Speech parameter encoding method capable of transmitting a spectrum parameter at a reduced number of bits

• US 5,271,089 A
• Filed: 11/04/1991
• Issued: 12/14/1993
• Est. Priority Date: 11/02/1990
• Status: Expired due to Term

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Ozawa, Kazunori
• Primary Examiner(s): Fleming, Michael R.
• Assistant Examiner(s): Hafiz, Tariq R.
• Application Number:
  US07/787,596
• Time in Patent Office: 771 Days
• Field of Search: 381/30, 381/36, 381/43, 395/2.2, 395/2.31, 395/2.39
• US Class Current: 704/222
• CPC Class Codes: G10L 19/07 Line spectrum pair [LSP] vocoders
• CPC Class Codes: G10L 2019/0005 Multi-stage vector quantisation
• CPC Class Codes: H03M 7/3082 Vector coding for television signals, see H04N19/94
• CPC Class Codes: H04L 2012/5652 Cell construction, e.g. including header, packetisation, depacketisation, assembly, reassembly
• CPC Class Codes: H04L 49/1553 Interconnection of ATM switching modules, e.g. ATM switching fabrics
• CPC Class Codes: H04L 49/255 Control mechanisms for ATM switching fabrics

Abstract

In a speech parameter encoding method of encoding an input speech signal into a sequence of encoded signals, vector quantization is carried out through a plurality of vector quantizers which are connected in cascade to one another through subtractors and which cooperate with code books, respectively. Such a cascade connection of the vector quantizers may be supplied with the input speech signal either at every frame or at every subframe shorter than the frame to produce a set of code vector candidates from each of the vector quantizers. Each set of the code vector candidates is sent to a cumulative distortion calculator to select an optimum combination of the code vector candidates and to produce, as the encoded signals, an index representative of the optimum combination. Alternatively, a sequence of coefficients which represent a spectrum parameter is calculated at every frame and is divided into a plurality of coefficient groups each of which is subjected to vector quantization to produce a plurality of code vector candidates for each coefficient group. Cumulative distortions are calculated for combinations of the code vector candidates to select an optimum combination.

Claims

1. A speech parameter encoding method of encoding a speech signal into a sequence of output encoded signals, said method comprising the steps of:

dividing said speech signal into a sequence of partial speech signals each of which has a predetermined period;

successively calculating a spectrum parameter at every one of said partial speech signals;

successively giving said spectrum parameter at a predetermined one of said partial speech signals to a vector quantizer circuit which comprises first through last stages of vector quantization units connected in cascade from the first stage to the last stage through a subtractor interposed between two adjacent ones of said first through said last stages of the vector quantization units, each of said vector quantization units carrying out vector quantization by the use of a code book which stores a plurality of code vectors determined in relation to quantization distortions from said spectrum parameter while said subtractor produces error signals;

calculating said first set of the code vector candidates in said first stage of the vector quantization units by the use of the code book of the first stage in consideration of said quantization distortions between said spectrum parameter and each of said code vectors stored in the code book of the first stage;

calculating the error signals representative of differences between said first set of the code vector candidates and the spectrum parameter by the use of said subtractor;

successively calculating, in response to the spectrum parameter and the error signals sent through said subtractor, other code vector candidates from a next following one of the stages to the last stage of said vector quantization units to successively produce the other code vector candidates in connection with the error signals calculated by said subtractor;

selecting a specific one of combinations of said first through last sets of the code vector candidates that is concerned with the quantization distortions successively calculated in each of said first through last stages in a time division fashion; and

producing said specific one of the combinations as the output encoded signals.

2. A speech parameter encoding method as claimed in claim 1, wherein each of said partial speech signals is located within a frame while said predetermined period is a frame period.

3. A speech parameter encoding method as claimed in claim 2, wherein said selecting step comprises the steps of:

calculating, from said quantization distortions, cumulative distortions for all of the combinations of the first through last sets of the code vector candidates; and

selecting, as said specific one of the combinations, one of the combinations that minimizes the cumulative distortions.

4. A speech parameter encoding method as claimed in claim 1, wherein said selecting step comprises the steps of:

calculating last stage distortions for all of the combinations of the first through last sets of the code vector candidates; and

selecting, as said specific one of the combinations, one of the combinations that minimizes the last stage distortions.

