CN101572093A - Method and device for transcoding - Google Patents

Abstract

Translated fromChinese

本发明实施例提供了一种转码方法和装置。所述方法包括：获取源端码流中的当前子帧的自适应码书增益g_p和固定码书增益的比例因子γ_gc；根据缩放因子α，对所述γ_gc进行处理得到α×γ_gc；根据目标端的量化规则，对所述α×γ_gc和g_p进行量化，实现转码。所述装置包括：获取模块、缩放模块和量化模块。本发明通过引入缩放因子对固定码书增益比例因子γ_gc进行缩放后，采用新的量化准则在目标编码器的增益码表中进行搜索，完成增益参数的转码，从而有效地降低增益参数转码的计算复杂度，提高转码语音的质量。

Embodiments of the present invention provide a transcoding method and device. The method includes: obtaining the adaptive codebook gain_gp of the current subframe in the source stream and the scaling factor γ_gc of the fixed codebook gain; according to the scaling factor α, processing the γ_gc to obtain α×γ_gc : Quantize the α×γ_gc and_gp according to the quantization rules of the target end to realize transcoding. The device includes: an acquisition module, a scaling module and a quantization module. In the present invention, after scaling the fixed codebook gain scale factor γ_gc by introducing a scaling factor, a new quantization criterion is used to search in the gain code table of the target encoder to complete the transcoding of the gain parameters, thereby effectively reducing the transcoding of the gain parameters. The computational complexity of the code is improved, and the quality of the transcoded voice is improved.

Description

A kind of code-transferring method and device

Technical field

The present invention relates to the encoding and decoding field, particularly a kind of code-transferring method and device.

Background technology

Based on CELP (Code-Excited Linear Prediction, Code Excited Linear Prediction) voice coding is current VoIP (Voice over Internet Protocol, the networking telephone) and one of the core technology in the mobile communication system, for compatibility and the intercommunication that realizes communication facilities between the different suppliers, need carry out the transcoding work of different CELP voice codings standard room.

At present, when carrying out the transcoding of different CELP voice codings standard room, the code-transferring method of the extensive different CELP voice codings standard room that is based on the novel gain switch technology of adopting, this method is transmitted the transcoding of realizing between the different coding standard respectively after necessarily handling by the parameter that will be transmitted in the code stream.The parameter that relates to has: line spectral frequencies, integer and mark pitch delay, fixed code book symbol and location parameter and gain parameter, referring to Fig. 1, this synoptic diagram based on the trans-coding system of the different CELP voice codings standard room of parameter is provided, wherein, communication facilities A is operated under the different CELP voice coding standards with B, parametric solution among the communication facilities C in the communication link between communication facilities A and communication facilities B/scrambler A receives the laggard line parameter decoding of code stream of communication facilities A, get access to relevant parameters, then by LSP (Line SpectrumPairs, the linear spectral parameter) transcoding has been realized the transcoding to wherein line spectral frequencies parameter, Pitch (fundamental tone) transcoding has been realized the transcoding to integer and mark pitch delay parameter, ACELP (Algorithm Code-Excited Linear Prediction, the algebraic codebook Code Excited Linear Prediction) transcoding has been realized the transcoding to fixed code book symbol and location parameter, Gain (gain) transcoding has been realized the transcoding to gain parameter, parameter coder/decoder B obtains through each parameter behind the transcoding, after encoding, code stream behind the coding is sent to opposite end communication facilities B, thereby realized the work of different CELP voice coding standard transcoding, successfully realized being operated in the communication between the substandard communication facilities of different CELP voice codings.

Wherein, when the Gain gain parameter being carried out the transcoding realization, can adopt direct transcoding DTE (Decode Then Encode, decode earlier and afterwards encode) method, subject encoder is by realizing the gain parameter transcoding asking for quantization operation of gain parameter: at first, utilize and ask for the self-adapting code book gain g that formula gets access to current subframe_pWith fixed codebook gain g_c, then,, adopt square error item e according to the combined optimization process of Optimum Excitation_jAs search criteria, the gain quantization code table in the ferret out scrambler makes e_jMinimum g_pAnd g_cBe the yield value after the final quantification; At last, the index value of optimum gain quantized value in the gain code table write in the code stream of subject encoder, promptly finished the transcoding operation of gain parameter.But, when adopting this DTE method to realize the gain parameter transcoding, owing to obtaining g_pAnd g_cThe time, need recover the self-adapting code book vector and the algebraic-codebook vector of current subframe, and need to recover the code book amount of gain simultaneously, carry out second quantization then, these operations have increased the complexity of calculating greatly, in addition because twice gain quantization causes the decline of transcoding voice quality.

In addition, when gain parameter being carried out the transcoding realization, can also adopt the self-adapting code book gain g that directly source encoder is passed over_pWith algebraic-codebook gain g_cCarry out the method for transcoding, this method has been omitted in the subject encoder self-adapting code book gain g_pWith algebraic-codebook gain g_cThe process of asking for, carrying out gain parameter when quantizing, adopt following square error item as search criteria: E (L)=[ex1 (i)-ex2 (i, L)]², (i=0,1 ..., N), wherein, N is a subframe lengths,

Wherein, ex1 be parametric excitation signal ex1 (i, L)=g_p(L) * v (i)+g_c(L) * c (i), (i=0,1 ... N), wherein, g_p(L) the self-adapting code book yield value for passing over from source encoder, g_cBe the algebraic-codebook yield value that passes over from source encoder, v (i), c (i) are respectively self-adapting code book vector and algebraic-codebook vector; Ex2 be pumping signal ex2 synthetic behind the transcoding (i, L)=g_p' (L) * v (i)+g_c' (L) * c (i), (i=0,1 ..., N), wherein, g_p' (L) be the self-adapting code book yield value of L code word in the subject encoder, g_c' (L) be the algebraic-codebook yield value of L code word in the subject encoder.

By search, make the minimum g of E in the following formula (L) to gain quantization code table in the subject encoder_p' (L) and g_c' (L) be yield value after the final quantification, the index value of optimum gain quantized value in the gain code table write in the code stream of subject encoder, promptly finished the transcoding operation of gain parameter.But, when adopting this method to realize the gain parameter transcoding, need to keep the prediction of gain process.Carrying out gain parameter when quantizing, need a large amount of calculating guarantee that signal transmits energy conservation, increase the complexity of calculating, and because the difference of different CELP standard gain quantization code books, influenced the quality of transcoding voice.

