Variable bit rate LPC filter quantizing and inverse quantizing device and method

• US 9,245,532 B2
• Filed: 07/10/2009
• Issued: 01/26/2016
• Est. Priority Date: 07/10/2008
• Status: Expired

• Current Assignee: VoiceAge Corporation
• Original Assignee: VoiceAge Corporation
• Inventors: Gournay, Philippe, Bessette, Bruno, Salami, Redwan
• Primary Examiner(s): BAKER, MATTHEW H
• Application Number:
  US12/501,201
• Publication Number:
  US 20100023325A1
• Time in Patent Office: 2,391 Days
• Field of Search: 704219-223
• US Class Current: 1/1
• 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 19/18 Vocoders using multiple modes
• CPC Class Codes: H03M 7/3082 Vector coding for television signals, see H04N19/94

Abstract

A device and a method for quantizing a LPC filter in the form of an input vector in a quantization domain, comprises a calculator of a first-stage approximation of the input vector, a subtractor of the first-stage approximation from the input vector to produce a residual vector, a calculator of a weighting function from the first-stage approximation, a warper of the residual vector with the weighting function, and a quantizer of the weighted residual vector to supply a quantized weighted residual vector. A device and a method for inverse quantizing of a LPC filter, comprises means for receiving coded indices representative of a first-stage approximation of a vector representative of the LPC filter in a quantization domain and of a quantized weighted residual version of the vector, a calculator of an inverse weighting function from the first-stage approximation, an inverse quantizer of the quantized weighted residual version of the vector to produce a weighted residual vector, a multiplier of the weighted residual vector by the inverse weighting function to produce a residual vector, and an adder of the first-stage approximation with the residual vector to produce the vector representative of the LPC filter in the quantization domain.

Claims

1. A device for quantizing a LPC filter in the form of an input vector in a quantization domain, comprising:

a plurality of quantization modules using respective, distinct quantization modes, wherein the quantization modes include an absolute quantization mode and differential quantization modes, wherein the differential quantization modes use respective, different references each based on a previously quantized LPC filter or a combination of previously quantized LPC filters, and wherein each quantization module comprises;

a calculator of a first-stage approximation of the input vector, wherein the first stage approximation includes (a) the result of absolute quantization of the input vector in the case of the absolute quantization mode and (b) the reference in the case of the differential quantization modes;

a subtractor of the first-stage approximation from the input vector to produce a residual vector;

a calculator of weights of a weighting function using a mathematical relation including the first-stage approximation as a variable;

a warper of the residual vector by applying the weights of the weighting function to the residual vector; and

a quantizer of the weighted residual vector to supply a quantized weighted residual vector; and

a quantization mode selector configured to select one of the quantization modules and the corresponding quantization mode based on a level of distortion.

2. An LPC filter quantizing device according to claim 1, further comprising means for converting the LPC filter in the quantization domain to form the input vector.

3. An LPC filter quantizing device according to claim 1, further comprising a multiplexer of the first-stage approximation of the input vector and the quantized weighted residual vector.

4. An LPC filter quantizing device according to claim 1, wherein the calculator of the weights of the weighting function uses the first-stage approximation and a scaling factor which is dependent upon the quantization mode selected by the quantization mode selector to calculate the weights.

5. An LPC filter quantizing device according to claim 4, wherein the scaling factor has a value chosen to attain at least one of a certain average bit rate and a certain average distortion.

6. An LPC filter quantizing device according to claim 1, wherein the quantization domain is a line spectral frequency domain.

7. An LPC filter quantizing device according to claim 1, wherein the weighting function uses a scaling factor based on a type of the LPC filter and on a selected quantization mode.

8. An LPC filter quantizing device according to claim 1, wherein the quantization mode selector is configured to select the quantization mode providing a lower level of distortion for a given bit rate.

9. An LPC filter quantizing device according to claim 1, wherein the quantization mode selector is configured to select the quantization mode providing a lower bit rate for a target level of distortion.

28. An LPC filter quantizing device according to claim 7, wherein a combination of the type of LPC filter, of the quantization mode, and of the first-stage approximation is selected from the group consisting of:

LPC4 filter, with absolute quantization, and with 8-bit vector quantization (VQ) approximation;

LPC0 filter, with absolute quantization, and with 8-bit VQ approximation;

LPC0 filter, with relative LPC4 quantization, and with quantized LPC4 approximation;

LPC2 filter, with absolute quantization, and with 8-bit VQ approximation;

LPC2 filter, with relative LPC4 quantization, and with quantized LPC4 approximation;

LPC1 filter, with absolute quantization, and with 8-bit VQ approximation;

LPC1 filter, with relative (LPC0+LPC2)/2 quantization, and with quantized (LPC0+LPC2)/2 approximation;

LPC1 filter, with relative LPC2 quantization, and with quantized LPC2 approximation;

LPC3 filter, with absolute quantization, and with 8-bit VQ approximation;

LPC3 filter, with relative (LPC2+LPC4)/2 quantization, and with quantized (LPC2+LPC4)/2 approximation;

LPC3 filter, with relative LPC2 quantization, and with quantized LPC2 approximation; and

LPC3 filter, with relative LPC4 quantization, and with quantized LPC4 approximation;

wherein the LPC0 filter is the LPC filter of a last frame of a previous super-frame, the LPC1 filter is the LPC filter of a first frame of a current super-frame, the LPC2 filter is the LPC filter of a second frame of the current super-frame, the LPC3 filter is the LPC filter of a third frame of the current super-frame and the LPC4 filter is the LPC filter of a fourth frame of the current super-frame.

