Method for voice signal coding and/or decoding by means of a long term prediction and a multipulse excitation signal

• US 6,246,979 B1
• Filed: 03/23/2000
• Issued: 06/12/2001
• Est. Priority Date: 07/10/1997
• Status: Expired due to Term

• Current Assignee: Grundig Multimedia B.V. (Arcelik A.S.)
• Original Assignee: Grundig AG (Arcelik A.S.)
• Inventors: Carl, Holger
• Primary Examiner(s): Korzuch, William R.
• Assistant Examiner(s): Azad, Abul K.
• Application Number:
  US09/446,613
• Time in Patent Office: 446 Days
• Field of Search: 704/212, 704/216, 704/217, 704/218, 704/219, 704/222, 704/223, 704/230, 704/221
• US Class Current: 704/219
• CPC Class Codes: G06T 9/004 Predictors, e.g. intraframe, interframe coding
• CPC Class Codes: G10L 19/06 Determination or coding of the spectral characteristics, e.g. of the short-term prediction coefficients
• CPC Class Codes: G10L 19/10 the excitation function being a multipulse excitation
• CPC Class Codes: G10L 2019/0011 Long term prediction filters, i.e. pitch estimation

Abstract

This invention relates to a method for voice signal coding and/or decoding. According to this method, a voice signal analysis for determining the prediction parameters is carried out from a digital voice signal. An excitation signal component (ENtp) is determined from an adaptive code book (3) built from a delayed integral excitation signal (Ev). Further, a multipulse component of the excitation signal is determined (4) by minimising the effect of the weighting filtered difference between the signal resulting from the respective excitation signal and the input voice signal.

Claims

1. A method of encoding and decoding voice signals, the method comprising the steps of:

a) analyzing a digital voice signal for determination of prediction parameters,

b) determining an excitation signal component from an adaptive code book, the adaptive code book being formed from a delayed total excitation signal, said adaptive code book being used for LTP analysis and being constructed from entries resulting from said delayed total excitation signal which are completely searched in a first subblock of each block, whereas, in each of additional subblocks, only a defined partial range is searched, the partial range being determined by a respective preceding delay;

c) minimizing an output of a weighted, filtered deviation of a signal resulting from a respective evaluated excitation signal from an input voice signal to determine a multiple pulse component of said excitation signal;

d) processing digital voice signals in sequential overlapping blocks having a constant number of sampling values, each of which is divided into a defined number of subblocks having a defined number of sampling values, for determination of the excitation signal parameters;

e) scalar quantizing reflection coefficients resulting from a block per block analysis of said voice signal for determination of prediction parameters using individually adjusted quantization tables;

f) encoding delayed values for all subblocks together in a delay code and linearly quantizing each individual associated amplification factor;

g) encoding pulse positions of a subblock together in a position code and subjecting associated amplitudes to a block-adaptive linear quantization, wherein an amplitude value maximum is quantized non-linearly and each of normalized pulse amplitudes are linearly quantized, wherein each multiple pulse excitation vector determined for one subblock contains, in addition to a defined number of pulses, only zero sampling values;

h) decoding parameters for reclaiming quantized transfer values using table memories for regenerating quantized reflection coefficients;

i) generating an excitation signal using a regenerated component from said adaptive code book and a regenerated multiple pulse component, wherein a regenerated voice signal is constructed by synthesis filtration using said excitation signal and regenerated prediction parameters;

j) calculating, during parameter decoding, delays for components obtained from said adaptive code book from a delay code, wherein a delay for a first subblock of a block lies in a range of sampling periods and delays for three additional subblocks lie in ranges defined by a respective preceding delay;

k) determining, during parameter decoding, linearly quantized amplification factors for components of said excitation signal obtained from one of said adaptive code book and a table;

l) calculating, during parameter decoding, corresponding pulse positions for a multiple pulse component of said excitation vector of a subblock, wherein position values of said subblock are each extracted from a position code; and

m) determining, during parameter decoding, pulse amplitudes for a multiple component of said excitation signal of a subblock, wherein each non-linearly quantized block maximum and each linearly quantized normalized pulse amplitude is extracted for multiplicative regeneration of respective quantized unnormalized pulse amplitudes of said subblock.