Summary of the invention

For in the transcoding process of different CELP voice codings standard room, reduce the computation complexity of gain parameter transcoding, improve the quality of transcoding voice, the embodiment of the invention provides a kind of code-transferring method and device.Described technical scheme is as follows:

On the one hand, provide a kind of code-transferring method, described method comprises:

Obtain the self-adapting code book gain g of the current subframe in the end code stream of source_pScale factor γ with fixed codebook gain_Gc

According to zoom factor α, to described γ_GcHandle and obtain α * γ_Gc

According to the quantizing rule of destination end, to described α * γ_GcAnd g_pQuantize, realize transcoding.

On the other hand, provide a kind of transcoding device, described device comprises:

Acquisition module is used for obtaining the self-adapting code book gain g of the current subframe of source end code stream_pScale factor γ with fixed codebook gain_Gc

Zoom module is used for according to zoom factor α, the γ that described acquisition module is got access to_GcCarry out convergent-divergent and handle, obtain α * γ_Gc

Quantization modules is used for the quantizing rule according to destination end, to described α * γ_GcAnd g_pQuantize, realize transcoding.

The beneficial effect of the technical scheme that the embodiment of the invention provides is:

By introducing zoom factor to fixed codebook gain scale factor γ_GcAfter carrying out convergent-divergent, adopt new quantification criterion in the gain code table of subject encoder, to search for, finish the transcoding of gain parameter, thereby reduce the computation complexity of gain parameter transcoding effectively, improve the quality of transcoding voice.

Description of drawings

Fig. 1 is the synoptic diagram based on the trans-coding system of the different CELP voice codings standard room of parameter that prior art provides;

Fig. 2 is the process flow diagram of the code-transferring method that provides of the embodiment of the invention 1;

Fig. 3 is the method flow diagram of definite zoom factor α of providing of the embodiment of the invention 1;

Fig. 4 is the gain parameter conversion block diagram that the embodiment of the invention 1 provides;

Fig. 5 is the gain parameter transcoding block diagram that the embodiment of the invention 2 provides;

Fig. 6 is that the embodiment of the invention 2 provides another parameter transcoding block diagram;

Fig. 7 is the transcoding schematic representation of apparatus that the embodiment of the invention 3 provides;

Fig. 8 is another synoptic diagram of the transcoding device that provides of the embodiment of the invention 3.

Embodiment

For making the purpose, technical solutions and advantages of the present invention clearer, embodiment of the present invention is described further in detail below in conjunction with accompanying drawing.

A kind of code-transferring method that the embodiment of the invention provides thes contents are as follows: the self-adapting code book gain g that obtains the current subframe in the end code stream of source_pScale factor γ with fixed codebook gain_GcAccording to zoom factor α, to γ_GcCarry out convergent-divergent and handle, obtain α * γ_GcAccording to the quantification criterion of destination end, to α * γ_GcAnd g_pQuantize, realize transcoding.By introducing zoom factor to fixed codebook gain scale factor γ_GcAfter carrying out convergent-divergent, adopt new quantification criterion in the gain code table of subject encoder, to search for again, finish the transcoding of gain parameter, thereby reduce the computation complexity of gain parameter transcoding effectively, improve the quality of transcoding voice.The method of the transcoding that provides at the embodiment of the invention is elaborated below, sees following each embodiment for details.

Embodiment 1

Referring to Fig. 2, the embodiment of the invention provides a kind of code-transferring method, and the system chart that provides with Fig. 1 is that example describes, and detailed content is as follows:

101: the self-adapting code book gain g of the current subframe in the code stream that the obtaining communication device A sends over_pScale factor γ with fixed codebook gain_Gc, concrete grammar is as follows:

At first, source demoder received communication device A transmitted stream is also decoded, and from the demoder code stream of source, gain index information is decoded, utilize the gain quantization code table of the gain index information searching source encoder that decoding obtains again, get access to the self-adapting code book gain g of current subframe_pScale factor γ with fixed codebook gain_Gc

The scale factor γ of the fixed codebook gain that wherein, relates in this step_Gc=g_c/ g '_c, wherein, g_cBe the current subframe fixed codebook gain value that coding is tried to achieve, g '_cThe current subframe fixed codebook gain value that obtains for prediction, wherein,$g_c^{\′}={10}^{0.05(\tilde{E}\left(n\right)+\stackrel{\‾}{E}-E_I)},$In this formula

Be the prediction term that obtains by prediction algorithm to current subframe gain; E_IBe the total energy term of current subframe fixed code book; E is the average energy quantifier (wherein, different CELP voice coding standard correspondence is different) of current subframe fixed code book.

102: utilize zoom factor α, to γ_GcCarry out convergent-divergent and handle, obtain α * γ_Gc

Wherein, α * γ_GcScale factor for the fixed codebook gain of target communication device B correspondence; Above-mentioned, α is γ_GcZoom factor, before gain parameter is carried out transcoding, with the scale factor γ of fixed codebook gain_GcThe purpose that adopts zoom factor α to carry out convergent-divergent is in order to remedy different (they specifically being because the average energy quantifier difference of current subframe fixed code book separately) of source end (communication facilities A) voice coding standard and destination end (communication facilities B) voice coding standard, and the influence of the amplitude distortion of the transcoding voice that cause.

103: according to quantizing criterion, to α * γ_GcAnd g_pCarry out quantification treatment, realize transcoding gain parameter.

