CN101572093B

CN101572093B - Transcoding method and device

Info

Publication number: CN101572093B
Application number: CN 200810105691
Authority: CN
Inventors: 鲍长春; 徐昊; 唐繁荣; 胡翔宇
Original assignee: Huawei Technologies Co Ltd; Beijing University of Technology
Current assignee: Huawei Technologies Co Ltd; Beijing University of Technology
Priority date: 2008-04-30
Filing date: 2008-04-30
Publication date: 2012-04-25
Anticipated expiration: 2028-04-30
Also published as: CN101572093A; WO2009132544A1

Abstract

The embodiment of the invention provides a transcoding method and a transcoding device. The method comprises the following steps: obtaining adaptive codebook gain g of current subframe in source code stream_pAnd a scale factor gamma of the fixed codebook gain_gc(ii) a According to a scaling factor alpha, for the gamma_gcProcessed to obtain alpha x gamma_gc(ii) a According to the quantization rule of the target end, the alpha multiplied by the gamma is measured_gcAnd g_pAnd quantizing to realize transcoding. The device comprises: the device comprises an acquisition module, a scaling module and a quantization module. The invention introduces a scaling factor to a fixed codebook gain scale factor gamma_gcAfter zooming, searching in the gain code table of the target coder by adopting a new quantization criterion to complete the transcoding of the gain parameters, thereby effectively reducing the computational complexity of the transcoding of the gain parameters and improving the quality of the transcoded voiceAmount of the compound .

Description

Transcoding method and device

Technical Field

The present invention relates to the field of encoding and decoding, and in particular, to a transcoding method and apparatus.

Background

Voice coding based on Code-Excited Linear Prediction (CELP) is one of the core technologies in current VoIP (Voice over Internet Protocol) and mobile communication systems, and in order to implement compatibility and intercommunication of communication devices between different providers, transcoding work between different CELP Voice coding standards is required.

At present, when transcoding between different CELP voice coding standards is performed, a transcoding method between different CELP voice coding standards based on a novel gain conversion technology is widely used, and the method realizes transcoding between different coding standards by respectively transmitting parameters transmitted in code streams after certain processing. The parameters involved are: line spectrum frequency, integer and fractional Pitch delay, fixed codebook symbol and position parameters and gain parameters, see fig. 1, which provides a schematic diagram of a transcoding system between different CELP speech coding standards based on the parameters, wherein, communication devices a and B operate under different CELP speech coding standards, a parameter decoder/encoder a in a communication device C in a communication link between the communication device a and the communication device B receives a Code stream of the communication device a and decodes the parameters to obtain corresponding parameters, and then transcoding of the Line spectrum frequency parameters and transcoding of Pitch (Pitch) are realized by LSP (Linear spectrum parameters) transcoding, and transcoding of integer and fractional Pitch delay parameters is realized by Pitch (Pitch Code-Excited Linear Prediction), and transcoding of fixed codebook symbol and position parameters is realized by ACELP (algebraic Code-Excited Linear Prediction) Gain transcoding realizes transcoding of Gain parameters, and the parameter coder/decoder B acquires each transcoded parameter, codes the parameter and sends the coded code stream to the opposite communication equipment B, thereby realizing transcoding work of different CELP voice coding standards and successfully realizing communication between communication equipment working under different CELP voice coding standards.

When the Gain parameters are transcoded, a method of directly transcoding a DTE (Decode Then Encode) can be adopted, and the target encoder transcodes the Gain parameters by solving and quantizing the Gain parameters: firstly, obtaining the adaptive codebook gain g of the current subframe by using an solving formula_pAnd fixed codebook gain g_cThen, according to the joint optimization process of the optimal excitation, the mean square error term e is adopted_jAs a search criterion, search a gain quantization code table in the target encoder so that e_jMinimum g_pAnd g_cThe gain value is the final quantized gain value; finally, the index value of the optimal gain quantization value in the gain code table is written into the code stream of the target encoder, and the transcoding operation of the gain parameter is completed. However, when the DTE method is adopted to realize the transcoding of the gain parameters, the g is obtained_pAnd g_cWhen the adaptive codebook vector and the generation of the current subframe need to be recoveredThe digital book vector needs to be restored, and then secondary quantization is carried out, so that the calculation complexity is greatly increased, and in addition, the quality of the transcoded voice is reduced due to the two-time gain quantization.