5. A speech parameter encoding method as claimed in claim 1, said predetermined period being a subframe which is shorter than a frame and which has said predetermined period, wherein said dividing step comprises the steps of:

dividing said speech signal into a plurality of frame speech signals each of which is located within the frame; and

subdividing each of said frame speech signals into said partial speech signals each of which is located within the predetermined period.

6. A speech parameter encoding method as claimed in claim 5, wherein said selecting step comprises the steps of:

calculating, from said quantization distortions, cumulative distortions for all of the combinations of the first through last sets of the code vector candidates; and

determining, as said specific one of the combinations, one of the combinations that is concerned with the cumulative distortions.

7. A speech parameter encoding method as claimed in claim 6, wherein said determining step comprises:

calculating the quantization distortion of a preselected one of said subframes by the use of said code vector candidates and a predetermined code book previously obtained by training; and

selecting, as said specific one of the combinations, one of said combinations that minimizes sums of the cumulative distortions and said quantization distortion of said preselected one of said subframes.

8. A speech parameter encoding method as claimed in claim 5, wherein said selecting step comprises the steps of:

calculating last stage distortions for all of the combinations of the first through last sets of the code vector candidates; and

determining, as said specific one of the combinations, one of the combinations that is concerned with the last stage distortions.

9. A speech parameter encoding method as claimed in claim 8, wherein said determining step comprises:

calculating the quantization distortion of a preselected one of said subframes by the use of said code vector candidates and a predetermined code book previously obtained by training; and

selecting, as said specific one of the combinations, one of said combinations that minimizes sums of the last stage distortions and said quantization distortion of said preselected one of said subframes.

10. A speech parameter encoding method of encoding a speech signal into a sequence of output encoded signals, said method comprising the steps of:

dividing said speech signal into a sequence of frame speech signals each of which has a predetermined period;

subdividing each of said frame speech signals into a plurality of subframe speech signals each of which lasts for a subframe period shorter than said frame period;

successively calculating a spectrum parameter for a predetermined one of said subframe speech signals;

successively giving said spectrum parameter calculated for said predetermined one of the subframe speech signals to a vector quantizer circuit which comprises first through last stages of vector quantization units connected in cascade from the first stage to the last stage through a subtractor interposed between two adjacent ones of said first through said last stages of the vector quantization units, each of said vector quantization units carrying out vector quantization by the use of a code book which stores a plurality of code vectors determined in relation to quantization distortions from said spectrum parameter while said subtractor produces error signals;

calculating said first set of the code vector candidates in said first stage of the vector quantization units by the use of the code book of the first stage in consideration of said quantization distortions between said spectrum parameter and each of said code vectors stored in the code book of the first stage;

calculating the error signals representative of differences between said first set of the code vector candidates and the spectrum parameter by the use of said subtractor;

successively calculating, in response to the spectrum parameter and the error signals sent through said subtractor, other code vector candidates from a next following one of the stages to the last stage of said vector quantization units to successively produce the other code vector candidates as second through the last sets of the code vector candidates in connection with the error signals calculated by said subtractor;

obtaining selected code vector sets which minimize one of cumulative distortions in the first through last stages and last stage distortions in the last stage;

preparing a coefficient code book which stores coefficients related to said spectrum parameter;

successively calculating the quantization distortions of a prescribed one of said subframes by the use of said selected code vector sets and said coefficient code book;

selecting a specific one of said coefficients that minimizes the quantization distortions successively calculated in each of said first through last stages in a time division fashion; and

producing said specific one of the coefficients as the output encoded signals.

11. A speech parameter encoding method of encoding a speech signal into a sequence of encoded speech signals by the use of a spectrum parameter having an order specified by dimensions of coefficients, said method comprising the steps of:

dividing said speech signal into a sequence of frame speech signals each of which lasts for a frame period;

calculating said spectrum parameter to a predetermined order for each of said frame speech signals, said predetermined order of the spectrum parameter being represented by the coefficients having a dimension number which is equal to the predetermined order;

dividing said coefficients into a plurality of divided coefficients each of which has a divided dimension number smaller than said order number;

preparing a plurality of code books each of which stores a plurality of code vectors for said divided coefficients;

successively searching each of said code books for each of said divided coefficients to read selected code vectors out of each of said code books and to produce said selected code vectors as code vector candidates for each of said divided coefficients;

calculating cumulative distortions for said predetermined order in connection with combinations of said code vector candidates;

selecting, as a specific code vector, one of said combinations that minimizes said cumulative distortions; and

producing said specific code vector as said encoded speech signals.