Wherein, according in the subject encoder to self-adapting code book gain g_pWith fixed code book scale factor γ_GcQuantization method, when adopting new quantification criterion in the gain code table of subject encoder, to search for again, because different voice coding standard is to self-adapting code book gain g_pWith fixed code book scale factor γ_GcWhen quantizing, can there be different quantification demands, for example, for the voice coding standard of G..729a pattern, need be to g_pAnd γ_GcThe associating quantification, and for the voice coding standard of AMR7.95kbit/s pattern, then need g_pAnd γ_GcCarry out independent quantification, so, correspondingly, carrying out g_pAnd γ_GcQuantification the time, can be in the following way:

(1) self-adapting code book gain g in subject encoder_pWith fixed code book scale factor γ_GcWhen quantizing, adopt the following quantification criterion of uniting for uniting:

$E=\underset{0\≤k\≤N}{\arg \min }\{{(\mathrm{\α}\×{\mathrm{\γ}}_{\mathrm{gc}}-{\mathrm{\γ}}_{\mathrm{gc}}^k)}^2+\mathrm{\β}{(g_p-g_p^k)}^2\},$

Wherein, β is g_p(size of β is according to g for weight factor_pAnd γ_GcThe scope of value is selected, and the β value is 1 usually), g in the following formula_pAnd γ_GcBe respectively the self-adapting code book gain of the current subframe that source end demoder obtains and the scale factor of fixed codebook gain; g_p^kAnd γ_Gc^kK code word when vector quantization is carried out in gain in the code table for the scrambler of destination end; N is the size of code table in the subject encoder, satisfies following formula and makes E value minimum by searching the gain code table of subject encoder, obtaining, and promptly satisfies near α * γ simultaneously_Gc, g_pK code word γ_Gc^kAnd g_p^kThereby, realize the quantification of gain parameter.

(2) self-adapting code book gain g in subject encoder_pWith fixed code book scale factor γ_GcFor carrying out independence when quantizing, adopts following independent quantification criterion:

$E_1=\underset{0\≤k_1\≤N_1}{\arg \min }{(\mathrm{\α}\×{\mathrm{\γ}}_{\mathrm{gc}}-{\mathrm{\γ}}_{\mathrm{gc}}^{k_1})}^2$

$E_2=\underset{0\≤k_2\≤N_2}{\arg \min }{(g_p-g_p^{k_2})}^2$

Wherein, γ_GcIt is the scale factor of the fixed codebook gain of the current subframe that obtains of source end demoder;Then the fixed code book scale factor is carried out k in the vector quantization code table for the target code end₁Individual code word; N₁For quantizing the size of fixed code book scale factor code table in the subject encoder; α is γ_GcZoom factor; g_pIt is the self-adapting code book gain of the current subframe that obtains of source end demoder;

Then be the k that quantizes in the subject encoder in the self-adapting code book gain code table₂Individual code word; N₂For quantizing the size of self-adapting code book gain code table in the subject encoder.

Further, when carrying out self-adapting code book gain g_pWith fixed code book scale factor γ_GcDuring independent the quantification, can also use following absolute quantification criterion according to the real needs of trans-coding system:

$E_1^{\′}=\underset{0\≤k_1\≤N_1}{\arg \min }|\mathrm{\α}\×{\mathrm{\γ}}_{\mathrm{gc}}-{\mathrm{\γ}}_{\mathrm{gc}}^{k_1}|;$$E_2^{\′}=\underset{0\≤k_2\≤N_2}{\arg \min }|g_p-g_p^{k_2}|,$Wherein, the concrete implication of each parameter of above-mentioned formula is the same, repeats no more.

To sum up, successfully realized the transcoding of the gain parameter in the transcoding process that different token sounds is carried out, reduced the computation complexity of gain parameter transcoding, improved the quality of transcoding voice by above-mentionedsteps 101 to 103.

Below, specifically how to determine that at above-mentioned zoom factor α is described in detail, usually when definite this zoom factor α, the data of training can obtain (so difference is all calculated α in each sequence by adopting the large sample statistics to obtain or each sequence being calculated the dual mode that obtains, still large sample statistics α is as the fixedly α value of each sequence), two kinds of method steps are identical, and referring to Fig. 3, the concrete steps of definite zoom factor α that the embodiment of the invention provides the contents are as follows:

201, according to the average energy E2 of the fixed code book of the average energy E1 of the fixed code book of source end voice coding standard and destination end voice coding standard, determine the objective ideal value α of zoom factor α_L

For example, adopt computing formula,${\mathrm{\α}}_L=\frac{{10}^{0.05\×\left(\stackrel{\‾}{E}1\right)}}{{10}^{0.05\×\left(\stackrel{\‾}{E}2\right)}},$E1 ≠ E2, as mentioned before, the average energy of the fixed code book of different CELP standard correspondences is inequality.

202, according to α_LDetermine the span [a1, a2] of zoom factor α, wherein, when determining this span, can choose, for example, work as α according to system's needs_LValue is 1.6 o'clock, can select this span to be [1.5,2], can also select this span to be [1,2], and the embodiment of the invention is not limited in concrete mode and the method that is adopted when determining this span.

203, select suitable quantification criterion according to the quantization method to gain parameter in the subject encoder, according to determined quantification criterion, get access to the α in the span of determined α according to default selective rule, each α that obtains is carried out gain parameter transcoding training experiment, test obtains MOS-LQO ((the Mean Opinion Score-ListeningQuality Objective of the synthetic speech of each α, objective voice quality average score), wherein, this mark has been represented the quality of synthetic speech, and the quality of the high more corresponding synthetic speech of mark is high more; And calculate the average segmentation energy of synthetic speech of each α and the average segmentation energy of raw tone;

For example, default selective rule be to serve as that the interval obtains element in the span [a1, a2] of zoom factor α with 0.05, and each the element α that gets access to is carried out gain parameter transcoding training experiment.

204, divide and the average segmentation energy of synthetic speech and the degree of closeness of the average segmentation energy of raw tone by MOS-LQO according to the synthetic speech of each α, select the α value of optimum, and this α value that will select is as fixed codebook gain scale factor γ_GcValues of zoom factor.Wherein, different zoom factor α can produce different synthetic speechs, can formulate according to the concrete needs of system and select the rule of optimum α value as zoom factor α, and the embodiment of the invention does not limit the formulation mode and the content of this concrete rule.

Referring to Fig. 4, the embodiment of the invention provides a kind of gain parameter conversion block diagram, wherein, only need obtain self-adapting code book gain g_pScale factor γ with fixed code book_Gc, adopt γ_GcCarrying out transcoding trains the zoom factor α that obtains to γ_GcCarried out after convergent-divergent handles, only need to after handling to α * γ_GcAnd g_pCarry out quantification treatment, thereby realized transcoding, reduced the computation complexity of gain parameter transcoding, improved the quality of transcoding voice gain parameter.

