- 660-Subjective assessment of time-varying speech quality

The last decade has seen a considerable evolution in the telecommunications sector with the development of real-time services on the Internet such as telephony. These services come up with several technical issues that have to be taken into account by operators. One of them is time-varying speech quality resulting from delays and random losses of packetized information during connection. In order to assess speech quality, typical methods described in recommendation ITU-T, P.800 use short stimuli (8 s) in subjective listening tests. These stimuli are speech samples impaired by the system under study. In the ACR method (Absolute Category Rating) which is a single stimulus assessment, listeners assess sample’s quality giving a mark on a five-category scale: 5 for “Excellent”, 4 for “Good”, 3 for “Fair”, 2 for “Poor“ and 1 for “Bad”. Subsequent analyses usually use averaged scores obtained from all listeners called MOS (for Mean Opinion Score). Although well suited for timeconstant speech quality, these methods do not take into consideration integration of non-steady quality over long periods. In this context, new protocols suitable to time-varying speech quality need to be developed. In this paper we present a few results obtained for timevarying speech quality with an appropriate methodology as follow.