12. A speech parameter encoding method of encoding a speech signal into a sequence of encoded speech signals by the use of a spectrum parameter which has an order and which is specified by coefficients having dimensions, said method comprising the steps of:

dividing said speech signal into a sequence of frame speech signals each of which lasts for a frame period;

calculating said spectrum parameter to a predetermined order for each of said frame speech signals to produce a restricted spectrum parameter of said predetermined order represented by the coefficients having a dimension number which is equal to the predetermined order;

carrying out vector quantization for said restricted spectrum parameter through a plurality of vector quantizer stages to produce a set of code vector candidates from at least one of said vector quantizer stages and to produce the other sets of code vector candidates from the other code vector quantizer stage by dividing the coefficients of said restricted spectrum parameter into a plurality of partial coefficients each of which has a partial dimension smaller than said dimension number of said restricted spectrum parameter, by preparing a plurality of code books for said partial coefficients, and by searching the code books for the respective partial coefficients;

calculating cumulative distortions for combinations of said set and the other sets of the code vector candidates;

selecting a specific one of the combinations that minimizes the cumulative distortions; and

producing said specific one of the combinations as the encoded speech signals.

13. A speech parameter encoding method of encoding a speech signal into a sequence of encoded speech signals by the use of a spectrum parameter which has an order and which is specified by coefficients having dimensions, said method comprising the steps of:

dividing said speech signal into a plurality of frame speech signals each of which lasts for a frame period;

subdividing each of said frame speech signals into a plurality of subframe speech signals each of which lasts for a subframe period shorter than said frame period;

calculating the spectrum parameter to a predetermined order for at least one of said subframe speech signals to produce at least one restricted spectrum parameter of said predetermined order composed of the coefficients having a dimension number equal to said predetermined order;

quantizing a selected one of said at least one restricted spectrum parameter into a quantized value by the use of code vector candidates stored in a code book; and

further quantizing another restricted spectrum parameter for remaining at least one of said subframe speech signals by the use of said quantized value.

14. A speech parameter encoding method as claimed in claim 13, wherein said further quantizing step comprises the steps of:

preparing a coefficient code book which stores code vectors representative of coefficients;

calculating another restricted spectrum parameter for remaining at least one of said subframe speech signals with reference to said code vectors stored in said coefficient code book together with said quantized value.

15. A speech parameter encoding method as claimed in claim 13, wherein said further quantizing step comprising the steps of:

nonlinearly processing another restricted spectrum parameter to produce a processed parameter; and

calculating another restricted spectrum parameter for remaining at least one of said subframe speech signals with reference to said processed parameter together with said quantized value.

16. A speech parameter encoding method of encoding a speech signal into a sequence of encoded speech signals by the use of a spectrum parameter which has an order and which is specified by coefficients having dimensions, said method comprising the steps of:

dividing said speech signal into a plurality of frame speech signals each of which lasts for a frame period;

subdividing each of said frame speech signals into a plurality of subframe speech signals each of which lasts for a subframe period shorter than said frame period;

calculating the spectrum parameter to a predetermined order for at least one of said subframe speech signals to produce at least one restricted spectrum parameter of said predetermined order composed of the coefficients having a dimension number equal to said predetermined order;

quantizing a selected one of said at least one restricted spectrum parameter into a quantized value by the use of code vector candidates stored in a code book to produce a quantized value signal representative of said quantized value;

calculating a difference between said quantized value and another restricted spectrum parameter for remaining at least one of said subframe speech signals to produce a difference signal representative of said difference;

further quantizing said difference signal to produce a quantized difference signal by the use of code vectors which are stored in a code book to represent differences; and

producing said quantized value signal and said quantized difference signal as said encoded speech signals.

17. A speech parameter encoder for use in encoding a speech signal into a sequence of output encoded signals, said encoder comprising:

dividing means supplied with said speech signal for dividing said speech signal into a sequence of partial speech signals each of which has a predetermined period;

calculation means supplied with said partial speech signal sequence for successively calculating a spectrum parameter at every one of said partial speech signals; and

a vector quantization circuit supplied with said spectrum parameter at a predetermined one of said partial speech signals for carrying out vector quantization of the spectrum parameter at said predetermined one of the partial speech signals to produce said output encoded signal sequence;

said vector quantization circuit comprising;

first through last stages of vector quantization units each of which carries out vector quantization of a stage input signal, said first through said last stages of the vector quantization units producing first through last stage sets of code vector candidates together with first through last stages of vector quantization distortions concerned with the first through the last stage sets of the code vector candidates, respectively;

