Vector quantization of a time sequential signal by quantizing an error between subframe and interpolated feature vectors

• US 5,583,888 A
• Filed: 09/12/1994
• Issued: 12/10/1996
• Est. Priority Date: 09/13/1993
• Status: Expired due to Term

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Ono, Shigeru
• Primary Examiner(s): Chin, Wellington
• Assistant Examiner(s): LUTHER, WILLIAM A
• Application Number:
  US08/304,480
• Time in Patent Office: 820 Days
• Field of Search: 375/240, 375/243, 375/245, 375/285, 375/340, 375/346, 375/354, 375/359, 375/362, 375/371, 375/372, 341/106, 341/51, 370/98, 370/100.1, 370/105.1, 370/109, 370/110.1, 348/512, 348/513, 348/439, 348/418, 348/409, 381/29, 381/30, 381/34, 381/35, 381/39, 381/40, 395/2, 395/2.21
• US Class Current: 375/240
• CPC Class Codes: G10L 19/00 Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis in musical instruments G10H
• CPC Class Codes: H03M 7/3082 Vector coding for television signals, see H04N19/94

Abstract

For quantization of subframe feature vectors representative of subframes identified by subframe indexes in each frame of a time sequential signal, such as an audio or a video signal, a first feature vector of a first subframe of a current frame is extracted with additional extraction of a second feature vector of a second subframe of the current frame and is quantized by a frame vector quantizer unit into a quantized frame vector for the current frame. Using a quantized frame vector for a previous frame and referring to interpolation coefficient vectors, an interpolator interpolates an interpolated feature vector for the current frame. An error vector is calculated between the interpolated and the second feature vectors and is quantized by an error vector quantizer unit into a quantized error vector. The subframe feature vectors are quantized by using the subframe indexes, the quantized frame vectors, the interpolated feature vector, and the quantized error vector. Preferably, the interpolation coefficient vectors are adjusted according to a mode decided for the current frame by its subframes.

Claims

1. A vector quantizing method of vector quantizing a time sequential signal divided into consecutive frames, each comprising a plurality of subframes identified by subframe indexes, comprising the steps of:

extracting first and second feature vectors at least from first and second subframes of a current frame of said time sequential signal;

calculating a quantized frame vector for said current frame by using said first feature vector;

interpolating an interpolated feature vector for said current frame based on interpolation coefficient vectors between said quantized frame vector and another quantized frame vector calculated for a previous frame which is at least one frame prior to said current frame;

calculating an error between said second feature vector and said interpolated feature vector as an error feature vector;

quantizing said error feature vector into a quantized error vector; and

producing quantized subframe vectors for said plurality of subframes by using said subframe indexes, the quantized frame vector for said current frame, said interpolated feature vector, and said quantized error vector.

2. A vector quantizing method as claimed in claim 1, further comprising the steps of:

deciding a fundamental mode of each frame by using the subframes of said each frame; and

deciding said interpolation coefficient vectors by the fundamental mode of said current frame.

3. A vector quantizing method as claimed in claim 2, wherein said fundamental mode represents whether said time sequential signal builds up, is stationary, or builds down in said current frame.

4. A vector quantizing device for vector quantizing a time sequential signal divided into consecutive frames, each comprising a plurality of subframes identified by subframe indexes, said vector quantizing device comprising:

extracting means for extracting first and second feature vectors at least from first and second subframes of a current frame of said time sequential signal;

frame vector calculating means for calculating a quantized frame vector for said current frame by using said first feature vector;

interpolating means for interpolating an interpolated feature vector for said current frame based on interpolation coefficient vectors between said quantized frame vector and another quantized frame vector calculated for a previous frame which is at least one frame prior to said current frame;

error calculating means for calculating an error between said second feature vector and said interpolated feature vector as an error feature vector;

error quantizing means for quantizing said error feature vector into a quantized error vector; and

quantized subframe vector calculating means for calculating quantized subframe vectors for said plurality of subframes by using said subframe indexes, the quantized frame vector for said current frame, said interpolated feature vector, and said quantized error vector.

5. A vector quantizing device as claimed in claim 4, further comprising fundamental frame mode deciding means for deciding a fundamental mode of said current frame by using the subframes of said current frame.

6. A vector quantizing device as claimed in claim 5, wherein said fundamental mode represents whether said time sequential signal builds up, is stationary, or builds down in said current frame.

7. A vector quantizing device as claimed in claim 6, wherein said fundamental frame mode deciding means calculates a quotient of a product of root mean squares of two trailing ones of the subframes of said current frame divided by another product of root mean squares of two leading ones of the subframes of said current frame to decide that said fundamental mode represents said build up, said stationary state, for said build down of said time sequential signal in said current frame when said quotient is not greater than 0.5, is between 0.5 and 2.0, both inclusive, or is not less than 2.0, respectively.

8. A vector quantizing device as claimed in claim 6, wherein said fundamental frame mode deciding means calculates an average adaptive predictive gain of said current frame to decide that said fundamental mode represents said build up, said stationary state, or said build down of said time sequential signal in said current frame when said predictive gain is greater than 10.0 dB, is between 10 dB and 5.0 dB, or is between 5.0 dB and 2.0 dB, respectively.