For the method for the transcoding that further embodiment of the invention provided describes, set forth with a concrete example below, wherein, the source end is for adopting the communication facilities A of the AMR 7.95kbit/s pattern in the CELP standard, destination end is for adopting the communication facilities B of the G..729a pattern in the CELP standard, when communication facilities A desire realizes communicating by letter with communication facilities B, because it is operated under the different voice coding standards, in order to realize its communication, need be by AMR 7.95kbit/s pattern to G, the transcoding of 729a pattern, this transcoding process are finished by the intermediate equipment C of the communication link that is arranged in communication facilities A and communication facilities B usually:

(1) when carrying out linear spectral LSP parameter transcoding, adopt the method for interpolation, as shown in the formula:

${\mathrm{\Ω}}_{2k-1}^{729a}=0.5{\mathrm{\Ω}}_{k-1}^{\mathrm{AMR}}+0.5{\mathrm{\Ω}}_k^{\mathrm{AMR}}$

${\mathrm{\Ω}}_{2k}^{729a}={\mathrm{\Ω}}_k^{\mathrm{AMR}}$

Wherein, Ω_k^AMRLSP parameter vector for AMR k frame; Ω_K-1^AMRLSP parameter vector for AMR k-1 frame; Ω_2k-1^729aBe G, the LSP parameter vector of 729a 2k-1 frame; Ω_2k^729aBe G, the LSP parameter vector of 729a 2k frame.

(2) when carrying out integer and mark pitch delay parameter transcoding, directly basis of integer value and the mark pitch value with each subframe of AMR passes to G.729a each subframe.Other need judge that whether the fundamental tone of AMR even number subframe surpasses fundamental tone scope G.729a, if exceed, then gets the up-and-down boundary of fundamental tone scope G.729a.

(3) when carrying out fixed code book (being algebraic-codebook) symbol, location parameter transcoding, need carry out earlier being passed to G.729a each subframe again behind the gray code decoder after AMR decodes algebraic-codebook information, G.729a the pulse position that transmission is obtained carries out writing code stream behind the recompile.

(4) when carrying out the gain parameter transcoding, adopt the mode of large sample statistics, be that training data is selected from NTT (NipponTelegraph and Telephone Corporation, NTT Co., Ltd.) Chinese speech in the database: each 4 people of men and women speaker, 8 seconds voice of each speaker 12 segment length, voice length is 768 seconds altogether.Wherein, the average energy of used fixed code book is 36dB under the AMR7.95kbit/s pattern, and the average energy of used fixed code book is 30dB in G.729a.Weight factor β is defaulted as 1.

At first, determine the objective ideal value α of zoom factor α_LAs follows:

${\mathrm{\α}}_L=\frac{{10}^{0.05\×\left(36\right)}}{{10}^{0.05\×\left(30\right)}}={10}^{0.3}=1.99526$

Secondly, according to α_LDetermine the subjective span of zoom factor α, because

So zoom factor α ∈ [1,2].Wherein,

Expression with the radix point among the digital W after second round up to obtain having only one fractional value behind the radix point.

Once more,, select each α in this scope according to preset rules according to the subjective span of zoom factor α, each α that obtains is carried out gain parameter transcoding training experiment after, obtain the value of optimum α, for example, α=1.3.

At last, owing to G.729a in the pattern quantification is united in self-adapting code book gain and fixed code book scale factor, select the associating quantizing rule, thereby successfully realized the gain parameter transcoding.Wherein, it is as follows with reference to formula to unite quantification:

$E=\underset{0\≤k\≤128}{\arg \min }\{{(\mathrm{\α}\×{\mathrm{\γ}}_{\mathrm{gc}}-{\mathrm{\γ}}_{\mathrm{gc}}^k)}^2+\mathrm{\β}{(g_p-g_p^k)}^2\}$

Wherein, g_pAnd γ_GcBe the self-adapting code book gain of the current subframe that decoding obtains in the AMR7.95kbit/s code stream and the scale factor of fixed codebook gain; g_p^kAnd γ_Gc^kThen be k code word in code table when G.729a coding side carries out vector quantization to gain; In coding G.729a is 128 for the used code table size of gain quantization, so N gets 128, satisfies following formula and makes E value minimum by searching the gain code table of subject encoder, obtaining, and promptly satisfies near α * γ simultaneously_Gc, g_pThe γ of k code word_Gc^kAnd g_p^kThereby, realized quantification.

The code-transferring method that the inventor adopts the embodiment of the invention to provide, by 480 language material objective examinations are tested, obtain following result: at AMR7.95 when G.729a carrying out transcoding, the average MOS-LQO of the synthetic speech that use DTE method obtains must be divided into 3.343035417, and the average MOS-LQO of the synthetic speech that the code-transferring method that utilizes the embodiment of the invention to provide obtains must be divided into 3.51400625, as can be seen, the code-transferring method that the embodiment of the invention provides not only can reduce the computation complexity of gain parameter transcoding, can also improve the quality of transcoding voice greatly.

In sum, the embodiment of the invention provides a kind of code-transferring method, by introducing zoom factor to fixed codebook gain scale factor γ_GcCarry out carrying out transcoding according to determined quantization error criterion behind the convergent-divergent, thereby reduce the computation complexity of gain parameter transcoding greatly, can also improve the quality of transcoding voice greatly, the code-transferring method that the embodiment of the invention provides is applicable to any situation based on the code book gain parameter transcoding between the different CELP speech coding algorithms of prediction of gain vector quantization.

Embodiment 2

The embodiment of the invention provides a kind of code-transferring method, utilizes the mode of setting up the mapping code table, reduces the operation of the re-quantization of gain parameter transcoding part, further reduces the computation complexity of gain parameter part, and method thes contents are as follows:

At first, shine upon according to the zoom factor α value of determining and selected quantification criterion code table gain quantization in the code table of gain quantization in the source encoder and the subject encoder.

Wherein, when definite zoom factor α value, definite method that can similarly adopt embodiment 1 to provide, present embodiment still is defaulted as 1 with the β value and describes.

