Audio encoder and audio decoder for encoding frames presented in form of audio signal samples

FIELD: information technology.

SUBSTANCE: audio encoder (100) for encoding frames presented in form of audio signal samples to obtain encoded frames, wherein a frame consists of a plurality of time domain audio signals, including a predictive coding analysis stage (110) and determining information on coefficients of a synthesis filter and prediction domain frame information based on a frame of audio samples. The audio encoder (100) further includes a domain converter (120) for converting a frequency domain audio sample frame and obtaining a frame spectrum and an encoding domain computer (130) for making a decision on encoded data for a frame based on information on coefficients and information on a prediction domain frame, or based on the frame spectrum. The audio encoder (100) includes a controller (140) for determining information on a switching coefficient for cases when the encoding domain computer decides that encoded data of the current frame are based on information on coefficients and information on a prediction domain frame, and [for cases] when data of a previous frame were encoded based on the spectrum of the previous frame and redundancy reducing encoder (150) for encoding information on the prediction domain frame, information on coefficients, information on the switching coefficient and/or frame spectrum.

EFFECT: improved concept of audio encoding using encoding domain switching.

14 cl, 29 dwg

The present invention relates to the field of audio encoding/decoding, and more particularly to a method of audio encoding using multiple domains encoding.

In the technique of encoding a known coding scheme in the frequency domain, such as MP3 or AAC. These encoders in the frequency domain based on the transformation of the "time domain"/"frequency domain", with the subsequent stage sampling in which the sampling error is controlled using information from the psychoacoustic module and the encoding step, in which discrete spectral coefficients and the corresponding additional information enables you to perform entropy encoding using a code table.

On the other hand there are encoders, which are very suitable for speech processing, such as AMR-WB+, as described in 3GPP TS 26,290. Such coding schemes perform speech LP (LP = Linear Prediction) filtering of the signal in the time domain. This LP filtering is obtained on the basis of linear prediction analysis of the input signal in the time domain. The resulting coefficients of the LP filter is then discretized/encoded and transmitted as additional information. The process is known as LPC (LPC = linear prediction coding). At the output of the filter differential signal prediction or signal error the prediction, which is also known as the excitation signal, is encoded using phase analysis-synthesis ACELP encoder or, on the contrary, it is encoded using the encoder conversion, which uses the Fourier transform with overlap. The choice between the ACELP coding and transform coded excitation, which is also referred to as TLC, encoding is performed using the algorithm with a closed or open loop.

The encoding of sound in the frequency domain, such as a highly efficient coding scheme AAC, combine coding scheme AAS and method for recovering spectral range, can be combined with instruments stereo or multi-channel encoding, which is known under the term "MPEG environment.

On the other hand, speech coders, such as AMR-WB+, also have a stage of amplification of high frequencies and stereo channel.

The coding scheme in the frequency domain advantageous in that they allow you to get high quality at low baratinha [low sampling frequency] for musical signals. However, it is problematic to obtain high-quality speech signals at low baratinha. The coding scheme of speech allow you to get high quality for speech signals even at low nitratine, and give low quality for music signals at low baratinha.

The schema to which tiravanija in the frequency domain often use the so-called MDCT (MDCT = Improved discrete cosine transformation). MDCT was originally described in J. Princen, A. Bradley, "Analysis/Synthesis Filter Bank Design Based on Time Domain Aliasing Cancellation", IEEE Trans. ASSP, ASSP-34(5):1153-1161, 1986., IEEE Trans. ASSP, ASSP-34 (5):1153-1161, 1986. MDCT or set of filters MDCT is widely used in modern and efficient audio encoders. This type of signal processing provides the following advantages: Smooth crossfade [transition] between processing units: Even if the signal in each processing unit is changed in different ways (for example, due to the discretization of the spectral coefficients), it does not disappear artifacts [deviation, distortion]associated with abrupt transitions from block to block, due to overlapping Windows/ [or] additional operations. Critical moment [MDCT]: the number of spectral values at the output of the filter block is equal to the number of input values for the temporal areas at the entrance [of the filter block] and thus should provide additional value.