subtractor means between two adjacent ones of said first through said last stages of the vector quantization units to connect said two adjacent ones of the first through the last stages of the vector quantization units in cascade to each other, said two adjacent ones of the first through the last stages of the vector quantization units being composed of a preceding and a following stages of the vector quantization units, said subtractor means being supplied with the spectrum parameter and a selected one of the first through the last stage sets of the code vector candidates that is sent from the preceding stage of the vector quantization units to produce error signals representative of differences between the spectrum parameter and the selected one of the first through the last stage sets of the code vector candidates, said error signal being sent to the following stage of the vector quantization units as said stage input signal; and

cumulative calculation means supplied with the first through the last stages of the vector quantization distortions for calculating a sum of the first through the last stages of the vector quantization distortions to produce a cumulative distortion as said sequence of the output encoded signals.

18. A speech parameter encoder as claimed in claim 17, wherein each of said first through said last stages of vector quantization units comprises:

a code book for storing a plurality of code vectors; and

a vector quantizer supplied with said stage input signal and connected to said code book for carrying out the vector quantization of the stage input signal with reference to the code vectors read out of the code book to produce each of said first through said last stage sets of the code vector candidates together with each of said first through said last stages of the vector quantization distortions.

19. A speech parameter encoder as claimed in claim 17, wherein each of said first through said last stages of vector quantization units comprises:

a code book for storing a plurality of code vectors; and

a vector quantizer supplied with said stage input signal and connected to said code book for carrying out the vector quantization of the stage input signal with reference to the code vectors read out of the code book to produce each of said first through said last stage sets of the code vector candidates together with each of said first through said last stages of the vector quantization distortions.

20. A speech parameter encoder for use in encoding a speech signal into a sequence of output encoded signals, said encoder comprising:

dividing means supplied with said speech signal for dividing said speech signal into a sequence of frame speech signals each of which has a predetermined period;

subdividing means supplied with said sequence of frame speech signals for subdividing each of said frame speech signals into a plurality of subframe speech signals each of which lasts for a subframe period shorter that said frame period;

calculating means supplied with said plurality of subframe signals for successively calculating a spectrum parameter for a predetermined one of said subframe speech signals;

a vector quantization circuit, comprising a plurality of vector quantization units which are connected in cascade from one another through a subtractor and which successively process the spectrum parameters in series, for said predetermined one of the subframe speech signals for carrying out vector quantization of said spectrum parameter at said predetermined one of the subframe speech signals to produce said output encoded signal sequence;

said vector quantization circuit comprising;

first through last stages of vector quantization units, each of which carries out vector quantization of a stage input signal, said first through said last stages of the vector quantization units producing first through last stage sets of code vector candidates together with first through last stages of vector quantization distortions concerned with the first through the last stage sets of the code vector candidates, respectively;

subtractor means between two adjacent ones of said first through said last stages of the vector quantization units to connect said two adjacent ones of the first through the last stages of the vector quantization units in cascade to each other, said two adjacent ones of the first through the last stages of the vector quantization units being composed of a preceding and a following stages of the vector quantization units, said subtractor means being supplied with the spectrum parameter and a selected one of the first through the last stage sets of the code vector candidates that is sent from the preceding stage of the vector quantization units to produce error signals representative of differences between the spectrum parameter and the selected one of the first through the last stage sets of the code vector candidates, said error signal being sent to the following stage of the vector quantization units as said stage input signal;

cumulative calculation means supplied with the first through the last stages of the vector quantization distortions for calculating a sum of the first through the last stages of the vector quantization distortions, for obtaining selected code vector sets which minimize one of cumulative distortions in the first through last stages and last stage distortions in the last stage;

a coefficient code book which stores coefficients related to said spectrum parameter being used with said selected code vector sets to successively calculate the quantization distortions of a prescribed one of said subframes;

said cumulative calculation means selecting a specific one of said coefficients that minimizes the quantization distortions successively calculated in each of said first through last stages in a time division fashion and producing said specific one of the coefficients as the output encoded signals.

