The need for low-power multimedia processing is integral to portable and embedded devices such as cell phones, wireless terminals, multimedia handhelds and PDAs. The multimedia processing in these devices is often done in software on programmable processors under the control of real-time operating systems (RTOSs). The paper describes a novel approach for power-aware scheduling of multimedia tasks in RTOSs that trades off computation deadlines against power consumption. Our proposed technique is based on exploiting (a) the inherent tolerance of many real-time multimedia systems to lost data samples due to factors such as communication noise and network congestion, and (b) the ability of many new processors to operate under dynamically changing voltage and frequency. We use history-based prediction of multimedia sample processing time to aggressively adapt the processor voltage and frequency for significantly reduced power consumption, at the cost of missed task deadlines that are handled by adaptive multimedia applications as a form of noise leading to data loss. Simulation results showing the effectiveness of our technique are presented.