Wherein, when the gain quantization code table shines upon in gain quantization code table and the subject encoder in the source encoder, can adopt following mapping method:

${\hat{G}}_g=Q_g\left[Q_s^{-1}\right[{\hat{G}}_s\left]\right]$

Wherein, in the following formula

Gain index for source encoder; Q_s^-1For source encoder gain inverse quantization is handled; Q_gFor after the gain information that decoding is obtained according to the zoom factor α that determines carries out convergent-divergent, carry out quantification treatment again according to the quantification criterion in the selected subject encoder;For carrying out the gain index value behind the re-quantization in the subject encoder.

Then, the index value that obtains after code words all in the source encoder handled by above-mentioned mapping is preserved, and for example, sets up the mapping code table, preserves with the form of form.

At last, when carrying out the gain parameter transcoding, can realize transcoding to gain parameter by directly looking into above-mentioned mapping code table.

Referring to Fig. 5, the embodiment of the invention provides a kind of gain parameter transcoding block diagram, receive the code stream of source encoder after, solve self-adapting code book gain and scale factor after, thereby successfully realized transcoding by searching default mapping code table to gain parameter.

Referring to Fig. 6, the embodiment of the invention provides another block diagram of gain parameter transcoding, further, those skilled in the art can also be known, the thought that code table is shone upon in the introducing of adopting present embodiment to provide, can also for, at first the gain code stream of source encoder is provided with the corresponding mapping code table of bit stream with the gain code stream of subject encoder according to the values of zoom factor of determining and selected quantification criterion, then, when carrying out transcoding, directly to the source encoder transmitted stream according to the mapping of tabling look-up of this mapping code table, referring to table 1, a kind of mapping code table is provided, as shown in table 1, when the gain code stream of source encoder is 100101, look into this table and can obtain under the situation of α=1.4, the subject encoder gain code stream of 100101 correspondences is 100011, has so just successfully realized the transcoding to gain parameter.

Table 1

α	Source encoder gain code stream	Subject encoder gain code stream
			1.4	100101	100011
2.0	100101	100100
			......	......	......

The code-transferring method that the embodiment of the invention provides, by searching default mapping code table, omit the re-quantization operation in the gain parameter transcoding algorithm, can realize under the prerequisite that the transcoding voice quality improves the computation complexity when reducing the gain parameter transcoding further.

Embodiment 3

Referring to Fig. 7, the embodiment of the invention provides a kind of transcoding device, and this device comprises:

Acquisition module is used for obtaining the self-adapting code book gain g of the current subframe of source end code stream_pScale factor γ with fixed codebook gain_Gc

Zoom module is used for the γ that acquisition module is got access to according to zoom factor α_GcCarry out convergent-divergent and handle, obtain α * γ_Gc

Quantization modules is used for the quantification criterion according to destination end, to α * γ_GcAnd g_pQuantize, realize transcoding.

Wherein, in order to determine above-mentioned zoom factor α, device also comprises:

Zoom factor α acquisition module is used to obtain zoom factor α;

Wherein, above-mentioned zoom factor α acquisition module specifically comprises:

First acquiring unit is used for the average energy according to the fixed code book of source end and destination end, determines the objective ideal value α of zoom factor α_L

Second acquisition unit is used for the α that obtains according to first acquiring unit_L, determine the span of zoom factor α;

The 3rd acquiring unit is used for obtaining the test number that is used to test according to default selective rule from the span that second acquisition unit obtains; Test number is at least one;

The 4th acquiring unit is used for getting access to zoom factor α according to default test order from the test number that the 3rd acquiring unit obtains.

Wherein, above-mentioned the 3rd acquiring unit specifically is used for according to default interval, and the numerical value in the span that second acquisition unit is obtained is sampled, and obtains the test number that is used to test;

The 4th acquiring unit specifically is used to obtain objective voice quality average score, the average segmentation energy of synthetic speech and the average segmentation energy of raw tone of the synthetic speech of test number; Average segmentation energy of objective voice quality average score, synthetic speech and the average segmentation energy of raw tone according to synthetic speech obtain zoom factor α from test number.

Since in the different CELP voice coding standards to self-adapting code book gain g_pScale factor γ with fixed codebook gain_GcCan adopt different quantization methods when quantizing, as unite quantification, independent quantification etc., correspondingly, the embodiment of the invention provides the quantization modules in the transcoding device, is specifically as follows:

(1) when destination end to the self-adapting code book g that gains_pScale factor γ with fixed codebook gain_GcUnite when quantizing;

The embodiment of the invention provides the transcoding device also to comprise:

Weight factor β acquisition module is used for according to self-adapting code book gain g_pScale factor γ with fixed codebook gain_Gc, determine weight factor β.

Correspondingly,

Quantization modules is specially quantifying unit, is used for according to the gain vector code table that obtains destination end, to α * γ_GcAnd g_pThe quantization code word in the gain vector code table is obtained in the associating quantification, realizes the transcoding of gain parameter, wherein, unites when quantizing, and quantizing rule is:

$E=\underset{0\≤k\≤128}{\arg \min }\{{(\mathrm{\α}\×{\mathrm{\γ}}_{\mathrm{gc}}-{\mathrm{\γ}}_{\mathrm{gc}}^k)}^2+\mathrm{\β}{(g_p-g_p^k)}^2\}$

(2), when destination end to the self-adapting code book g that gains_pScale factor γ with fixed codebook gain_GcWhen carrying out the independence quantification;

Quantization modules is specially quantifying unit, is used for according to destination end fixed code book scale factor code table and self-adapting code book gain code table, respectively to α * γ_GcAnd g_pQuantize, obtain the quantification self-adaptation code word in quantification fixed codeword in the quantization scaling factor code table and the self-adapting code book gain code table, realize the transcoding of gain parameter, wherein, quantizing rule is:

$E_1=\underset{0\≤k_1\≤N_1}{\arg \min }{(\mathrm{\α}\×{\mathrm{\γ}}_{\mathrm{gc}}-{\mathrm{\γ}}_{\mathrm{gc}}^{k_1})}^2$

