Vector joint encoding/decoding method and vector joint encoder/decoder

• US 10,089,995 B2
• Filed: 12/26/2017
• Issued: 10/02/2018
• Est. Priority Date: 01/26/2011
• Status: Active Grant

• Current Assignee: Huawei Technologies Co., Ltd. (Huawei Investment & Holding Co., Ltd.)
• Original Assignee: Huawei Technologies Co., Ltd. (Huawei Investment & Holding Co., Ltd.)
• Inventors: Ma, Fuwei, Zhang, Dejun, Miao, Lei, Qi, Fengyan
• Primary Examiner(s): Guerra-Erazo, Edgar
• Application Number:
  US15/854,130
• Publication Number:
  US 20180137870A1
• Time in Patent Office: 280 Days
• Field of Search: 704500-504, 704216, 704217, 704219, 704223, 704225, 704229, 704258, 704262, 704263, 704264
• CPC Class Codes: G01L 19/12 Alarms or signals
• CPC Class Codes: G10L 19/038 Vector quantisation, e.g. TwinVQ audio
• CPC Class Codes: G10L 19/04 using predictive techniques
• CPC Class Codes: G10L 19/107 Sparse pulse excitation, e.g. by using algebraic codebook
• CPC Class Codes: G10L 19/12 the excitation function being a code excitation, e.g. in code excited linear prediction [CELP] vocoders
• CPC Class Codes: G10L 2019/0005 Multi-stage vector quantisation
• CPC Class Codes: G10L 21/00 Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility G10L19/00 takes precedence

Abstract

A vector joint encoding/decoding method and a vector joint encoder/decoder are provided, more than two vectors are jointly encoded, and an encoding index of at least one vector is split and then combined between different vectors, so that encoding idle spaces of different vectors can be recombined, thereby facilitating saving of encoding bits, and because an encoding index of a vector is split and then shorter split indexes are recombined, thereby facilitating reduction of requirements for the bit width of operating parts in encoding/decoding calculation.

Claims

1. A method for vector joint encoding for a voice signal on a CELP (code excited linear prediction) model, comprising:

calculating, by a processor of an encoder, an encoding index Ind_t of each vector from a plurality of vectors, wherein the each vector is determined based on the voice signal, wherein a subscript t denotes a t^th vector and t∈

[0, T−

1], and wherein T represents number of vectors and is an integer greater than or equal to 2;

splitting the Ind_t at least once into two split indexes Ind_t0 and Ind_t1 according to a set factor SLF_t, wherein the set factor SLF_t is a set split factor SLF for the t^th vector that varies according to the t^th vector, wherein the set factor SLF_t is a positive integer, wherein the Ind_t0 denotes a serial number of an interval to which the Ind_t belongs, wherein the Ind_t1 denotes a serial number of the Ind_t in the interval to which the Ind_t belongs, wherein a length of the interval is not greater than the SLF_t, and wherein Ind_t≤

Ind_t0×

SLF_t+Ind_t1;

obtaining a combined index Ind_SLF based on a split index of at least one vector and a split index of other vector for a further splitting if any of the Ind_t0 and Ind_t1 is uncombined; and

encoding the final output of the previous steps if all of the Ind_t0 and Ind_t1 is combined.

2. The method according to claim 1, wherein each of the vectors is represented as pulse distribution on a track, and wherein the encoding index is an index used for indicating the pulse distribution on the track.

3. The method according to claim 1, wherein the Ind_t0 split from the Ind_t is equal to

Ind_t0=Int(Ind_t/SLF_t), wherein Int( ) denotes rounding down to an integer, and the Ind_t1 split from the Ind_t is equal to Ind_t1=Ind_t% SLF_t, and wherein % denotes taking a remainder.

4. The method according to claim 3, wherein SLF_t=2^(K_t), wherein K_t is a positive integer, and wherein the obtaining the combined index Ind_SLF comprises:

selecting the Ind_t0 to participate in a combination for a vector t providing a split index to participate in the combination;

or

setting SLF_t=Int(Ind_t,max/2(K_t)), wherein Ind_t,max denotes a maximal value of the Ind_t, and wherein obtaining a combined index Ind_SLF comprises selecting the Ind_t1 to participate in a combination for the vector t providing a split index to participate the combination.

5. The method according to claim 4, wherein K_t=K_t,max−

2, K_t=K_t,max−

3, or K_t=K_t,max−

4, and wherein the K_t,max is a length of a bit sequence of the Ind_t,max.

6. The method according to claim 1, wherein the obtaining the combined index Ind_SLF comprises selecting a split index for embodying space occupancy characteristics of the Ind_t to participate in a combination for a vector t providing a split index to participate in the combination, wherein the embodying space occupancy characteristics of the Ind_t refers to that, compared with other split indexes, an occupancy rate of a value range of a selected split index for an encoding space is closest to an occupancy rate of a value range of the Ind_t for the encoding space.

7. The method according to claim 1, wherein the splitting the Ind_t at least once into the two split indexes Ind_t0 and Ind_t1 according to the set factor SLF_t comprises:

selecting a value of bits of the Ind_t as the Ind_t0, the bits being located at a set position; and

selecting a value of bits at a remaining position as the Ind_t1, wherein SLF_t=2^(K_t), wherein number of the bits at the set position is K_t0, wherein number of the bits at the remaining position is K_t, wherein K_t0+K_t=K_t,max, wherein K_t,max is a length of a bit sequence of Ind_t,max and wherein the Ind_t,max denotes a maximal value of the Ind_t.