In addition, when the transcoding of the gain parameter is implemented, the adaptive codebook gain g directly transmitted by the source encoder can be adopted_pSum algebraic codebook gain g_cMethod for transcoding without adaptive codebook gain g in target encoder_pSum algebraic codebook gain g_cWhen the gain parameter quantization is carried out, the following mean square error terms are adopted as search criteria: e (L) ([ ex1(i) -ex2(i, L))]²(i ═ 0, 1.., N), where N is the subframe length,

where ex1 is the reference excitation signal ex1(i, L) ═ g_p(L)×v(i)+g_c(L) x c (i), (i ═ 0, 1,. N), where g is_p(L) is the adaptive codebook gain value, g, passed from the source encoder_cV (i), c (i) are adaptive codebook vector and algebraic codebook vector, respectively, for algebraic codebook gain values passed from the source encoder; ex2 is the resultant excitation signal ex2(i, L) ═ g after transcoding_p′(L)×v(i)+g_c' (L) × c (i), (i ═ 0, 1.., N), where g is_p' (L) is the adaptive codebook gain value for the Lth codeword in the target encoder, g_c' (L) is the algebraic codebook gain value for the L-th codeword in the target encoder.

G for minimizing E (L) in the above formula by searching the gain quantization code table in the target encoder_p' (L) and g_c' (L) is the gain value after the final quantization, the index value of the optimal gain quantization value in the gain code table is written into the code stream of the target encoder, namely the transcoding operation of the gain parameter is completed. However, when the method is used for realizing the transcoding of the gain parameters, the gain prediction process needs to be reserved. When the gain parameter quantization is carried out, a large amount of calculation is needed to ensure the conservation of signal transmission energy, the calculation complexity is increased, and different CELP standard gain quantization codes are adoptedThe difference in the text affects the quality of the transcoded speech.

Disclosure of Invention

In order to reduce the computational complexity of gain parameter transcoding and improve the quality of transcoded voice in the transcoding process between different CELP voice coding standards, embodiments of the present invention provide a transcoding method and apparatus. The technical scheme is as follows:

in one aspect, a transcoding method is provided, where the method includes:

obtaining adaptive codebook gain g of current subframe in source code stream_pAnd a scale factor gamma of the fixed codebook gain_gc；

According to a scaling factor alpha, for the gamma_gcProcessed to obtain alpha x gamma_gc；

According to the quantization rule of the target end, the alpha multiplied by the gamma is measured_gcAnd g_pQuantizing to realize transcoding, specifically comprising:

obtaining a weight factor beta, and according to the weight factor beta and a gain vector code table of the target end, carrying out the treatment on the alpha multiplied by the gamma_gcAnd g_pPerforming joint quantization to obtain a quantization code word in the gain vector code table; or

According to the target fixed codebook scale factor code table and the adaptive codebook gain code table, respectively aligning the alpha x gamma_gcAnd g_pAnd quantizing to obtain a quantized fixed code word in a quantization scale factor code table and a quantized adaptive code word in an adaptive code book gain code table.

In another aspect, a transcoding device is provided, the device comprising:

an obtaining module, configured to obtain adaptive codebook gain g of a current subframe in a source-end code stream_pAnd a scale factor gamma of the fixed codebook gain_gc；

A scaling module, configured to apply the gamma obtained by the obtaining module according to a scaling factor α_gcScaling to obtain alpha x gamma_gc；

A quantization module for quantizing the α × γ according to a quantization rule of a target end_gcAnd g_pQuantizing to realize transcoding, specifically: when the target end pair is to the adaptive codebook gain g_pAnd a scale factor gamma of the fixed codebook gain_gcWhen joint quantization is performed, the method comprises the following steps; a weight factor beta obtaining module for obtaining the gain g according to the self-adaptive codebook_pAnd a scale factor gamma of the fixed codebook gain_gcDetermining the weighting factor β; the quantization module is specifically a quantization unit, configured to perform the quantization on α × γ according to the gain vector code table at the target end and the β acquired by the weighting factor β acquisition module_gcAnd g_pPerforming joint quantization to obtain a quantized code word in the gain vector code table to realize transcoding; or

When the target end pair is to the adaptive codebook gain g_pAnd a scale factor gamma of the fixed codebook gain_gcWhen performing independent quantization, the quantization module is specifically a quantization unit, configured to separately perform quantization on α × γ according to the target fixed codebook scale factor code table and the adaptive codebook gain code table_gcAnd g_pAnd quantizing to obtain a quantized fixed code word in a quantized scale factor code table and a quantized adaptive code word in an adaptive code book gain code table, so as to realize transcoding.

The technical scheme provided by the embodiment of the invention has the beneficial effects that:

gain scale factor gamma to fixed codebook by introducing scaling factor_gcAfter scaling, searching in a gain code table of a target encoder by adopting a new quantization criterion to complete transcoding of gain parameters, thereby effectively reducing the computational complexity of transcoding of the gain parameters and improving the quality of transcoded voice.

