Voice quality constitutes the key element of the customer’s feeling and trust towards his telecom operator. Consequently, it is essential that France Telecom possesses means of voice quality assessment as it is a vital aspect for VoIP service provision. Quality should be supervised at the end-point where it presents the most faithful reflection of customer’s perception. However, most of the existing methods seem inadequate for this use. Even though the signal-based measuring methods evaluate well end-to-end quality
Voice quality constitutes the key element of the customer’s feeling and trust towards his telecom operator. Consequently, it is essential that France Telecom possesses means of voice quality assessment as it is a vital aspect for VoIP service provision. Quality should be supervised at the end-point where it presents the most faithful reflection of customer’s perception. However, most of the existing methods seem inadequate for this use. Even though the signal-based measuring methods evaluate well end-to-end quality, their implementation in customer gateways is impossible, mainly because of lack of resources. For the purpose of supervision it is recommended to use parametrical models, such as the E model (ITU-T G.107) or models compliant with ITU-T P.564, which do not require neither complex signal processing nor a reference signal. For practical reasons, France Telecom chose to work on the E-model instead of developing a totally new model.
Nevertheless, our experience regarding the use of the E model in end-point voice quality assessing probes has demonstrated some substantial drawbacks of the model. According to several published tests, the model judges too severely the degradations due to packet loss, jitter and delay, and therefore underpredicts quality. On the other hand, the information about packet loss, jitter and delay is often the only available in an end-point. Therefore, in the first place it is important to construct a new model which will evaluate better the influence of these three factors on quality. However, if time allows we do not exclude the possibility of including other parameters, such as noise or echo.
The most important part of the study will concern the study of the influence of packet loss on voice quality for the purposes of real-time measurement. The impact of certain parameters such as delay or noise has already been studied by France Telecom and elsewhere. It is only required to integrate them into one unique quality evaluation model. We would also like, that the model took into account dynamic phenomena such as recency effect. At the end, the model should evaluate quality score in narrow band (300-3400 Hz) as well as wide band (50-7000 Hz) on a single, mixed, quality scale.
In order to construct the model the creation of new subjective databases will be necessary. Some work can be based on objective tests (PESQ), using the test chain which has been developed in our team.
A validation of the performance of the model (formally at the end of the development, but also probably on intermediate versions) is necessary. In particular, the use of the method described in ITU-T P.564 will be deeply appreciated.
Applicant profile (not all are required)
C and Matlab programming
Data analysis, statistical methods
Mathematics
Understanding of human perception and judgment, especially of psychophysics
QoS, voice quality
VoIP comprehension (in particular of voice flows: protocols, codecs, jitter buffer, aspects related to packet loss, jitter and delay)
IP comprehension
Experience in neural networks is considered as an optional plus
contact
Please send applications to:
Anna CZUCZMAN
France Telecom – Division R&D
TECH/SSTP – Lannion
anna.czuczman [a-t] orange-ftgroup.com
(+33) 2 96 05 22 90
Read more...