MDCT block filter provides a high frequency selectivity and coding gain.

These useful properties are achieved through the use of the method exceptions overlay in the time domain. Exclude overlapping in the time domain performs the convolution synthesis of overlap of the two signals adjacent Windows. If between the stages of analysis and synthesis of MDCT does not apply discretization, get a quality restoration of the original signal. Od is ako, MDCT is used for encoding, which is specially adapted for musical signals. Such an encoding scheme in the frequency domain, as noted above, reduce the quality of speech signals at low bit rate, while specially adapted speech encoders are of a higher quality at a comparable speed transmission or even have much lower data rate for the same quality as compared with coding schemes in the frequency domain. Methods of encoding speech, such as AMR-WB+ (AMR-WB+ = adaptive multi-speed broadband) encoder, as defined in technical specification "Extended Adaptive Multi-Rate - Wideband (AMR-WB+) codec", 3GPP TS 26.290 V6.3.0, 2005-06, [methods of encoding speech] do not use MDCT and, therefore, may not use any of the advantages of the excellent properties of the MDCT which, in particular, on the one hand, rely critically on the selected process, and on the other hand, use a transition from one block to another.

Thus, the transition from one block to another is achieved by using MDCT without loss in data rate and, therefore, a critical moment MDCT still does not occur in the speech encoders. You could combine speech encoders and audio encoders within a hybrid coding scheme, but there is still the problem is mA switching from one mode encoding to another at low speed and with high quality.

Conventional approaches to the encoding of sound is usually intended for the beginning of the audio file, or for communication. The use of these traditional approaches, filter structures, such as filters, prediction, allows to reach a stationary state at a certain time from the beginning of the procedure of encoding or decoding. However, to activate the system audio encoding, for example, on the one hand, using transform-based coding, and, on the other hand, [using] encoding speech according to a preliminary analysis on the input, the corresponding patterns of filters will not be active and constantly updated. For example, speech encoders within a short period of time may be reused [boot]. After rebooting again begins the start-up period, the internal States are reset. For example, the required duration reach a steady state for the speech encoder may be crucial, especially for the quality of the transitions. Conventional approaches, such as, for example, AMR-WB+ "with the technical characteristics of the Extended Adaptive Multi-Rate - Wideband (AMR-WB+) codec", 3GPP TS 26,290 V6.3.0, 2005-06, used for General reset of the encoder speech during the transition or switching between the conversion of the main encoder and the encoder speech. AMR-WB+ optimized condition is, he only runs once, when the signal is lost in the assumption that there is no intermediate stops or discharges. Therefore, all memory of the encoder can be updated for a frame using the frame. When the AMR-WB+ is used in the middle of the signal, called reset, and all the memory used for encoding or decoding, is set to zero. Thus, conventional approaches have the problem that are too long to reach a steady state of the encoder speech, and, in addition, make a strong distortion in the instability phase.

Another problem with conventional approaches is that they use large segments overlap when switching areas of coding, making ceilings that give adverse effects to the efficiency of encoding.

The object of the present invention is the improved audio coding using switch regions encoding.

This is achieved by the audio encoder in accordance with claim 1, the method for audio encoding in accordance with claim 7, device, audio decoding of claim 8, the method of audio decoding in accordance with 14, and a computer program according to § 15. The present invention is based on the assumption that the above problems can be solved in the device D. the coding by reviewing the information on the status of the filter after a reset. For example, after a reset, when the state of a particular filter is reset, start-up procedure or transfer of the filter to the operating state can be reduced if the filter starts not with phase switching, i.e. when all of the States or the memory is set to zero, and [begins] with information about a state from which it can be implemented fast start or a small period before the operation.

Following the invention is that the above information about the state of the switch can be generated in the device for encoding or decoding. For example, when choosing between the approach to coding based on predictions and on the basis of conversion, additional information may be provided to switch in order for the decoder to use filters synthesis predictions in steady state, before you can use the results of their [filters].