$E_2=\underset{0\≤k_2\≤N_2}{\arg \min }{(g_p-g_p^{k_2})}^2$

Further, can also be according to the concrete needs of transcoding, above-mentioned quantization modules is specially absolute quantifying unit, is used for according to destination end fixed code book scale factor code table and self-adapting code book gain code table, respectively to α * γ_GcAnd g_pDefinitely quantize, obtain the quantification self-adaptation code word in quantification fixed codeword in the quantization scaling factor code table and the self-adapting code book gain code table, realize the transcoding of gain parameter, wherein, absolute quantizing rule is:

$E_1^{\′}=\underset{0\≤k_1\≤N_1}{\arg \min }|\mathrm{\α}\×{\mathrm{\γ}}_{\mathrm{gc}}-{\mathrm{\γ}}_{\mathrm{gc}}^{k_1}|;E_2^{\′}=\underset{0\≤k_2\≤N_2}{\arg \min }|g_p-g_p^{k_2}|,$

In order to reduce the operation of gain parameter transcoding re-quantization partly, further reduce the computation complexity of gain parameter part, referring to Fig. 8, the transcoding device that the embodiment of the invention provides can also comprise:

Mapping code table generation module is used for according to zoom factor α and weight factor β, gain quantization code table in source end gain quantization code table and the subject encoder is shone upon obtain shining upon code table;

Processing module is used for the self-adapting code book gain g according to the current subframe of the source end code stream that obtains_pScale factor γ with fixed codebook gain_Gc, search the mapping code table, realize transcoding.

Wherein, mapping code table generation module specifically comprises:

First processing unit according to the gain index of source end scrambler, is handled the gain index inverse quantization of source end scrambler;

Second processing unit is used for according to zoom factor α, and convergent-divergent is carried out in the gain that obtains after the first processing unit inverse quantization is handled;

The 3rd processing unit is used for the quantizing rule according to destination end, quantizes the gain behind the second processing unit convergent-divergent;

The manages the unit everywhere, is used to obtain the gain index value after the 3rd processing unit quantizes, and the gain index value after the quantification forms the mapping code table.

Those skilled in the art can also be known, the thought that code table is shone upon in the introducing of in like manner adopting above-mentioned present embodiment to provide, and the transcoding device that the embodiment of the invention provides can also comprise:

Mapping code table generation module is used for according to zoom factor α and weight factor β, and the quantizing rule of destination end, and the code stream of source end scrambler and the code stream of destination end scrambler are shone upon, and obtains shining upon code table;

Processing module is used to obtain the code stream of source end, searches the mapping code table, realizes transcoding.

In sum, the transcoding device that the embodiment of the invention provides is by introducing zoom factor to fixed codebook gain scale factor γ_GcAfter carrying out convergent-divergent, adopt new quantification criterion in the gain code table of subject encoder, to search for again, finish the transcoding of gain parameter, thereby reduce the computation complexity of gain parameter transcoding effectively, improve the quality of transcoding voice.

Part steps in the embodiment of the invention can utilize software to realize that corresponding software programs can be stored in the storage medium that can read, as CD or hard disk etc.

The above only is specific embodiments of the invention;, for those skilled in the art, not all under the prerequisite that does not break away from the principle of the invention in order to restriction the present invention; any modification of being done, be equal to replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (21)

