Conclusions

In this Chapter, we showed who to use neural networks to learn how humans evaluate the perceived quality of degraded speech signals transmitted over a packet network. We toke into consideration the effect of language, codec type, packetization interval, loss rate, and loss distribution. We showed that the trained neural network reproduces accurately the subjective evaluations. It is also capable of evaluating subjective quality in presence of new sets of parameters' values. In this way, one can automatically measure in real time and with good confidence the subjective speech quality. In order to build the database used for training and testing the neural network, we carried out a series of field tests. This helped us in the choice of the mentioned parameters and their ranges. ChapterChapter