In other words, this disclosure of the present invention, which is particularly important when switching between the conversion and the area of predicting when the switching device, audio encoding, additional information about the condition of the filter closer to the actual switching on the area to depict the Azania, can eliminate the problem of generating artifacts [distortion] switch.

Another aspect of the invention is that such information about switching can be transmitted to the decoder only when the analysis of its release shortly before you perform the actual switch, and the basic process of running the encoder is based on processing of the output and determine the information about the filter or the memory status shortly before switching. In some embodiments, this can use conventional encoders and reduction of artifacts when switching to be associated exclusively with the operation of the decoding device. Taking into account the above information, for example, filters, prediction can be operable before the actual switching, i.e. by analyzing the output of the transformation corresponding decoding device. Embodiments of the present invention will be specified with use of the accompanying drawings, on which:

figure 1 shows a variant of the device, audio encoding;

figure 2 shows a variant of the device, audio decoding;

figure 3 shows the shape of the window used in the embodiment;

figa and 4b show MDCT and temporal overlapping area;

figure 5 shows the block diagram of the embodiment to cancel the time domain cash the taxpayer;

figa-6g illustrate the signals to be processed to cancel the imposition of the temporary area in the embodiment;

figa-7g illustrate a chain of signal processing to cancel the imposition of the temporary area in the embodiment that uses the device decoding a linear prediction;

figa-8g shows a chain of signal processing in the variant with the abolition of overlapping time domain; and

figa and 9b show a signal processing device for encoding and decoding options.

1 shows a variant of the audio encoding device 100. The audio encoding device 100 is intended to encode the frames presented in the form of samples of the audio signal to obtain encoded frames in which the frame consists of several audio samples in the time domain. The embodiment of the device audio coding includes the analysis stage when coding with prediction 110 to determine information about the filter coefficients of the synthesis and the frame information of the prediction region on the basis of a frame of audio samples. In variants of the embodiment of the frame of the prediction area can match the frame excitation or a filtered version of the frame excitation. Later in Pego may be included coding region prediction when encoding information about the filter coefficients of the synthesis and information about the milling is IU the field of prediction of PA-based frame of audio samples. In addition, the embodiment of the audio encoding device 100 includes a transducer region 120 for converting a frame of audio samples in the frequency domain to obtain a spectrum of the frame. Subsequently, it can be used to convert the coding region, when the coded frame of the spectrum. In addition, the embodiment of the audio encoding device 100 includes a transmitter coding region 130 for a decision to be encoded data for a frame based on the information on the coefficients and on the frame information region predictions, or [data for a frame is based] on the spectrum of the frame. The embodiment of the audio encoding device 100 includes a controller 140 to determine the information about the ratio of the switch when the transmitter coding region determines that the encoded data of the current frame based on the information on the coefficients and the frame information region prediction, and encoded data of the previous frame is encoded based on a previous spectrum of the frame.

The embodiment of the audio encoding device 100 further comprises an encoder redundancy reduction 150 to encode the frame information region predictions, information on odds, information on the ratio of the switch and/or on the frame of the spectrum. In other words, the transmitter region Kodirov is of 130 defines a region encoding, while the controller 140 provides information on the ratio of switching when switching from the transformation to the field of prediction.

In figure 1 some compounds appear broken lines. They point to different ways in the embodiments. For example, information about the coefficients of the switch can be simply obtained a permanent job analysis stage coding prediction 110 so that the information on the coefficients and the information on the frame area predictions are always available at the respective output. Then, the controller 140 may indicate redundancy reduction in the encoding device 150, when the encoding of the output of the analysis stage of coding prediction 110 and, when the encoding of the output spectrum of a frame in the frequency Converter region 120 after the decision to switch is performed by computer coding 130. Therefore, the controller 140 can detect a redundancy reduction encoder redundancy reduction 150 and encode information about the coefficient switch to switch from the transformation to the field of prediction.

