermore, effects of error concealment techniques (FEC or ARQ) (e.g. silence, noise, repetition, and waveform substitution for lost audio packets) [38] can affect the quality.
• Encoding parameters: Parameters concerning the ability to encode or compress the original media-type signals to fit the output stream into certain bit rate. There are two types of compression (lossless and lossy), however, the only one that affects the quality is the lossy one. The majority of the existing codecs use both types. Hence some of the parameters due to the encoding impairments can be classified into the type of the codec used, the sampling rate, and the number of bits per sample, the number of layers in the case of layered codecs. For video, we have in addition the bit rate, the frame rate, the encoded frame types, etc.
• Other parameters: like the echo (which may occur due to the long end-to-end delay), crosstalk effect (when two or more people start talking at the same time), or the number of participating sources. These effects occur in bi-directional sessions or multi-party conferences. In addition, there is the audio-video synchronization aspect for the case of multimedia applications [40].

There is also the aspect of the session types. Actually, session-types may evoke additional parameters to be taken into account. Two-way sessions present interactivity between the end-users. In the case of speech, echo, end-to-end delay, crosstalk effects may affect the quality. These factors do not affect the quality for one-way sessions. One-way sessions can be equivalent to bi-directional sessions under the following hypothesis: when there is a mechanism to control the echo or if we assume ideal echo suppression. The crosstalk effect can be solved for the case of half-duplex conversions. The delay jitter effect can be reduced by implementing a dejittering mechanism [37]. These parameters may not affect the quality for some other cases of the media-types. For example, for pure video transmission in real time, even for the two-way sessions as the videoconferencing applications, it is easy to map them into two one-way sessions. As there is no echo, nor crosstalk problems. However, the end-to-end delay should have to be parameterized (considered as a quality-affecting parameter). In addition, if our method is used to assess the quality of multimedia (audio and video) applications, we recommend to use two separate modules to assess the quality for each one. An extra module should be used to assess the overall quality given the measures of the other two modules plus the value of the synchronization parameter. In this case, during the subjective quality tests, users are asked to assess the quality of each media type and the overall quality.