Drawings

FIG. 1 is a schematic diagram of a transcoding system between different CELP speech coding standards provided by the prior art based on parameters;

fig. 2 is a flowchart of a transcoding method provided in embodiment 1 of the present invention;

fig. 3 is a flowchart of a method for determining the scaling factor α according to embodiment 1 of the present invention;

fig. 4 is a conversion block diagram of gain parameters provided in embodiment 1 of the present invention;

fig. 5 is a block diagram of transcoding the gain parameter provided in embodiment 2 of the present invention;

fig. 6 is a block diagram of another parameter transcoding provided in embodiment 2 of the present invention;

fig. 7 is a schematic diagram of a transcoding device provided in embodiment 3 of the present invention;

fig. 8 is another schematic diagram of a transcoding device according to embodiment 3 of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

The embodiment of the invention provides a transcoding method, which comprises the following contents: obtaining adaptive codebook gain g of current subframe in source code stream_pAnd a scale factor gamma of the fixed codebook gain_gc(ii) a According to the scaling factor alpha, to gamma_gcScaling to obtain alpha x gamma_gc(ii) a According to the quantization criterion of the target end, the method is used for the alpha multiplied by the gamma_gcAnd g_pAnd quantizing to realize transcoding. Gain scale factor gamma to fixed codebook by introducing scaling factor_gcAfter scaling, the new quantization criterion is adopted to search again in the gain code table of the target encoder to complete the transcoding of the gain parameters, thereby effectively reducing the transcoding count of the gain parametersCalculating complexity and improving the quality of the transcoded voice. The following describes the transcoding method provided by the embodiment of the present invention in detail, and the details are shown in the following embodiments.

Example 1

Referring to fig. 2, an embodiment of the present invention provides a transcoding method, which is described by taking the system block diagram provided in fig. 1 as an example, and details are as follows:

101: obtaining adaptive codebook gain g of current subframe in code stream sent by communication equipment A_pAnd a scale factor gamma of the fixed codebook gain_gcThe specific method comprises the following steps:

firstly, a source decoder receives and decodes a code stream sent by communication equipment A, and decodes gain index information from the code stream of the source decoder, and then searches a gain quantization code table of the source encoder by using the gain index information obtained by decoding to obtain a self-adaptive codebook gain g of a current subframe_pAnd a scale factor gamma of the fixed codebook gain_gc。

Wherein the scale factor γ of the fixed codebook gain involved in this step_gc＝g_c/g′_cWherein g is_cCurrent subframe fixed codebook gain value, g ', found for coding'_cFixing the codebook gain value for the current subframe obtained for the prediction, wherein,in the formula

The prediction item of the current subframe gain is obtained through a prediction algorithm; e_IFixing the total energy item of the codebook for the current subframe;

the average energy term of the codebook is fixed for the current subframe (where different CELP speech coding standards correspond to different ones).

102：Using the scaling factor α, for γ_gcScaling to obtain alpha x gamma_gc。

Wherein, α × γ_gcA scaling factor of a fixed codebook gain corresponding to the target communication device B; in which α is γ_gcBefore transcoding the gain parameter, the scaling factor gamma of the codebook gain is fixed_gcThe purpose of scaling by the scaling factor α is to compensate for the influence of amplitude distortion of the transcoded speech caused by the difference between the source-side (communication device a) speech coding standard and the target-side (communication device B) speech coding standard (specifically, the difference between the average energy items of the current subframe fixed codebooks).

103: for α × γ according to the quantization criterion_gcAnd g_pAnd carrying out quantization processing to realize transcoding of the gain parameters.

Wherein the gain g is based on the adaptive codebook gain in the target encoder_pScale factor gamma with fixed codebook_gcWhen a new quantization criterion is adopted to search in the gain code table of the target encoder again, the adaptive codebook gain g is obtained according to different speech coding standards_pScale factor gamma with fixed codebook_gcWhen quantization is performed, there may be different quantization requirements, for example, for g.729a mode speech coding standard, g needs to be quantized_pAnd gamma_gcJoint quantization is performed, and for the speech coding standard of AMR7.95kbit/s mode, g is required_pAnd gamma_gcIndependent quantization is performed, and accordingly, pair g is performed_pAnd gamma_gcThe following method can be adopted for quantization:

adaptive codebook gain g in target encoder_pScale factor gamma with fixed codebook_gcFor joint quantization, the following joint quantization criterion is adopted:

wherein β is g_pWeight factor (size of beta according to g)_pAnd gamma_gcThe range of values is chosen, usually beta is 1), in which g is defined as_pAnd gamma_gcRespectively obtaining the scale factors of the adaptive codebook gain and the fixed codebook gain of the current subframe obtained by the source-end decoder;

and

a kth code word in a code table when vector quantization is carried out on the gain for an encoder of the target end; n is the size of the code table in the target encoder, and the gain code table meeting the above formula and minimizing the value of E is obtained by searching the gain code table of the target encoder, namely, the gain code table closest to alpha multiplied by gamma is met simultaneously_gc、g_pK < th > code word

And

thereby achieving quantization of the gain parameter.