21. A speech parameter encoder for encoding a speech signal into a sequence of encoded speech signals by the use of a spectrum parameter having an order by dimensions of coefficients comprising:

dividing means for dividing said speech signal into a sequence of frame speech signals each of which lasts for a frame period;

calculation means for calculating said spectrum parameter to a predetermined order for each of said frame speech signals, said predetermined order of the spectrum parameter being represented by the coefficients having a dimension number which is equal to the predetermined order;

a vector quantization circuit supplied with said spectrum parameter for dividing said coefficients into a plurality of divided coefficients each of which has a divided dimension number smaller than said order number;

a plurality of code books each of which stores a plurality of code vectors for said divided coefficients;

said vector quantization circuit successively searching each of said code books for each of said divided coefficients to read selected code vectors out of each of said code books and to produce said selected code vectors as code vector candidates for each of said divided coefficients;

cumulative calculation means for calculating cumulative distortions for said predetermined order in connection with combinations of said code vector candidates;

minimum detecting means for selecting, as a specific code vector, one of said combinations that minimizes said cumulative distortions and for producing said specific code vector as said encoded speech signals.

22. A speech parameter encoder for encoding a speech signal into a sequence of encoded speech signals by the use of a spectrum parameter which has an order and which is specified by coefficients having dimensions, said encoder comprising:

dividing means for dividing said speech signal into a sequence of frame speech signals each of which lasts for a frame period;

calculation means for calculating said spectrum parameter to a predetermined order for each of said frame speech signals to produce a restricted spectrum parameter of said predetermined order represented by the coefficients having a dimension number which is equal to the predetermined order;

a vector quantization circuit supplied with said spectrum parameter for carrying out vector quantization for said restricted spectrum parameter through a plurality of vector quantizer stages to produce a set of code vector candidates from at least one of said vector quantizer stages and to produce the other sets of code vector candidates from the other code vector quantizer stage by dividing the coefficients of said restricted spectrum parameter into a plurality of partial coefficients each of which has a partial dimension smaller than said dimension number of said restricted spectrum parameter, by preparing a plurality of code books for said partial coefficients, and by searching the code books for the respective partial coefficients;

cumulative calculation means for calculating cumulative distortions for combinations of said set and other sets of the code vector candidates; and

a minimum detecting means for selecting a specific one of the combinations that minimizes the cumulative distortions and for producing said specific one of the combinations as the encoded speech signals.

23. A speech parameter encoder as claimed in claim 22, further comprising:

a coefficient code book which stores code vectors representative of coefficients; and

said calculation means calculating another restricted spectrum parameter, to be supplied to said vector quantization unit, for remaining at least one said subframe speech signals with reference to said code vectors stored in said coefficient code book together with said quantized value.

24. A speech parameter encoder for encoding a speech signal into a sequence of encoded speech signals by the use of a spectrum parameter which has an order and which is specified by coefficients having dimensions, said encoder comprising:

dividing means for dividing said speech signal into a plurality of frame speech signals each of which lasts for a frame period;

subdividing means for subdividing each of said frame speech signals into a plurality of subframe speech signals each of which lasts for a subframe period shorter than said frame period;

calculation means for calculating the spectrum parameter to a predetermined order for at least one of said subframe speech signals to produce at least one restricted spectrum parameter of said predetermined order composed of the coefficients having a dimension number equal to said predetermined order;

a vector quantization circuit supplied with said spectrum parameter for quantizing a selected one of said at least one restricted spectrum parameter into a quantized value by the use of code vector candidates stored in a code book and further quantizing another restricted spectrum parameter for remaining at least one of said subframe speech signals by the use of said quantized value.

25. A speech parameter encoder for encoding a speech signal into a sequence of encoded speech signals by the use of a spectrum parameter which has an order and which is specified by coefficients having dimensions, said encoder comprising:

dividing means for dividing said speech signal into a plurality of frame speech signals each of which lasts for a frame period;

subdividing means for subdividing each of said frame signals into a plurality of subframe speech signals each of which lasts for a subframe period shorter than said frame period;

calculation means for calculating the spectrum parameter to a predetermined order for at least one of said subframe speech signals to produce at least one restricted spectrum parameter of said predetermined order composed of the coefficients having a dimension number equal to said predetermined order;

a vector quantization circuit supplied with said spectrum parameter for quantizing a selected one of said at least one restricted spectrum parameter into a quantized value by the use of code vector candidates stored in a code book to produce a quantized value signal representative of said quantized value;

a difference vector quantization circuit for calculating a difference between said quantized value and another restricted spectrum parameter for remaining at least one of said subframe speech signals to produce a difference signal representative of said difference, for further quantizing said difference signal to produce a quantized difference signal by the use of code vectors which are stored in a code book to represent differences, and for producing said quantized value signal and said quantized difference signal as said encoded speech signals.