Abstract:
The Bluetooth Special Interest Group (SIG) has standardized the subband coding (SBC) audio codec to connect headphones via wireless Bluetooth links. SBC compresses audio at high fidelity while having an ultra-low algorithm delay. To make SBC suitable for the Internet, we extend it by using a time and packet loss concealment (PLC) algorithm that is based on ITU's G.711 Appendix I. The design is novel in the aspect of the interface between codec and speech receiver. We developed a new approach on how to distribute the functionality of a speech receiver between codec and application. Our approach leads to easier implementations of high quality VoIP applications.
We conducted subjective and objective listening tests of the audio quality of SBC and PLC in order to determine an optimal coding mode and the trade-off between coding mode and packet loss rate. More precisely, we conducted MUSHRA listening tests for selected sample items. These tests results are then compared with the results of multiple objective assessment algorithms (ITU P.862 PESQ, ITU BS.1387-1 PEAQ, Creusere's algorithm). We found out that a combination of the PEAQ basic and advanced values best matches---after third order linear regression---the subjective MUSHRA results . The linear regression has coefficient of determination of R²=0.907². By comparison, our individual human ratings show a correlation of about R=0.9 compared to our averaged human rating results.
Using the combination of both PEAQ algorithms, we calculate hundred thousands of objective audio quality ratings varying audio content and algorithmic parameters of SBC and PLC. The results show which set of parameters value are best suitable for a bandwidth and delay constrained link. The transmission quality of SBC is enhanced significantly by selecting optimal encoding parameters as compared to the default parameter sets given in the standard.
Finally, we present preliminary objective tests results on the comparison of the audio codecs SBC, CELT, APT-X and ULD coding speech and audio transmission. They all allow a mono and stereo transmission of music at ultra-low coding delays (<10ms), which is especially useful for distributed ensemble performances over the Internet.