(II) adaptive codebook gain g in target encoder_pScale factor gamma with fixed codebook_gcFor independent quantization, the following independent quantization criteria are used:

wherein, γ_gcIs the scaling factor of the fixed codebook gain of the current subframe obtained by the source decoder;

then vector quantize kth in code table for fixed codebook scale factor for target encoding end₁A code word; n is a radical of₁Quantizing the size of the fixed codebook scale factor code table in the target encoder; alpha, gamma_gcA scaling factor of (d); g_pThe adaptive codebook gain of the current subframe obtained by the source decoder;

then it is the kth in the quantization adaptive codebook gain code table in the target encoder₂A code word; n is a radical of₂The size of the adaptive codebook gain codebook table in the target encoder is quantized.

Further, when adaptive codebook gain g is performed_pScale factor gamma with fixed codebook_gcWhen independently quantizing, the following absolute quantization criteria can be used according to the specific requirements of the transcoding system:

the specific meanings of the parameters of the above formula are the same as the above, and are not repeated.

In summary, the transcoding of the gain parameters in the transcoding process of different voice standards is successfully realized through the steps 101 to 103, the computational complexity of the transcoding of the gain parameters is reduced, and the quality of the transcoded voice is improved.

In the following, a detailed description is given specifically on how to determine the scaling factor α, and generally, when determining the scaling factor α, the training data can be obtained by two ways of obtaining through large sample statistics or obtaining through calculation for each sequence (the difference is that each sequence calculates α, or the large sample statistics α is used as a fixed α value for each sequence), and the steps of the two ways are the same, and referring to fig. 3, the specific steps provided by the embodiment of the present invention for determining the scaling factor α are as follows:

201, average energy of fixed codebook according to source speech coding standardMeasurement of

And average energy of fixed codebook of target-side speech coding standardDetermining an objective ideal value alpha of a scaling factor alpha_L；

For example, using a calculation formula,

as mentioned above, the average energies of the fixed codebooks are different for different CELP standards.

202 according to α_LDetermining the value range [ a1, a2] of the scaling factor alpha]Wherein, when determining the value range, the value range can be selected according to the system requirement, for example, when alpha is_LWhen the value is 1.6, the value range can be selected to be [1.5, 2]]The value range can also be selected to be [1, 2]]The embodiments of the present invention do not limit the specific manner and method used in determining the value range.

203, selecting a proper quantization criterion according to a quantization method of a gain parameter in a target encoder, acquiring alpha in a value range of the determined alpha according to the determined quantization criterion and a preset selection rule, performing a gain parameter transcoding training experiment on each acquired alpha, and testing to obtain MOS-LQO ((Mean Opinion Score-listing quality Objective) of the synthesized voice of each alpha, wherein the Score represents the quality of the synthesized voice, and the higher the Score is, the higher the quality of the corresponding synthesized voice is, and calculating the average segmentation energy of the synthesized voice of each alpha and the average segmentation energy of the original voice;

for example, the preset selection rule is to acquire elements in the value range [ a1, a2] of the scaling factor α at intervals of 0.05, and perform a gain parameter transcoding training experiment on each acquired element α.

204, selecting the optimal alpha value according to the MOS-LQO score of the synthesized voice of each alpha and the proximity degree of the average segmentation energy of the synthesized voice and the average segmentation energy of the original voice, and taking the selected alpha value as a fixed codebook gain scaling factor gamma_gcThe scaling factor value of (c). Different synthetic voices are generated by different scaling factors alpha, and a rule for selecting an optimal alpha value as the scaling factor alpha can be formulated according to specific needs of a system.

Referring to fig. 4, an embodiment of the present invention provides a gain parameter conversion block diagram, where only the adaptive codebook gain g needs to be obtained_pAnd a scale factor gamma of the fixed codebook_gcBy using a pair of gamma_gcScaling factor alpha to gamma obtained by transcoding training_gcAfter the scaling processing is performed, only the processed pairs of alpha x gamma are needed_gcAnd g_pAnd quantization processing is carried out, so that transcoding of the gain parameters is realized, the computational complexity of transcoding of the gain parameters is reduced, and the quality of transcoded voice is improved.

