Method and system for processing the pre-echoes of an audio-digital signal coded by frequency transformation

A system for processing the pre-echoes of an audio-digital signal, said audio-digital signal comprising blocks comprising N successive samples, said blocks being transformed by a frequency transform into transform coefficients y(k) over weighting windows formed over said blocks, said system comprising:

• a coder including a plurality of modules arranged in cascade, said coder comprising:
• a module for reading said samples x(n) of said audio-digital signal,
• a module for receiving said read samples, for calculating said frequency transform and for outputting a corresponding transform coefficient y(k) for each current sample block,
• a module for receiving said read samples, for detecting a transition in amplitude of said audio-digital signal at the level of at least one sample of rank (N-q) of said current sample block and for outputting a transition indicator and a position of said transition in said current block,
• a module for receiving said read samples, and for extracting, from said samples x(n), parameters of energy .sigma..sup.2 x(n) of each sample or of a group of successive samples of said current sample block of said audio-digital signal,
• a module for coding by quantizing said transform coefficients, and
• a module for coding said parameters of energy .sigma..sup.2 x(n), and said position of said transition;
• a digital transmission channel transmitting said transition indicator and coded values of said transform coefficients, said parameters of energy .sigma..sup.2 x(n), and said position of said transition;
• and
• a decoder including a plurality of modules arranged in cascade for receiving said coded values and said transition indicator from said transmission channel, said decoder comprising:
• a module for decoding said transmitted transform coefficients y(k),
• a module for calculating the inverse frequency transform of decoded transform coefficients, and for outputting a noisy decoded signal x'(n),
• a module for reconstructing said transmitted coded parameters of energy .sigma..sup.2 x(n) and position of said transition,
• a module for receiving said noisy decoded signal x'(n) and said reconstructed parameters of energy .sigma..sup.2 x(n), for extracting prediction coefficients of said samples x(n) of the block of samples x(n-1) preceding said current sample block and for evaluating the energies of the innovation noise .sigma..sup.2 w(n) and of the coding noise .sigma..sup.2 b(n) for said sample x(n) of rank n of said current sample block, and
• a KALMAN filtering module for receiving as a filtering input said noisy decoded signal x'(n) and as a control input said prediction coefficients and said energies of said coding noise .sigma..sup.2 b(n) and said innovation noise .sigma..sup.2 w(n), and for outputting a filtered decoded signal xr(n), thereby reducing the resulting additive noise when coding the quantization of said transform coefficients.