Translated fromChinese

1、一种转码方法，其特征在于，所述方法包括：1. A transcoding method, characterized in that the method comprises:获取源端码流中的当前子帧的自适应码书增益g_p和固定码书增益的比例因子γ_gc；Obtain the adaptive codebook gain g_p and the scaling factor γ_gc of the fixed codebook gain of the current subframe in the source code stream;根据缩放因子α，对所述γ_gc进行处理得到α×γ_gc；According to the scaling factor α, the γ_gc is processed to obtain α×γ_gc ;根据目标端的量化规则，对所述α×γ_gc和g_p进行量化，实现转码。Quantize the α×γ_gc and_gp according to the quantization rules at the target end to implement transcoding.2、如权利要求1所述的转码方法，其特征在于，所述方法还包括：获取缩放因子α，具体为：2. The transcoding method according to claim 1, further comprising: obtaining the scaling factor α, specifically:根据所述源端和目标端的固定码书的平均能量，确定缩放因子α的客观理想值α_L；Determine an objective ideal value α_L of the scaling factor α according to the average energy of the fixed codebook at the source end and the target end;根据所述α_L，确定所述缩放因子α的取值范围；According to the α_L , determine the value range of the scaling factor α;根据预设的选择规则，从所述取值范围中获取用于进行测试的测试数值；所述测试数值至少为一个；According to a preset selection rule, a test value for testing is obtained from the value range; the test value is at least one;根据预设的测试规则，从所述测试数值中获取缩放因子α。According to a preset test rule, the scaling factor α is obtained from the test value.3、如权利要求2所述的转码方法，其特征在于，所述预设的选择规则，具体为：3. The transcoding method according to claim 2, wherein the preset selection rules are specifically:根据预设的间隔，对所述取值范围内的数值进行抽样。The values within the value range are sampled according to a preset interval.4、如权利要求2所述的转码方法，其特征在于，所述根据预设的测试规则，从所述测试数值中获取到缩放因子α，具体为：4. The transcoding method according to claim 2, wherein the scaling factor α is obtained from the test value according to the preset test rules, specifically:获取所述测试数值的合成语音的客观语音质量平均评分、合成语音的平均分段能量和原始语音的平均分段能量；Obtaining the objective voice quality average score of the synthesized speech of the test value, the average segmental energy of the synthesized speech and the average segmented energy of the original speech;根据所述合成语音的客观语音质量平均评分、所述合成语音平均分段能量与原始语音平均分段能量，从所述测试数值中获取缩放因子α。The scaling factor α is obtained from the test value according to the objective speech quality average score of the synthesized speech, the average segmental energy of the synthesized speech and the average segmental energy of the original speech.5、如权利要求1所述的转码方法，其特征在于，所述对所述α×γ_gc和g_p进行量化，具体为：5. The transcoding method according to claim 1, wherein said quantizing said α×γ_gc and_gp is specifically:获取权重因子β，根据所述权重因子β和所述目标端的增益矢量码表，对所述α×γ_gc和g_p进行联合量化，获取所述增益矢量码表中的量化码字。Acquiring a weight factor β, performing joint quantization on the α×γ_gc and g_p according to the weight factor β and the gain vector code table of the target end, and obtaining quantized codewords in the gain vector code table.6、如权利要求1所述的转码方法，其特征在于，所述对所述α×γ_gc和g_p进行量化，具体为：6. The transcoding method according to claim 1, wherein said quantizing said α×γ_gc and_gp is specifically:根据所述目标端固定码书比例因子码表和自适应码书增益码表，分别对所述α×γ_gc和g_p进行量化，获取所述量化比例因子码表中的量化固定码字和自适应码书增益码表中的量化自适应码字。Quantize the α×γ_gc and_gp respectively according to the fixed codebook scale factor code table and the adaptive codebook gain code table at the target end, and obtain the quantized fixed codeword and the quantized fixed codeword in the quantized scale factor code table. Quantized adaptive codewords in the adaptive codebook gain code table.7、如权利要求6所述的转码方法，其特征在于，所述分别对所述α×γ_gc和g_p进行量化，还可以为：7. The transcoding method according to claim 6, wherein said quantizing said α×γ_gc and_gp respectively can also be:分别对所述α×γ_gc和g_p进行绝对量化。Absolute quantification is performed on the α×γ_gc and_gp respectively.8、如权利要求5所述的转码方法，其特征在于，所述获取权重因子β，具体为：8. The transcoding method according to claim 5, wherein said obtaining the weight factor β is specifically:根据所述自适应码书增益g_p和固定码书增益的比例因子γ_gc，确定所述权重因子β。The weighting factor β is determined according to the adaptive codebook gain g_p and the scaling factor γ_gc of the fixed codebook gain.9、如权利要求5所述的转码方法，其特征在于，所述方法还包括：9. The transcoding method according to claim 5, further comprising:根据缩放因子α和权重因子β，对所述源端增益量化码表与目标编码器中增益量化码表进行映射得到映射码表；According to the scaling factor α and the weighting factor β, the source-end gain quantization code table is mapped to the gain quantization code table in the target encoder to obtain a mapping code table;获取源端码流中的当前子帧的自适应码书增益g_p和固定码书增益的比例因子γ_gc，查找所述映射码表，实现增益参数的转码。Obtain the adaptive codebook gain g_p of the current subframe in the source code stream and the scaling factor γ_gc of the fixed codebook gain, search the mapping code table, and realize the transcoding of the gain parameters.10、如权利要求9所述转码方法，其特征在于，所述对所述源端增益量化码表与目标编码器中增益量化码表进行映射得到映射码表，具体为：10. The transcoding method according to claim 9, characterized in that the mapping code table is obtained by mapping the source gain quantization code table and the gain quantization code table in the target encoder, specifically:根据所述源端编码器的增益索引，对所述源端编码器的增益索引反量化处理；Dequantize the gain index of the source encoder according to the gain index of the source encoder;根据所述缩放因子α，将反量化处理后得到的增益进行缩放；Scaling the gain obtained after inverse quantization processing according to the scaling factor α;根据目标端的量化规则，量化所述缩放后的增益；Quantizing the scaled gain according to a quantization rule at the target end;获取量化后的增益索引值，所述量化后的增益索引值形成映射码表。A quantized gain index value is acquired, and the quantized gain index value forms a mapping code table.11、如权利要求5所述的转码方法，其特征在于，所述方法还包括：11. The transcoding method according to claim 5, further comprising:根据所述缩放因子α和权重因子β，以及目标端的量化规则，对所述源端编码器的码流和目标端编码器的码流进行映射，得到映射码表；Mapping the code stream of the source encoder and the code stream of the target encoder according to the scaling factor α and weight factor β, and the quantization rule of the target end, to obtain a mapping code table;获取所述源端的码流，查找所述映射码表，实现增益参数的转码。Obtain the code stream of the source, search the mapping code table, and realize the transcoding of the gain parameters.12、一种转码装置，其特征在于，所述装置包括：12. A transcoding device, characterized in that the device comprises:获取模块，用于获取源端码流中的当前子帧的自适应码书增益g_p和固定码书增益的比例因子γ_gc；An acquisition module, configured to acquire the adaptive codebook gain_gp and the scaling factor γ_gc of the fixed codebook gain of the current subframe in the source stream;缩放模块，用于根据缩放因子α，对所述获取模块获取到的γ_gc进行缩放处理，得到α×γ_gc；A scaling module, configured to perform scaling processing on the γ_gc acquired by the acquisition module according to the scaling factor α, to obtain α×γ_gc ;量化模块，用于根据目标端的量化规则，对所述α×γ_gc和g_p进行量化，实现转码。The quantization module is used to quantize the α×γ_gc and_gp according to the quantization rules of the target end to realize transcoding.13、如权利要求12所述的转码装置，其特征在于，所述装置还包括：13. The transcoding device according to claim 12, further comprising:缩放因子α获取模块，用于获取所述缩放因子。