On the Scalability of Audio Quality for Networked Multimedia Environments

As a result of listening tests performed with users of multimedia applications we present constraints for a dynamic adaptation (scaling) of audio data streams to the available system and network resources. For the speech medium any severe deformation of frequency characteristics should be avoided, while shortening and expanding of speech pauses is suitable to cope with varying delays. For the music medium the best scaling algorithm is a smooth fade between quality levels. Scaling resulting in lower quality should be performed in multiple steps, while in scaling towards higher quality any approach can be chosen. It appears that six levels of quality are sufficient for scaling of music from high quality (CD) to very low quality (cellular phone) and vice versa. In our proposal of a Dynamic QoS Centered Architecture, where the bitrate and scheduling of associated media streams are adapted dynamically, the highest possible QoS for the user has to be taken into account.