Compensation for missing voice and image coding frames in packet switched networks

Real-time communication through packet-switched networks, such as the Internet, is increasing in popularity. These networks were designed for bursty computer data with no stringent timing requirements. Real-time applications, however, require low delay transmission of information. Iterative postprocessing algorithms are presented to combat this problem for voice and image data, respectively. This paper describes the use of the naturalness preserving transform for data reconstruction assuming a packet erasure model. Algorithm complexity and qualitative measures are also noted. Both speech and image quality after reconstruction are shown to be significantly improved.