To further describe the transcoding method provided in the embodiment of the present invention, a specific example is described below, where a source end is a communication device a using an AMR7.95kbit/s mode in a CELP standard, a target end is a communication device B using a G..729a mode in the CELP standard, and when the communication device a wants to communicate with the communication device B, since the communication device a operates under different voice coding standards, in order to implement communication thereof, transcoding from the AMR7.95kbit/s mode to the G, 729a mode is required, and the transcoding process is usually performed by an intermediate device C located in a communication link between the communication device a and the communication device B:

(1) when linear spectrum LSP parameter transcoding is carried out, an interpolation method is adopted, and the following formula is adopted:

wherein,LSP parameter vector for AMR k frame;

LSP parameter vector of AMR (adaptive multi-rate) k-1 frame;is G, 729a frame 2k-1 LSP parameter vector;

and G, 729a LSP parameter vector of 2k frame.

(2) When the integer and fractional fundamental tone delay parameter transcoding is carried out, the integer fundamental tone value and the fractional fundamental tone value of each AMR subframe are directly transmitted to each G.729a subframe. In addition, it is necessary to determine whether the pitch of the AMR even subframe exceeds the pitch range of g.729a, and if so, the upper and lower boundaries of the pitch range of g.729a are taken.

(3) When transcoding the symbol and position parameters of the fixed codebook (i.e. algebraic codebook), decoding algebraic codebook information from AMR, then decoding algebraic codebook information by gray code, and then transmitting to each sub-frame of g.729a, and g.729a recodes the pulse position obtained by transmission and writes into the code stream.

(4) In the gain parameter transcoding, a large sample statistical approach is adopted, i.e. the training data is selected from the chinese speech in NTT (nippon telegraph and Telephone Corporation, japan teletelephony Corporation) database: the male and female speakers have 4 persons each, each speaker has 12 segments of voice with the length of 8 seconds, and the total voice length is 768 seconds. Wherein, the average energy of the fixed code book used in the AMR7.95kbit/s mode is 36dB, and the average energy of the fixed code book used in G.729a is 30 dB. The weighting factor β defaults to 1.

First, an objective ideal value α of the scaling factor α is determined_LThe following were used:

secondly, according to α_LDetermining scalingThe subjective value range of the factor alpha is because

So the scaling factor α ∈ [1, 2]]. Wherein,

meaning that the second digit after the decimal point in the number W is rounded up to get a decimal value that has only one digit after the decimal point.

And thirdly, selecting each alpha in the range according to a preset rule according to the subjective value range of the scaling factor alpha, and obtaining the value of the optimal alpha after performing a gain parameter transcoding training experiment on each obtained alpha, wherein for example, alpha is 1.3.

Finally, the adaptive codebook gain and the fixed codebook scale factor are jointly quantized in the G.729a mode, and a joint quantization rule is selected, so that the gain parameter transcoding is successfully realized. Wherein, the joint quantization reference formula is as follows:

wherein, g_pAnd gamma_gcThe scale factor is the adaptive codebook gain and the fixed codebook gain of the current subframe obtained by decoding in the AMR7.95kbit/s code stream;

and

then the kth code word in the code table is used for vector quantization of the gain for the G.729a coding end; in G.729a coding, the size of the code table used for gain quantization is 128, so N is 128, and the gain code table of the target coder is searched to obtain the code table which satisfies the above expression and minimizes the value of E, namely satisfies the closest value of alpha multiplied by gamma simultaneously_gc、g_pOf the kth code word

And

thereby achieving quantization.

The inventor adopts the transcoding method provided by the embodiment of the invention, and objective test experiments are carried out on 480 corpora to obtain the following results: when AMR7.95 is used for transcoding to G.729a, the average MOS-LQO score of the synthesized voice obtained by using the DTE method is 3.343035417, while the average MOS-LQO score of the synthesized voice obtained by using the transcoding method provided by the embodiment of the invention is 3.51400625.

In summary, the embodiments of the present invention provide a transcoding method, which introduces a scaling factor to a fixed codebook gain scaling factor γ_gcTranscoding is performed according to the determined quantization error criterion after scaling is performed, thereby greatly reducing the computational complexity of gain parameter transcodingThe transcoding method provided by the embodiment of the invention is suitable for transcoding codebook gain parameters among different CELP voice coding algorithms based on gain prediction vector quantization.

