Wideband Stereo Speech Coding for Teleconferencing Applications

Almost a ll voice communications are based on monophonic narrowband speech. Wideband stereophonic communications provide a more natural sounding environment. This holds especially true for teleconferencing applications, where the localization information in the stereo signal adds a new dimension to the communication. As of today, there is no standardized speech codec with full stereo support. Statistical analysis shows that there exists correlation between left and right channels of a stereo speech signal. Instead of coding both channels independently (2 x mono bit-rate) stereo parameters are extracted which, combined with the mono signal, allows for low bit rate stereo reproduction. This thesis report describes a parametric stereo coding algorithm. The algorithm, which is targeted at conversational applications such as teleconferencing, reproduces the stereo signal from a down-mixed mono signal and additional stereo parameters. During the thesis work, different stereo coding methods have been studied and evaluated. Given design constraints such as low complexity, low delay and low bit rate, the stereo coding algorithm was based on adaptive inter-channel prediction techniques. In the thesis, novel solutions to inherent problems of the inter-channel prediction framework are presented. A parametric stereo extension, based on the new findings, has been implemented into the standardized Adaptive Multirate Wideband speech codec. Although the bit-rate cost for the stereo extensions is as low as 1 kbps, MUSHRA tests show that the new codec is able to reproduce a high-quality stereo speech signal.