Auditory assessment of conversational speech quality of traditional and spatialized teleconferences

We present a new auditory test methodology to assess the conversational speech quality of audio conferences. The method was developed for tests with three participants that are seated at different remote locations. The paper presents the development of the test scenarios that are intended to stimulate the participants' communication in a controlled way. The overall test methodology is outlined. We have applied the method in a conversation test to measure the quality impact of different bandwidths (narrowband, 300-3400 Hz; wideband, 50- 7000 Hz; fullband, 0-22050 Hz), spatial versus non-spatial rendering over headphones, and channels with and without talker echo. The primary goal of the conversation test was to investigate whether the newly developed test scenarios lead to equally long test conversations, and whether each scenario yields a good balance in terms of the speech activity of the three interlocutors. The advantages of spatial audio conferencing in terms of speaker recognition, focal assurance (remember who said what in a teleconference) and user preference have already been addressed in a listening-only context. In addition to characterizing the conferencing scenarios, we now studied the users' preferences and ability to discriminate the quality of different audio conferencing settings in a conversational context.