Example 2

The embodiment of the invention provides a transcoding method, which reduces the operation of requantization of a gain parameter transcoding part and further reduces the calculation complexity of the gain parameter part by using a mode of establishing a mapping code table, and comprises the following steps:

first, the gain quantized code table in the source encoder is mapped with the gain quantized code table in the target encoder according to the determined scaling factor α value and the selected quantization criterion.

When determining the value of the scaling factor α, the determination method provided in embodiment 1 may be similarly adopted, and this embodiment is still described by default of the value β as 1.

When the gain quantization code table in the source encoder is mapped with the gain quantization code table in the target encoder, the following mapping method may be adopted:

{\hat{G}}_{g} = Q_{g} [Q_{s}^{- 1} [{\hat{G}}_{s}]]

wherein, in the above formula

For gain indexing of source encoder；

Carrying out inverse quantization processing on the gain of the source encoder; q_gAfter the gain information obtained by decoding is scaled according to the determined scaling factor alpha, the quantization processing is carried out again according to the selected quantization standard in the target encoder;

the gain index value after re-quantization in the target encoder.

Then, all the code words in the source encoder are stored by the index values obtained after the mapping process, for example, a mapping code table is established and stored in a table form.

And finally, when the gain parameter is transcoded, transcoding the gain parameter can be realized by directly looking up the mapping code table.

Referring to fig. 5, an embodiment of the present invention provides a block diagram of gain parameter transcoding, where after a code stream of a source encoder is received, adaptive codebook gains and scale factors are decoded, and then a preset mapping code table is searched, so that transcoding of gain parameters is successfully achieved.

Referring to fig. 6, another block diagram of transcoding a gain parameter is provided in the embodiment of the present invention, and further, a person skilled in the art may also know that, by adopting the idea of introducing a mapping code table provided in this embodiment, a mapping code table corresponding to a bit stream is first set for a gain code stream of a source encoder and a gain code stream of a target encoder according to a determined scaling factor value and a selected quantization criterion, and then, during transcoding, table look-up mapping is directly performed on a code stream sent by the source encoder according to the mapping code table, see table 1, a mapping code table is provided, as shown in table 1, when the gain code stream of the source encoder is 100101, and when α is 1.4, looking up the table can obtain that the target encoder gain code stream corresponding to 100101 is 100011, so that transcoding of the gain parameter is successfully achieved.

TABLE 1

α	Source encoder gain code stream	Target encoder gain code stream
			1.4	100101	100011
2.0	100101	100100
			……	……	……

According to the transcoding method provided by the embodiment of the invention, the re-quantization operation in the gain parameter transcoding algorithm is omitted by searching the preset mapping code table, and the calculation complexity in the transcoding of the gain parameters is further reduced on the premise of improving the quality of the transcoded voice.

Example 3

Referring to fig. 7, an embodiment of the present invention provides a transcoding apparatus, including:

an obtaining module for obtaining the adaptive codebook gain of the current subframe in the source code streamg_pAnd a scale factor gamma of the fixed codebook gain_gc；

A scaling module for scaling the gamma acquired by the acquisition module according to the scaling factor alpha_gcScaling to obtain alpha x gamma_gc；

A quantization module for quantizing the alpha x gamma according to a quantization rule of the target end_gcAnd g_pAnd quantizing to realize transcoding.

Wherein, in order to determine the scaling factor α, the apparatus further comprises:

a scaling factor alpha obtaining module, configured to obtain a scaling factor alpha;

the scaling factor α obtaining module specifically includes:

a first obtaining unit, configured to determine an objective ideal value α of the scaling factor α according to average energy of fixed codebooks at the source end and the target end_L；

A second acquisition unit for acquiring alpha according to the alpha acquired by the first acquisition unit_LDetermining the value range of the scaling factor alpha;

the third acquisition unit is used for acquiring a test value for testing from the value range acquired by the second acquisition unit according to a preset selection rule; a test value of at least one;

and the fourth obtaining unit is used for obtaining the scaling factor alpha from the test value obtained by the third obtaining unit according to a preset test rule.

The third obtaining unit is specifically configured to sample, according to a preset interval, a value within a value range obtained by the second obtaining unit, and obtain a test value for testing;

the fourth obtaining unit is specifically used for obtaining the objective voice quality average score of the synthesized voice of the test value, the average segmentation energy of the synthesized voice and the average segmentation energy of the original voice; and obtaining a scaling factor alpha from the test value according to the objective voice quality average score of the synthesized voice, the average segmentation energy of the synthesized voice and the average segmentation energy of the original voice.

