Speaker dependent speech compression for low bandwidth communication

Abstract

Standard Object Systems has developed novel signal processing method for the compression and decompression of speech signals. The approach provides a dramatic reduction in communication bandwidth requirements for speech data by using a two phase process. In the first phase, a speaker dependent voice information file is created to model the speaker. This file is used in the second phase to both compress the speaker's speech at the transmitter and decompress the speaker's speech at the receiver. This speech compression has demonstrated communications quality speech at bit rates as low as 500 bits per second. Current aerospace applications employ standards such as multi band excitation as used on INMARSAT-M (MBE) at 4,150 BPS, as code excited linear prediction (CELP) for the U.S. DoD 4800 BPS STU III system, or the North American digital cellular standard using vector sum excited LPC (VSELP) at 8000 BPS.