Conversational Speech Quality - The Dominating Parameters in VoIP Systems

Speech quality, especially in voice over IP systems is influenced by many parameters: Delay has significant impact on the conversational quality mainly for two reasons: On the one hand the conversation between both subscribers and their interaction is more difficult. On the other hand the delay has a strong influence on the echo perception, longer transmission delay leads to an increased sensitivity in the users echo perception. It is important to evaluate the echo performance of a system in all conversational situations since typically additional artifacts like voice switching and background noise modulation may occur due to echo cancellation processes. Voice switching may be introduced at many stages in a VoIP scenario. Voice switching may lead to speech quality degradation perceived as missing syllables or missing words. Since again the effect is subjectively perceived different in single talk situations as compared to double talk situations both situations have to be taken into account. The transmission of background noise is a very critical parameter for the naturalness of a conversation. Due to clipping caused e.g. by VAD or the use of background noise reduction algorithms, background noise modulation-in combination with comfort noise injection-may degrade the perceived quality significantly. Packet loss may lead to speech quality degradation during single talk and double talk periods. The article describes objective methods which allow the assessment of various speech quality parameters relevant for the conversational quality. A. INTRODUCTION In modern IP systems the communication using speech can no longer be regarded as it used to be in traditional PSTN networks. The differentiation between terminals and network is no longer possible. Typically the same signal processing algorithms are used in networks and in terminals. Basically three different configurations have to be taken into account: IP gateway to IP gateway , IP terminal to PSTN terminal and IP terminal to IP terminal including the acoustical interfaces like hands-free phones, headsets or handset (see figure 1). Since the interaction of the signal processing between terminal and network highly influences the speech quality, typically a complete configuration has to be considered for all testing including network and traffic load simulations. The acoustical access is realized by acoustical components such as hands-free units, headsets or handset terminals. In any case sufficient echo control has to be provided in order to guarantee a sufficient echo loss even under worst case conditions. Due the influence of delay on echo perception (see[1]) these …