Due to the gain g of the adaptive codebook in different CELP speech coding standards_pAnd a scale factor gamma of the fixed codebook gain_gcWhen quantization is performed, different quantization methods, such as joint quantization, independent quantization, and the like, are adopted, and accordingly, the embodiment of the present invention provides a quantization module in a transcoding device, which may specifically be:

when target end is opposite to self-adaptive code book gain g_pAnd a scale factor gamma of the fixed codebook gain_gcWhen joint quantization is performed;

the embodiment of the invention provides the transcoding device, which further comprises:

a weight factor beta obtaining module for obtaining gain g according to the adaptive code book_pAnd a scale factor gamma of the fixed codebook gain_gcA weighting factor β is determined.

Accordingly, the number of the first and second electrodes,

the quantization module is specifically a quantization unit for performing a quantization on α × γ according to the gain vector code table of the acquisition target side_gcAnd g_pPerforming joint quantization to obtain quantization code words in a gain vector code table, and realizing transcoding of gain parameters, wherein when performing joint quantization, the quantization rule is as follows:

(II) when the target end is opposite to the gain g of the adaptive code book_pAnd a scale factor gamma of the fixed codebook gain_gcWhen independent quantization is performed;

the quantization module is specifically a quantization unit for respectively quantizing the alpha x gamma values according to a target fixed codebook scale factor code table and a self-adaptive codebook gain code table_gcAnd g_pAnd quantizing to obtain a quantized fixed code word in a quantized scale factor code table and a quantized adaptive code word in an adaptive code book gain code table, and realizing transcoding of gain parameters, wherein the quantization rule is as follows:

further, according to the specific need of transcoding, the quantization module is specifically an absolute quantization unit, and is configured to separately perform scaling factor code table and adaptive code gain code table for the target fixed code book, respectively for the α × γ_gcAnd g_pAbsolute quantization is carried out, a quantized fixed code word in a quantization scale factor code table and a quantized adaptive code word in an adaptive code book gain code table are obtained, and transcoding of gain parameters is achieved, wherein the absolute quantization rule is as follows:

in order to reduce the operation of re-quantization of the gain parameter transcoding portion and further reduce the computational complexity of the gain parameter portion, referring to fig. 8, the transcoding apparatus provided in the embodiment of the present invention may further include:

the mapping code table generating module is used for mapping the source end gain quantization code table and the gain quantization code table in the target encoder according to the scaling factor alpha and the weight factor beta to obtain a mapping code table;

a processing module for obtaining the adaptive codebook gain g of the current subframe in the source code stream_pAnd a scale factor gamma of the fixed codebook gain_gcAnd searching a mapping code table to realize transcoding.

The mapping code table generating module specifically comprises:

the first processing unit performs inverse quantization processing on the gain index of the source-end encoder according to the gain index of the source-end encoder;

the second processing unit is used for scaling the gain obtained after the dequantization processing of the first processing unit according to the scaling factor alpha;

the third processing unit is used for quantizing the gain scaled by the second processing unit according to the quantization rule of the target end;

and the fourth processing unit is used for acquiring the gain index value quantized by the third processing unit, and the quantized gain index value forms a mapping code table.

Those skilled in the art may also know that, similarly, by adopting the idea of introducing the mapping code table provided in the foregoing embodiment, the transcoding device provided in the embodiment of the present invention may further include:

the mapping code table generating module is used for mapping the code stream of the source-end encoder and the code stream of the target-end encoder according to the scaling factor alpha, the weighting factor beta and the quantization rule of the target end to obtain a mapping code table;

and the processing module is used for acquiring the code stream of the source end, searching the mapping code table and realizing transcoding.

In summary, the transcoding apparatus provided in the embodiment of the present invention introduces the scaling factor to the fixed codebook gain scaling factor γ_gcAfter scaling, searching in the gain code table of the target encoder again by adopting a new quantization criterion to complete transcoding of the gain parameters, thereby effectively reducing the computational complexity of transcoding the gain parameters and improving the quality of transcoded voice.

Some steps in the embodiments of the present invention may be implemented by software, and the corresponding software program may be stored in a readable storage medium, such as an optical disc or a hard disk.

The above description is only exemplary of the present invention and should not be construed as limiting the present invention, and any modification, equivalent replacement, or improvement made by those skilled in the art without departing from the principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method of transcoding, the method comprising:

2. The transcoding method of claim 1, wherein the method further comprises: obtaining a scaling factor α, specifically:

determining an objective ideal value alpha of the scaling factor alpha according to the average energy of the fixed code books of the source end and the target end_L；

According to the alpha_LDetermining the value range of the scaling factor alpha;

according to a preset selection rule, obtaining a test value for testing from the value range; the test value is at least one;

and obtaining a scaling factor alpha from the test value according to a preset test rule.