2. The method of claim 1, wherein a sampling frequency between 11 and 13 kilohertz is utilized for sampling said digital voice signals and a number of sampling values of a block is set as a multiple of a number of subblocks.

3. The method of claim 2, wherein a number of sampling values is fixed at 288 and a number of subblocks is set to four such that a number of sampling values per subblock is set to 72.

4. The method of claim 1, wherein a number of reflection coefficients is set to 14 and 5, 5, 4, 4, 4, 4, 4, 4, 3, 3, 3, 3, 3, and one of 3 or 2 bits respectively are used for quantization thereof.

5. The method of claim 1, wherein step b) comprises the step of utilizing total excitation signal values, delayed by between 36 and 186 sampling periods, for generating said adaptive code book, wherein, after a second subblock, each subregion which is to be searched is set to 48 delay values, and a delay code is set to a length of 24 bits and linear quantization of an amplification factor is effected using 5 bits.

6. The method of method of claim 1, further comprising the step of prefiltering said voice signal prior to step a) to suppress low-frequency components and to slightly emphasize high-frequency components.

7. The method of claim 1, wherein prediction parameters are determined through analysis of said voice signal in defined time regions which respectively extend blocks by a defined amount of time into a past and into a future such that drifts in said voice signal are included in results of said analysis in view of overlap between neighboring analyzing sections.

8. The method of claim 7, wherein a defined length of an analysis region is 27 ms and said region is centered with respect to blocks to be processed.

9. The method of claim 1, wherein analysis sections are subjected to a trapezoidal windowing having a window side length of 14 sample values prior to performing an LPC analysis of 14^th order in accordance with an autocorrelation method with stabilization and prefiltering is carried out using a limiting frequency of 50 Hz with the scalar quantization of reflection coefficients being effected through a simple minimization of a Euclidean distance between unquantized values and the entries of a respective individually adjusted representation table, wherein resulting LPC parameters are subjected to a weighting using a factor of 0.7 for carrying out LTP analysis to determine an excitation signal component from said adaptive code book, wherein a power of a weighted and filtered deviation between a respective signal resulting from a respective code book entry and said input voice signal is minimized and LPC parameters derived from prediction parameter determination are weighted using a factor of 0.9 for determination of a multiple pulse component of said excitation signal.

10. The method of claim 1, wherein voice signal encoding and decoding is carried out using a variable bit rate and said input signal is monitored for occurrence of speech pauses and speech pauses are recognized, wherein a signal power of said input signal is evaluated for monitoring of said speech pauses, with speech monitoring entailing a comparison between a signal power of said input signal and an adaptive threshold value, wherein evaluation of said signal power of said input signal is carried out in blocks and a speech pause is detected if a determined signal power lies below a threshold value throughout a plurality of sequential blocks, wherein a speech recognition bit is stored for characterizing a presence of said speech pause, and when blocks occur having no voice signals, input sided production of an excitation vector is carried out using a noise generator.

11. A device for encoding and decoding voice signals, comprising:

means for analyzing a digital voice signal for determination of prediction parameters;

means for determining an excitation signal component from an adaptive code book, the adaptive code book being formed from a delayed total excitation signal, said adaptive code book being used for LTP analysis and being constructed from entries resulting from said delayed total excitation signal which are completely searched in a first subblock of each block, whereas, in each of additional subblocks, only a defined partial range is searched, the partial range being determined by a respective preceding delay;

means for minimizing an output of a weighted, filtered deviation of a signal resulting from a respective evaluated excitation signal from an input voice signal to determine a multiple pulse component of said excitation signal;

means for processing digital voice signals in sequential overlapping blocks having a constant number of sampling values, each of which is divided into a defined number of subblocks having a defined number of sampling values, for determination of the excitation signal parameters;

means for scalar quantizing reflection coefficients resulting from a block per block analysis of said voice signal for determination of prediction parameters using individually adjusted quantization tables;

means for encoding delayed values for all subblocks together in a delay code and linearly quantizing each individual associated amplification factor;

