Method and apparatus for dynamically adapting a vector-quantizing coder codebook

Claim 2

In a vector quantizing coder for coding a sequence of input signal samples into a codebook address pointing at a prestored codeword selected for providing the lowest mean square error with said sequence, a method for dynamically updating said codebook during coding, said method including: splitting the codebook into a fixed content portion and a fixed length adaptive content portion; updating said codebook by shifting into said fixed length adaptive content portion each selected codeword, whereby a previously stored codeword is shifted out of said fixed length adaptive content portion.

Claim 3

A method according to claim 1 wherein said input signal samples are a voice signal.

Claim 4

A method according to claim 2 wherein said voice signal is coded during a conversation including periods of voice activity, "talkspurts", separated by long pauses of predetermined length, said method being characterized in that said fixed length adaptive portion of said codebook is reset to its original contents at the beginning of each talkspurt identified through conventional voice activity detection operations.

Claim 5

A method according to claim 3, further including a fast start-up step including: detecting the first sequence of input signal samples of each said talkspurt; diverting said first sequence from the vector quantizing operation; coding said first sequence using block companded, pulse code modulated (BCPCM) techniques; and using said BCPCM coded sequence for the initial step of each talkspurt codebook updating.

Claim 6

A Code Excited Linear Prediction/Long-Term Prediction (CELP/LTP) speech coding device including adaptive filtering means for filtering a speech signal s(n) and deriving therefrom a residual signal r(n); means for generating a predicted residual signal rp(n); means for subtracting said predicted residual signal from said residual signal and deriving therefrom an error residual signal e(n); vector quantizing means connected to said subtracting means, said vector quantizing means including a codebook initially loaded with predefined codewords, and search means for converting sequences of consecutive e(n) samples into a codeword address, said codeword being selected within said codebook for providing the lowest mean square error with respect to said sequence, and a gain factor; a codeword coder for encoding said selected codeword, a codeword decoder connected to said codeword coder for providing a decoded error signal e'(n); a long-term prediction (LTP) feedback loop connected to said codeword decoder for providing a decoded residual signal r'(n) and said rp(n) signal; said vector quantizing means being characterized in that it includes: means for splitting said codebook codewords into a fixed content codebook portion and a fixed length adaptive contents codebook portion; normalizing means connected to said codeword decoder and to said LTP feedback loop for generating normalized selected codewords CB(i) wherein: CB(i)=NORM (a·e'(i)+(1-a)·r'(i)) wherein NORM stands for a normalization operation according to: ##EQU13## wherein a<1 represents a preselected weighting coefficient and SQRT represents the square root function, and C is a normalizing predefined fixed value. loading means connected to said normalizing means for shifting said normalized selected codeword into said fixed length adaptive contents codebook portion, whereby a previously stored codeword therein is shifted out therefrom.

Claim 7

A Code Excited Linear Prediction/Long-Term Prediction (CELP/LTP) speech coding device including adaptive filtering means for filtering a speech signal s(n) and deriving therefrom a residual signal r(n); means for generating a predicted residual signal rp(n); means for subtracting said predicted residual signal from said residual signal and deriving therefrom an error residual signal e(n); vector quantizing means connected to said subtracting means, said vector quantizing means including a codebook initially loaded with predefined codewords, and search means for converting sequences of consecutive e(n) samples into a codeword address, said codeword being selected within said codebook for providing the lowest mean square error with respect to said sequence, and a gain factor; a codeword coder for encoding said selected codeword, a codeword decoder connected to said codeword coder for providing a decoded error signal e'(n); a long-term prediction (LTP) feedback loop connected to said codeword decoder for providing a decoded residual signal r'(n) and said rp(n) signal; said vector quantizing means being characterized in that it includes: means for splitting said codebook codewords into a fixed content codebook portion and a fixed length adaptive contents codebook portion; normalizing means connected to said codeword decoder and to said LTP feedback loop for generating normalized selected codewords CB(i) wherein: CB(i)=NORM (a·e'(i)+(1-a)·r'(i)) wherein NORM stands for a normalization operation according to: ##EQU13## wherein a<1 represents a preselected weighting coefficient and SQRT represents the square root function, and C is a normalizing predefined fixed value. loading means connected to said normalizing means for shifting said normalized selected codeword into said fixed length adaptive contents codebook portion, whereby a previously stored codeword therein is shifted out therefrom.

Claim 8

A speech coding device according to claim 5 further including: a voice activity detector (VAD) sensitive to said speech signal S(n) for detecting the occurrence of beginning and end of any silences longer than a predefined duration, said silences delimiting so-called talkspurts, and for generating a VAD flag at the beginning of each said talkspurt; resetting means sensitive to said VAD flag for resetting said fixed length adaptive contents codebook portion to its initial contents.

Claim 9

A speech coding device according to claim 6 further including: Block companding PCM (BCPCM) coding and decoding means; delay means fed with said residual signal r(n) and sensitive to said VAD flag for diverting the first sequence of speech signal samples to be encoded at the beginning of each talkspurt from said CELP/LTP coding device into said BCPCM coding and decoding means; means for feeding a decoded sequence of samples from said BCPCM coding and decoding means into said fixed length adaptive contents codebook portion.