3. The transcoding method of claim 2, wherein the preset selection rule specifically is:

and sampling the numerical value in the value range according to a preset interval.

4. The transcoding method of claim 2, wherein the obtaining of the scaling factor α from the test value according to a preset test rule specifically includes:

acquiring objective voice quality average score of the synthesized voice of the test value, average segmentation energy of the synthesized voice and average segmentation energy of the original voice;

and obtaining a scaling factor alpha from the test value according to the objective voice quality average score of the synthesized voice, the average segmentation energy of the synthesized voice and the average segmentation energy of the original voice.

5. The transcoding method of claim 1, wherein said separately for said axy γ_gcAnd g_pThe quantization may be performed as follows:

respectively to the alpha x gamma_gcAnd g_pAbsolute quantization is performed.

6. The transcoding method of claim 1, wherein the obtaining of the weighting factor β specifically comprises:

according to the adaptive codebook gain g_pAnd a scale factor gamma of the fixed codebook gain_gcAnd determining the weight factor beta.

7. The transcoding method of claim 1, wherein the method further comprises:

mapping the source end gain quantization code table and a gain quantization code table in a target encoder to obtain a mapping code table according to the scaling factor alpha and the weight factor beta;

obtaining adaptive codebook gain g of current subframe in source code stream_pAnd a scale factor gamma of the fixed codebook gain_gcAnd searching the mapping code table to realize the transcoding of the gain parameters.

8. The transcoding method of claim 7, wherein the mapping the source-side gain quantization code table and the gain quantization code table in the target encoder to obtain a mapping code table specifically comprises:

carrying out inverse quantization processing on the gain index of the source-end encoder according to the gain index of the source-end encoder;

according to the scaling factor alpha, scaling the gain obtained after the inverse quantization processing;

quantizing the scaled gain according to a quantization rule of a target end;

and obtaining a quantized gain index value, wherein the quantized gain index value forms a mapping code table.

9. The transcoding method of claim 1, wherein the method further comprises:

mapping the code stream of the source-end encoder and the code stream of the target-end encoder according to the scaling factor alpha, the weighting factor beta and the quantization rule of the target end to obtain a mapping code table;

and acquiring the code stream of the source end, searching the mapping code table and realizing the transcoding of the gain parameters.

10. A transcoding device, the device comprising:

11. The transcoding device of claim 10, wherein the device further comprises:

and the scaling factor alpha acquisition module is used for acquiring the scaling factor alpha.

12. The transcoding device of claim 11, wherein the scaling factor α obtaining module specifically comprises:

A second obtaining unit for obtaining the alpha according to the alpha obtained by the first obtaining unit_LDetermining the value range of the scaling factor alpha;

the third acquisition unit is used for acquiring a test value for testing from the value range acquired by the second acquisition unit according to a preset selection rule; the test value is at least one;

13. The transcoding device of claim 12, wherein the third obtaining unit is specifically configured to sample, according to a preset interval, a value within the value range obtained by the second obtaining unit, and obtain a test value for testing;

the fourth obtaining unit is specifically configured to obtain an objective voice quality average score of the synthesized voice of the test value, an average segmentation energy of the synthesized voice, and an average segmentation energy of the original voice; and obtaining a scaling factor alpha from the test value according to the objective voice quality average score of the synthesized voice, the average segmentation energy of the synthesized voice and the average segmentation energy of the original voice.

14. The transcoding device of claim 10, wherein the quantization module is specifically an absolute quantization unit configured to perform the respective scaling for the α x γ according to the target fixed codebook scale factor codebook and the adaptive codebook gain codebook_gcAnd g_pAnd carrying out absolute quantization to obtain a quantized fixed code word in the quantization scale factor code table and a quantized adaptive code word in the adaptive code book gain code table, and realizing transcoding.

15. The transcoding device of claim 10, wherein the device further comprises:

a processing module for obtaining the adaptive codebook gain g of the current subframe in the source code stream_pAnd a scale factor gamma of the fixed codebook gain_gcAnd searching the mapping code table to realize transcoding.

16. The transcoding device of claim 15, wherein the map code table generating module specifically comprises:

the first processing unit is used for carrying out inverse quantization processing on the gain index of the source-end encoder according to the gain index of the source-end encoder;

the third processing unit is used for quantizing the gain scaled by the second processing unit according to a quantization rule of a target end;

17. The transcoding device of claim 10, wherein the device further comprises:

the mapping code table generating module is used for mapping the code stream of the source end encoder and the code stream of the target end encoder according to the scaling factor alpha, the weighting factor beta and the quantization rule of the target end to obtain a mapping code table;