10. A device for quantizing a LPC filter in the form of an input vector in a quantization domain, comprising:

a plurality of quantization modules using respective, distinct quantization modes, wherein the quantization modes include an absolute quantization mode and differential quantization modes, wherein the differential quantization modes use respective, different references each based on a previously quantized LPC filter or a combination of previously quantized LPC filters, and wherein each quantization module comprises;

a calculator of a first-stage approximation of the input vector, wherein the first stage approximation includes (a) the result of absolute quantization of the input vector in the case of the absolute quantization mode and (b) the reference in the case of the differential quantization modes;

a subtractor of the first-stage approximation from the input vector to produce a residual vector;

a calculator of weights of a weighting function using a mathematical relation including the first-stage approximation as a variable;

a warper of the residual vector by applying the weights of the weighting function to the residual vector; and

a quantizer of the weighted residual vector to supply a quantized weighted residual vector; and

a quantization mode selector configured to select one of the quantization modules and the corresponding quantization mode based on a level of distortion;

wherein, in each of the quantization modules, the quantizer of the weighted residual vector comprises a variable bit rate quantizer.

11. An LPC filter quantizing device according to claim 10, wherein the variable bit rate quantizer comprises an algebraic vector quantizer.

12. A device for inverse quantizing a LPC filter quantized in an encoder, in the form of an input vector in a quantization domain, by a LPC filter quantizing device comprising (a) a plurality of quantization modules using respective, distinct quantization modes, wherein the quantization modes include an absolute quantization mode and differential quantization modes, wherein the differential quantization modes use respective, different references each based on a previously quantized LPC filter or a combination of previously quantized LPC filters, and wherein each quantization module comprises (i) a calculator of a first-stage approximation of the input vector, wherein the first stage approximation includes the result of absolute quantization of the input vector in the case of the absolute quantization mode and the reference in the case of the differential quantization modes, (ii) a subtractor of the first-stage approximation from the input vector to produce a residual vector, (iii) a calculator of weights of a weighting function using a mathematical relation including the first-stage approximation as a variable, (iv) a warper of the residual vector by applying the weights of the weighting function to the residual vector, and (v) a quantizer of the weighted residual vector to supply a quantized weighted residual vector;

and (b) a quantization mode selector configured to select one of the quantization modules and the corresponding quantization mode based on a level of distortion, the LPC filter inverse quantizing device comprising;

a demultiplexer for receiving, from the encoder, and for demultiplexing coded indices representative of (a) the first-stage approximation from the selected quantization module, (b) the selected quantization mode, and (c) the quantized weighted residual vector from the selected quantization module;

a determiner of the selected quantization mode using the demultiplexed coded indices;

a calculator of the first-stage approximation in response to the demultiplexed coded indices, and as a function of the determined quantization mode;

a calculator of an inverse weighting function from the first-stage approximation;

an inverse quantizer of the quantized weighted residual vector responsive to the demultiplexed coded indices to produce a weighted residual vector;

a multiplier of the weighted residual vector by the inverse weighting function to produce a residual vector; and

an adder of the first-stage approximation with the residual vector to produce a vector representative of the LPC filter in the quantization domain.

13. An LPC filter inverse quantizing device according to claim 12, wherein the inverse quantizer comprises a variable bit rate inverse algebraic vector quantizer.

14. An LPC filter inverse quantizing device according to claim 12, wherein the quantization domain is a line spectral frequency domain.

15. A method implemented in an encoder for quantizing a LPC filter in the form of an input vector in a quantization domain, comprising:

providing a plurality of distinct quantization modes, wherein the quantization modes include an absolute quantization mode and differential quantization modes, wherein the differential quantization modes use respective, different references each based on a previously quantized LPC filter or a combination of previously quantized LPC filters;

performing, for each of the plurality of distinct quantization modes, the following operations;

computing a first-stage approximation of the input vector, wherein the first stage approximation includes (a) the result of absolute quantization of the input vector in the case of the absolute quantization mode and (b) the reference in the case of the differential quantization modes;

subtracting the first-stage approximation from the input vector to produce a residual vector;

calculating weights of a weighting function using a mathematical relation including the first-stage approximation as a variable;

applying the weights of the weighting function to the residual vector; and

quantizing the weighted residual vector to supply a quantized weighted residual vector;

selecting one of the quantization modes based on a level of distortion; and

transmitting from the encoder to a decoder the selected quantization mode, the first-stage approximation of the input vector obtained using the selected quantization mode, and the quantized weighted residual vector obtained using the selected quantization mode.