A scaling factor α acquiring module, configured to acquire the scaling factor.14、如权利要求13所述的转码装置，其特征在于，所述缩放因子α获取模块具体包括：14. The transcoding device according to claim 13, wherein the scaling factor α acquisition module specifically includes:第一获取单元，用于根据所述源端和目标端的固定码书的平均能量，确定缩放因子α的客观理想值α_L；The first acquisition unit is configured to determine an objective ideal value α_L of the scaling factor α according to the average energy of the fixed codebooks at the source end and the target end;第二获取单元，用于根据所述第一获取单元获取的所述α_L，确定所述缩放因子α的取值范围；The second acquisition unit is configured to determine the value range of the scaling factor α according to the α_L acquired by the first acquisition unit;第三获取单元，用于根据预设的选择规则，从所述第二获取单元获取的取值范围中获取用于进行测试的测试数值；所述测试数值至少为一个；A third acquisition unit, configured to acquire a test value for testing from the range of values acquired by the second acquisition unit according to a preset selection rule; the test value is at least one;第四获取单元，用于根据预设的测试规则，从所述第三获取单元获取的所述测试数值中获取到缩放因子α。The fourth obtaining unit is configured to obtain the scaling factor α from the test value obtained by the third obtaining unit according to a preset test rule.15、如权利要求14所述的转码装置，其特征在于，所述第三获取单元具体用于根据预设的间隔，对所述第二获取单元获取的取值范围内的数值进行抽样，获取用于进行测试的测试数值；15. The transcoding device according to claim 14, wherein the third acquisition unit is specifically configured to sample the values within the value range acquired by the second acquisition unit according to a preset interval, Obtain test values for testing;所述第四获取单元具体用于获取所述测试数值的合成语音的客观语音质量平均评分、合成语音的平均分段能量和原始语音的平均分段能量；根据所述合成语音的客观语音质量平均评分、所述合成语音平均分段能量与原始语音平均分段能量，从所述测试数值中获取缩放因子α。The fourth obtaining unit is specifically used to obtain the average score of objective speech quality of the synthesized speech of the test value, the average segmental energy of the synthesized speech, and the average segmental energy of the original speech; score, the average segmental energy of the synthesized speech and the average segmental energy of the original speech, the scaling factor α is obtained from the test values.16、如权利要求12所述的转码装置，其特征在于，当所述目标端对自适应码书增益g_p和固定码书增益的比例因子γ_gc进行联合量化时；16. The transcoding device according to claim 12, wherein when the target end jointly quantizes the adaptive codebook gain_gp and the scaling factor γ_gc of the fixed codebook gain;所述装置还包括；The device also includes;权重因子β获取模块，用于根据所述自适应码书增益g_p和固定码书增益的比例因子γ_gc，确定所述权重因子β。A weight factor β acquiring module, configured to determine the weight factor β according to the adaptive codebook gain g_p and the scaling factor γ_gc of the fixed codebook gain.相应地，所述量化模块具体为量化单元，用于根据所述目标端的增益矢量码表和所述权重因子β获取模块获取的β，对所述α×γ_gc和g_p进行联合量化，获取所述增益矢量码表中的量化码字，实现转码。Correspondingly, the quantization module is specifically a quantization unit, configured to jointly quantize the α×γ_gc and_gp according to the gain vector code table of the target end and the β obtained by the weight factor β acquisition module, to obtain The quantized codewords in the gain vector code table realize transcoding.17、如权利要求12所述的转码装置，其特征在于，当所述目标端对自适应码书增益g_p和固定码书增益的比例因子γ_gc进行独立量化时；17. The transcoding device according to claim 12, wherein when the target end independently quantizes the adaptive codebook gain_gp and the scaling factor γ_gc of the fixed codebook gain;所述量化模块具体为量化单元，用于根据所述目标端固定码书比例因子码表和自适应码书增益码表，分别对所述α×γ_gc和g_p进行量化，获取所述量化比例因子码表中的量化固定码字和自适应码书增益码表中的量化自适应码字，实现转码。The quantization module is specifically a quantization unit, configured to quantize the α×γ_gc and_gp respectively according to the fixed codebook scale factor code table and the adaptive codebook gain code table at the target end, and obtain the quantized The quantized fixed codewords in the scale factor code table and the quantized adaptive codewords in the adaptive codebook gain code table realize transcoding.18、如权利要求17所述的转码装置，其特征在于，所述量化模块具体为绝对量化单元，用于根据所述目标端固定码书比例因子码表和自适应码书增益码表，分别对所述α×γ_gc和g_p进行绝对量化，获取所述量化比例因子码表中的量化固定码字和自适应码书增益码表中的量化自适应码字，实现转码。18. The transcoding device according to claim 17, wherein the quantization module is specifically an absolute quantization unit, configured to, according to the fixed codebook scale factor code table and the adaptive codebook gain code table at the target end, Absolute quantization is performed on the α×γ_gc and_gp respectively, and the quantized fixed codewords in the quantization scale factor code table and the quantized adaptive codewords in the adaptive codebook gain code table are obtained to realize transcoding.19、如权利要求12所述的转码装置，其特征在于，所述装置还包括：19. The transcoding device according to claim 12, further comprising:映射码表生成模块，用于根据所述缩放因子α和权重因子β，对所述源端增益量化码表与目标编码器中增益量化码表进行映射得到映射码表；A mapping code table generating module, configured to map the source gain quantization code table and the gain quantization code table in the target encoder to obtain a mapping code table according to the scaling factor α and the weighting factor β;处理模块，用于根据获取的源端码流中的当前子帧的自适应码书增益g_p和固定码书增益的比例因子γ_gc，查找所述映射码表，实现转码。The processing module is configured to search the mapping code table according to the adaptive codebook gain g_p of the current subframe and the scaling factor γ_gc of the fixed codebook gain in the obtained source code stream, so as to realize transcoding.20、如权利要求19所述的转码装置，其特征在于，所述映射码表生成模块具体包括：20. The transcoding device according to claim 19, wherein the mapping code table generation module specifically includes:第一处理单元，根据所述源端编码器的增益索引，对所述源端编码器的增益索引反量化处理；The first processing unit dequantizes the gain index of the source encoder according to the gain index of the source encoder;第二处理单元，用于根据所述缩放因子α，将所述第一处理单元反量化处理后得到的增益进行缩放；The second processing unit is configured to scale the gain obtained after inverse quantization processing by the first processing unit according to the scaling factor α;第三处理单元，用于根据目标端的量化规则，量化所述第二处理单元缩放后的增益；a third processing unit, configured to quantize the scaled gain of the second processing unit according to a quantization rule of the target;第四处理单元，用于获取第三处理单元量化后的增益索引值，所述量化后的增益索引值形成映射码表。The fourth processing unit is configured to obtain the quantized gain index value of the third processing unit, and the quantized gain index value forms a mapping code table.21、如权利要求12所述的转码装置，其特征在于，所述装置还包括：21. The transcoding device according to claim 12, further comprising:映射码表生成模块，用于根据所述缩放因子α和权重因子β，以及目标端的量化规则对所述源端编码器的码流和目标端编码器的码流进行映射，得到映射码表；A mapping code table generating module, configured to map the code stream of the source encoder and the code stream of the target encoder according to the scaling factor α and weight factor β, and the quantization rules of the target end, to obtain a mapping code table;处理模块，用于获取所述源端的码流，查找所述映射码表，实现转码。The processing module is used to obtain the code stream of the source, search the mapping code table, and implement transcoding.

Images