means for encoding pulse positions of a subblock together in a position code and subjecting associated amplitudes to a block-adaptive linear quantization, wherein an amplitude value maximum is quantized non-linearly and each of normalized pulse amplitudes are linearly quantized, wherein each multiple pulse excitation vector determined for one subblock contains, in addition to a defined number of pulses, only zero sampling values;

means for decoding parameters for reclaiming quantized transfer values using table memories for regenerating quantized reflection coefficients;

means for generating an excitation signal using a regenerated component from said adaptive code book and a regenerated multiple pulse component, wherein a regenerated voice signal is constructed by synthesis filtration using said excitation signal and regenerated prediction parameters;

means for calculating, during parameter decoding, delays for components obtained from said adaptive code book from a delay code, wherein a delay for a first subblock of a block lies in a range of sampling periods and delays for three additional subblocks lie in ranges defined by a respective preceding delay;

means for determining, during parameter decoding, linearly quantized amplification factors for components of said excitation signal obtained from one of said adaptive code book and a table;

means for calculating, during parameter decoding, corresponding pulse positions for a multiple pulse component of said excitation vector of a subblock, wherein position values of said subblock are each extracted from a position code; and

means for determining, during parameter decoding, pulse amplitudes for a multiple component of said excitation signal of a subblock, wherein each non- linearly quantized block maximum and each linearly quantized normalized pulse amplitude is extracted for multiplicative regeneration of respective quantized unnormalized pulse amplitudes of said subblock.

12. The device of claim 11, wherein said voice signal encoder comprises:

a pre-processing stage;

a voice signal analyzer for determination of prediction parameters from a digital voice signal;

an inversion filter connected to an output of said voice signal analyzer for determination of a residual signal from a digital voice signal using prediction parameters;

a circuit for determination of an excitation signal component from an adaptive code book, to an input side of which said prediction parameters, a residual signal and a delayed total excitation signal are applied; and

a circuit for determination of a multiple pulse component of an excitation vector, to an input side of which said prediction parameters, said residual signal, an excitation signal component from an adaptive code book and said delayed total excitation signal are applied;

wherein said voice signal decoder comprises;

a parameter decoder, containing table memories and calculation circuits for reconstructing said prediction parameters and parameters for said excitation signal component from said adaptive code book and said multiple pulse component from one of transmitted and stored values;

a circuit for determination of an excitation signal from a reconstructed component from said adaptive code book and from a reconstructed multiple pulse component; and

a synthesis filter to which said excitation signal and reconstructed prediction parameters are applied for forming a reconstructed voice signal.

13. The device of claim 12, wherein said pre-processing stage comprises a high-pass filter.

14. The device of claim 12, wherein said pre-processing circuit comprises a circuit for dividing a signal into signal blocks.

15. The device of claim 12, wherein said voice signal analyzer is an LPC analyzer.

16. The device of claim 12, wherein said voice signal analyzer comprises a circuit for weighting an input signal block using a window.

17. The device of claim 12, wherein said voice signal analyzer comprises a circuit for calculation of an auto-correlation sequence.

18. The device of claim 12, wherein said voice signal analyzer comprises a circuit for calculation of prediction parameters using a Durbin-Levinson recursion algorithm.

19. The device of claim 12, wherein said voice signal analyzer comprises a circuit for broadening a bandwidth.

20. The device of claim 12, wherein said voice signal analyzer comprises a circuit for determination of reflection coefficients.

21. The device of claim 12, wherein reflection coefficients are fed to a quantizer in which scalar quantization of said reflection coefficients is effected by a memory table.

22. The device of claim 12, wherein each signal block comprises several subblocks and said inversion filter determines a residual signal for each subblock.

23. The device of claim 12, wherein said circuit for determination of said excitation signal component from said adaptive code book comprises a first weighting circuit in which said prediction parameters are weighted with a first weighting factor.

24. The device of claim 12, wherein said circuit for determination of said multiple pulse component comprises a second weighting circuit in which said prediction parameters are weighted with a second weighting factor which differs from a first weighting factor.

25. The device of claim 12, wherein said voice signal encoder and said voice signal decoder work at a variable bit rate.

26. The device of claim 25, further comprising a speech pause detector which monitors said input signal for occurrence of speech pauses.

27. The device of claim 26, wherein said speech pause detector evaluates a signal power of said input signal.