16. An LPC filter quantizing method according to claim 15, further comprising converting the LPC filter in the quantization domain to form the input vector.

17. An LPC filter quantizing method according to claim 15, further comprising multiplexing the first-stage approximation of the input vector and the quantized weighted residual vector obtained using the selected quantization mode.

18. An LPC filter quantizing method according to claim 15, wherein calculating the weights of the weighting function comprises using the first-stage approximation and a scaling factor which is dependent upon the selected quantization mode.

19. An LPC filter quantizing method according to claim 18, wherein the scaling factor has a value chosen to attain at least one of a certain average bit rate and a certain average distortion.

20. An LPC filter quantizing method according to claim 15, wherein the quantization domain is a line spectral frequency domain.

21. A method implemented in an encoder for quantizing a LPC filter in the form of an input vector in a quantization domain, comprising:

providing a plurality of distinct quantization modes, wherein the quantization modes include an absolute quantization mode and differential quantization modes, wherein the differential quantization modes use respective, different references each based on a previously quantized LPC filter or a combination of previously quantized LPC filters;

performing, for each of the plurality of distinct quantization modes, the following operations;

computing a first-stage approximation of the input vector, wherein the first stage approximation includes (a) the result of absolute quantization of the input vector in the case of the absolute quantization mode and (b) the reference in the case of the differential quantization modes;

subtracting the first-stage approximation from the input vector to produce a residual vector;

calculating weights of a weighting function using a mathematical relation including the first-stage approximation as a variable;

applying the weights of the weighting function to the residual vector; and

quantizing the weighted residual vector to supply a quantized weighted residual vector;

selecting one of the quantization modes based on a level of distortion; and

transmitting from the encoder to a decoder the selected quantization mode, the first-stage approximation of the input vector obtained using the selected quantization mode, and the quantized weighted residual vector obtained using the selected quantization mode;

wherein quantizing the weighted residual vector comprises using a variable bit rate quantizer.

22. An LPC filter quantizing method according to claim 21, wherein using the variable bit rate quantizer comprises using an algebraic vector quantizer.

23. A method implemented in a decoder for inverse quantizing a LPC filter quantized in an encoder, in the form of an input vector in a quantization domain, by a LPC filter quantizing method comprising (a) providing a plurality of distinct quantization modes, wherein the quantization modes include an absolute quantization mode and differential quantization modes, wherein the differential quantization modes use respective, different references each based on a previously quantized LPC filter or a combination of previously quantized LPC filters, (b) performing, for each of the plurality of distinct quantization modes, the following operations:

(i) computing a first-stage approximation of the input vector, wherein the first stage approximation includes the result of absolute quantization of the input vector in the case of the absolute quantization mode and the reference in the case of the differential quantization modes, (ii) subtracting the first-stage approximation from the input vector to produce a residual vector, (iii) calculating weights of a weighting function using a mathematical relation including the first-stage approximation as a variable, (iv) applying the weights of the weighting function to the residual vector, and (v) quantizing the weighted residual vector to supply a quantized weighted residual vector, (c) selecting one of the quantization modes based on a level of distortion, and (d) transmitting from the encoder to the decoder the selected quantization mode, the first-stage approximation of the input vector obtained using the selected quantization mode, and the quantized weighted residual vector obtained using the selected quantization mode, the LPC filter inverse quantizing method comprising;

receiving at the decoder, from the encoder, coded indices representative of (a) the first-stage approximation from the selected quantization mode, (b) the selected quantization mode, and (c) the quantized weighted residual vector from the selected quantization mode;

determining the selected quantization mode using the received coded indices;

calculating, in response to the received coded indices, the first-stage approximation as a function of the determined quantization mode;

calculating an inverse weighting function from the first-stage approximation;

inverse quantizing the quantized weighted residual vector in response to the received coded indices to produce a weighted residual vector;

applying the inverse weighting function to the weighted residual vector to produce a residual vector; and

adding the first-stage approximation with the residual vector to produce a vector representative of the LPC filter in the quantization domain.

24. An LPC filter inverse quantizing method according to claim 23, wherein receiving the coded indices comprises demultiplexing the coded indices.

25. An LPC filter inverse quantizing method according to claim 23, wherein inverse quantizing the quantized weighted residual vector comprises variable bit rate inverse algebraic vector quantizing the quantized weighted residual vector.

26. An LPC filter inverse quantizing method according to claim 23, wherein applying the inverse weighting function to the weighted residual vector comprises multiplying the weighted residual vector by the inverse weighting function.

27. An LPC filter inverse quantizing method according to claim 23, wherein the quantization domain is a line spectral frequency domain.