Naturalness preserving transform for missing frame compensation

Voice 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. Voice, however, requires real time, low delay transmission of information. Packet loss, packet delay, and packet errors all result in lost voice frames. At frame loss rates above 10%, these lost frames must be replaced or intelligible communication is not achieved. This paper describes the use of the Naturalness Preserving Transform (NPT) for reconstruction of missing voice frames. The algorithm was tested with compressed samples for three different waveform coders (PCM, DPCM, and μlaw) and the results of comparison testing with frame repetition are presented