8. The method according to claim 7, wherein the obtaining the combined index Ind_SLF comprises selecting, in a bit sequence with a length being K_t,max, a split index of the Ind_t to participate in a combination for a vector t providing a split index to participate in the combination, wherein the split index at least comprises a value of highest two bits.

9. The method according to claim 1, wherein the further splitting comprises:

splitting the combined index Ind_SLF into T1 recombined indexes Ind_t0′

according to a set value range, wherein T1 is less than or equal to number of vectors generating the combined index, wherein a value range of at least one Ind_t0′

is greater than or equal a value range of the split index of a corresponding vector t, wherein the split index participates in a combination, wherein a value range of at least one Ind_t0′

is less than or equal the value range of the split index of the corresponding vector t, wherein the split index participates in another combination;

combining each recombined index and an uncombined split index of a corresponding vector;

encoding the combined each recombined index and an uncombined split index of a corresponding vector; and

encoding an uncombined split index of the vector when the vector without having an allocated recombined index exists.

10. The method according to claim 9, wherein the splitting the combined index into T1 recombined indexes Ind_t0′

according to a set value range comprises splitting a total length K_SLF of the bit sequence of the combined index into T1 sections according to a set length, wherein a value of each section corresponds to one Ind_t0′

, wherein the K_SLF is the length of the bit sequence of an Ind_SLF,max, and wherein the Ind_SLF,max denotes a maximal value of the Ind_SLF.

11. The method according to claim 10, wherein the splitting the Ind_t into the two split indexes Ind_t0 and Ind_t1 according to the set factor SLF_t comprises:

selecting in the bit sequence with the length being K_t,max, a value of K_t0 bits starting from highest bit of the Ind_t as the Ind_t0;

selecting a value of remaining bits as the Ind_t1, wherein SLF_t=2^(K_t), wherein K_t0+K_t=K_t,max, wherein the K_t,max is a length of a bit sequence of the Ind_t,max, wherein the Ind_t,max denotes a maximal value of the Ind_t,

wherein Ind_t0 of a vector t that needs to provide a split index is selected as a base to obtain the combined index Ind_SLF,

wherein the splitting the total length K_SLF of the bit sequence of the combined index into the T1 sections according to the set length comprises splitting the K_SLF according to the K_t0 value used by the vector t generating the combined index, and

wherein number of bits split by each Ind_t0′

is less than or equal to the value of K_t0 used by corresponding vector t.

12. The method according to claim 1, wherein encoding the combined index Ind_SLF comprises:

comparing the combined index Ind_SLF;

adjusting a threshold value THR, wherein THR≤

2^(K_SLF)−

Ind_SLF,max, wherein the K_SLF is a length of a bit sequence of the Ind_SLF,max, and wherein the Ind_SLF,max denotes a maximal value of the Ind_SLF;

encoding the Ind_SLF by using a first number of encoding bits when the Ind_SLF is less than the THR;

encoding the Ind_SLF added with an offset value THR₀ by using a second number of encoding bits when the Ind_SLF is not less than the THR, wherein THR≤

THR₀≤

2^(K_SLF)−

Ind_SLF,max, wherein the first number is less than the second number, wherein the second number is less than or equal to the K_SLF, and wherein the first number and the second number are both positive integers; and

encoding other uncombined split indices.

13. A vector joint pulse encoder working on a CELP (code excited linear prediction) model for encoding a voice signal, comprising:

a processor;

a memory in communication with the processor containing computer instructions that when executed by the processor cause the processor to;

calculate an encoding index Ind_t of each vector from a plurality of vectors, wherein the each vector is determined based on the voice signal, wherein a subscript t denotes a t^th vector and t∈

[0, T−

1], and wherein T represents number of vectors and is an integer greater than or equal to 2;

split the Ind_t at least once into two split indexes Ind_t0 and Ind_t1 according to a set factor SLF_t, wherein the set factor SLF_t is a set split factor SLF for the t^th vector that varies according to the t^th vector, wherein the set factor SLF_t is a positive integer, wherein the Ind_t0 denotes a serial number of an interval to which the Ind_t belongs, wherein the Ind_t1 denotes a serial number of the Ind_t in the interval to which the Ind_t belongs, wherein a length of the interval is not greater than the SLF_t, and wherein Ind_t≤

Ind_t0×

SLF_t+Ind_t1; and

obtain a combined index Ind_SLF based on a split index of at least one vector and a split index of other vector for a further splitting if any of the Ind_t0 and Ind_t1 is uncombined yet, and encoding the final output of the previous steps if all of the Ind_t0 and Ind_t1 is combined.

14. The encoder according to claim 13, wherein the computer instructions that when executed by the processor cause the encoder further to:

split the combined index into T1 recombined indexes Ind_t0′

according to a set value range, wherein T1 is less than or equal to number of vectors generating the combined index, wherein a value range of at least one Ind_t0′

is greater than or equal a value range of the split index, participating in a combination, of a corresponding vector t, and wherein a value range of at least one Ind_t0′

is less than or equal the value range of the split index, participating in another combination, of the corresponding vector t; and

combine each recombined index and an uncombined split index of a corresponding vector respectively and then perform encoding; and

encode an uncombined split index of the vector when a vector without having an allocated recombined index exists.