Monitoring, identification, and selection of audio signal poles with characteristic behaviors, for separation and synthesis of signal contributions

Abstract

A system for processing a signal representing acoustical information performs a linear predictive coding (LPC) analysis and segments the signal into music, speech and noise components (including channel noise and acoustic artifacts) in accordance with behavior, over time, of the poles describing the signal, resulting from the LPC analysis. Poles exhibiting behavior characteristic of speech, music and channel noise of interest may then be selected while other poles representing random noise or information which is not of interest are suppressed. A "cleaned" signal can then be synthesized, with or without additional pre-processing to further suppress unwanted components of the signal. Additionally or alternatively, tags can be applied to frames or groups of frames of the original signal to control application of decoding procedures or speech recognition algorithms. Alternatively, the synthesized "cleaned" signal may be used as an input to a vector quantizer for training of codebooks and channel assignments for optimal processing of the original signal.