Signal processing apparatus concealing impulse noise by autoregressive modeling

摘要

A signal processing apparatus (100) comprising a noise detector (102) configured to: receive a stream of information representative of a stream of audio signal samples (112); and detect samples in the received stream of information (112) that are distorted by impulse noise in order to generate a noise detection signal (114), wherein the noise detection signal (114) also identifies preceding and succeeding samples that are undistorted. The signal processing apparatus (100) also comprises a processor (104) configured to replace distorted samples in the received stream of information with composite predicted values (426; 526) to provide a reconstructed stream of audio signal samples (116).

主权项

1. A signal processing apparatus comprising:
a noise detector configured to:
receive a stream of information representative of a stream of audio signal samples;detect samples in the received stream of information that are distorted by impulse noise in order to generate a noise detection signal, wherein the noise detection signal also identifies preceding and succeeding samples that are undistorted; and a processor configured to:
receive a stream of information representative of the stream of audio signal samples;receive the noise detection signal from the noise detector;determine, based on preceding and/or succeeding undistorted samples in the received stream of audio signal samples, auto-regressive model parameters;apply the auto-regressive model parameters to the preceding undistorted samples to determine forward linear predicted values of the distorted samples;apply the auto-regressive model parameters to the succeeding undistorted samples to determine backward linear predicted values of the distorted samples;combine the forward and backward linear predicted values by means of a window function to determine composite predicted values of the distorted samples;replace the distorted samples in the received stream of information with the composite predicted values to provide a reconstructed stream of audio signal samples; anddetermine the stability of the auto-regressive model parameters, and if the stability is determined to be below a predetermined threshold value then replace the estimated auto-regressive model parameters with previously determined stable auto-regressive model parameters for further processing.