9. A vector quantizing device as claimed in claim 5, wherein said interpolating means comprises:

a quantized frame vector buffer for storing therein, as stored frame vectors, quantized frame vectors calculated for frames preceding said current frame;

an interpolation codebook keeping a prescribed number of code vectors as said interpolation coefficient vectors; and

an interpolating circuit for interpolating said interpolated feature vector with reference to said code vectors between the quantized frame vector for said current frame and at least one of said stored frame vectors.

10. A vector quantizing device as claimed in claim 9, wherein said fundamental mode represents whether said time sequential signal builds up, is stationary, or builds down in said current frame.

11. A vector quantizing device as claimed in claim 10, wherein said interpolation codebook is trained in response to the fundamental modes of previous frames preceding said current frame.

12. A vector quantizing device as claimed in claim 10, wherein said fundamental frame mode deciding means calculates a quotient of a product of root mean squares of two trailing ones of the subframes of said current frame divided by another product of root mean squares of two leading ones of the subframes of said current frame to decide that said fundamental mode represents said build up, said stationary state, or said build down of said time sequential signal in said current frame when said quotient is not greater than 0.5, is between 0.5 and 2.0, both inclusive, or is not less than 2.0, respectively.

13. A vector quantizing device as claimed in claim 10, wherein said fundamental frame mode deciding means calculates an average adaptive predictive gain of said current frame to decide that said fundamental mode represents said build up, said stationary state, or said build down of said time sequential signal in said current frame when said predictive gain is greater than 10.0 dB, is between 10.0 dB and 5.0 dB, or is between 5.0 dB and 2.0 dB, respectively.

14. A vector quantizing device as claimed in claim 4, wherein said frame vector calculating means comprises:

a vector quantization codebook keeping a preselected number of frame vector quantization code vectors; and

a frame vector quantizer for quantizing said current frame into the quantized frame vector for said current frame by selecting one of said frame vector quantization code vectors that minimizes a square of a difference between said first feature vector and said one of frame vector quantization code vector.

15. A vector quantizing device as claimed in claim 14, further comprising fundamental frame mode deciding means for deciding a fundamental mode of said current frame by using the subframes of said current frame, said fundamental mode representing whether said time sequential signal builds up, is stationary, or builds down in said current frame.

16. A vector quantizing device as claimed in claim 15, wherein said interpolating means comprises:

a quantized frame vector buffer for storing therein, as stored frame vectors, quantized frame vectors calculated for previous frames preceding said current frame;

an interpolation codebook trained in response to the fundamental modes of said previous frames to keep a prescribed number of code vectors as said interpolation code vectors; and

an interpolating circuit for interpolating said interpolated feature vector with use of said code vectors between the quantized frame vector for said current frame and at least one of said stored frame vectors.

17. A vector quantizing device as claimed in claim 4, wherein said error quantizing means comprises:

a vector quantization codebook keeping a preselected number of quantized error code vectors for said current frame; and

an error vector quantizer for quantizing said error feature vector into said quantized error vector by selecting one of said quantized error code vectors as said quantized error vector that minimizes a square of a difference between said error vector and said one of quantized error feature code vectors.

18. A vector quantizing device as claimed in claim 17, further comprising fundamental frame mode deciding means for deciding a fundamental mode of said current frame by using one of said plurality of subframes of said current frame, said fundamental mode representing whether said time sequential signal builds up, is stationary, or builds down in said current frame.

19. A vector quantizing device as claimed in claim 18, wherein said interpolating means comprises:

a quantized frame vector buffer for storing therein, as stored frame vectors, quantized frame vectors calculated for previous frames preceding said current frame;

an interpolation codebook trained in response to the fundamental modes of said previous frame to keep a prescribed number of code vectors as said interpolation coefficient vectors; and

an interpolating circuit for interpolating said interpolated feature vector with use of said stored code vectors between the quantized frame vector for said current frame and at least one of said stored frame vectors.

20. A vector quantizing device as claimed in claim 4, wherein said current frame being identified by a frame number M and next following a preceding frame of another frame number (M-1), where M represents an integer which depends on said current frame in said time sequential signal and is equal at least to two, said error quantizing means producing an (M-1)-th quantized error vector {f(M-1)} for said preceding frame and an M-th quantized error vector {f(M)} for said current frame, where f represents one of vector components, a hat over f indicating that the vector component is a component of a quantized vector, arguments (M-1) and M representing that the vector components are components of (M-1)-th and M-th quantized vectors, said subframe indexes being represented for each frame in general by nsb, where n is variable between 1 and N so that the subframe indexes are variable between 1sb and Nsb with N made to represent a predetermined integer decided by said plurality, said first subframe being identified by the subframe index Nsb, said second subframe being identified by a subframe index Nnd, where Nnd represents one of the subframe indexes between 1sb and (N-1)sb, wherein said quantized subframe vector calculating means first calculates a vector sum of said interpolated feature vector and said quantized error vector as a quantized feature vector {f(M, Nnd)} for the second subframe of said current frame and subsequently calculates said quantized subframe vectors with each vector component of an nsb-th vector of the quantized subframe vectors of said current frame calculated:

for the subframe index nsb between 1sb and Nnd, both inclusive, by;

##EQU8## and for the subframe indexes nsb between Nnd plus one and Nsb, both inclusive, by;

##EQU9##