Prediction of Background Noise Degradations in Fullband Speech Communication Scenarios

Authors: Sebastian Möller, Andreas Bütow

Abstract:

In this paper, which is mostly based on the Master thesis of the second author [2], we analyze the impact of background noise at the sending and receiving side on perceived speech quality in fullband (20-20,000 Hz) speech communication scenarios. Whereas the effect of noise has been a research topic in narrowband (300-3,400 Hz) speech telephony for a long time, experimental data addressing wider speech transmission bandwidths are still scarce. In order to fill this gap, a listening-only experiment has been carried out in which background noise levels were carefully controlled at both sending and receiving side. Three types of noise have been considered, including speech-like babble as well as white and pink noise. Overall speech quality was assessed by 25 normal-hearing participants following ITU-T Rec. P.800. The results were analyzed regarding the effect of noise level and noise type on speech quality, revealing significant impacts of both factors. More precisely, babble noise showed a much smaller impact on overall quality than white and pink noise. Finally, the results were used to augment a standardized network planning model, the fullband E-model described in ITU-T Rec. G.107.2. The augmented model showed a very good correlation to the average test results, for all noise levels at sending and receiving side, while still neglecting the type of noise which might not be known during network planning applications anyhow.