If a switch occurs, the controller 140 may indicate redundancy reduction in the encoding device 150 to encode overlapping frame in the previous frame redundancy reduction ustroystva encoding 150 can be controlled by the controller 140 so to a stream of bits contained as the previous frame, and information on the coefficients, the frame information region predictions, as well as the spectrum of the frame. In other words, in variants of the embodiments, the controller may control the redundancy reduction in the encoding device 150 so that the encoded frames include the above information. In other embodiments, the transmitter coding 130 may take the decision to change the coding region and to perform switching from the stage of analysis coding prediction 110 to the inverter frequency domain 120.

In these embodiments, the controller 140 may perform some internal analysis, in order to obtain the coefficients of the switch. In embodiments, the information ratio switching may correspond to information about the States of the filter, adapted to the content of the code table, the memory, information about the excitation signal, LPC coefficients, etc.

Information about the ratio of the switch may contain any information that allows you to put in a working condition or initialize phase synthesis predictions 220.

The transmitter coding 130 may determine your decision to switch region encoding on the basis of frames or samples of audio signals, which are also shown in dotted lines nafig 1. In other embodiments, this decision may be made based on factors information about the prediction frame field and/or frame of the spectrum.

In General, the options are not limited in a way that is embodied in computer coding 130 to change the region coding, and, most importantly, change the region coding is determined by the evaluator coding 130, during which arise the problems described above. In some embodiments, embodiments of the audio encoding device 100 is coordinated in such a way that the above-described significant drawbacks, at least partly compensated. In variants of the embodiments, the transmitter coding 130 can be adapted to a decision based on the properties of the signal or audio frames. As already known, the properties of an audio signal can determine the coding efficiency, i.e. for some characteristics of the audio signal, can be more efficient to use a transform-based coding, other characteristics may be more efficient using a prediction coding. In some embodiments, the transmitter coding 130 can be adapted for making decisions about the use of transform-based coding, when the signal which is mixed or voice type. If the signal is mixed or voice type, the transmitter coding 130 can be adapted for making decisions about the use of the frame region prediction, which is used for encoding.

In accordance with the broken lines and arrows in figure 1, the controller 140 may be provided with information on the coefficients, the frame information region prediction and spectrum of the frame, and the controller 140 may be adapted to determine information about the coefficient switching on the basis of the above information. In other embodiments, the controller 140 may provide information for the analysis stage when coding with prediction SOFTWARE to determine the coefficients of the switch. In some embodiments, embodiments of the coefficients switch can correspond to the information on the coefficients, and in other embodiments, they may be defined in various ways.

Figure 2 shows a variant of the device audio decoder 200. The embodiment of the device audio decoder 200 is intended for decoding encoded frames to obtain a frame of samples of the audio signal, and the frame consists of several audio samples in the time domain. The embodiment of the device audio decoder 200 includes dekodirovok obtain redundancy 210 for decoding encoded frames and receive, and the formation of the frame region prediction information on coefficients for a filter of the synthesis and/or spectrum of the frame. In addition, the embodiment of the device audio decoder 200 includes a stage of synthesis predictions 220 to determine frame prediction audio samples on the basis of information about the coefficients for filter synthesis and the frame information region predictions, and the time domain Converter 230 to convert the frame of the spectrum in the time domain and obtain the transformed frame from the spectrum of the frame. The embodiment of the device audio decoder 200 further comprises an adder 240 to combine the converted frame and frame prediction and receive frames presented in the form of samples of the audio signal.

In addition, the embodiment of the device audio decoder 200 includes a controller 250 for controlling the switching process. The switching process is carried out effectively when the previous frame based on the converted frame and the current frame based on the frame prediction. The controller 250 allows to obtain the coefficients of the switching stage of the synthesis predictions 220 for the preparation of initialization or transfer to operational status phase of the synthesis of predictions 220, so that the synthesis phase predictions 220 is initialized when the process of transition.

In accordance with the dotted arrows on IG controller 250 may be adapted to manage parts or all components of the device audio decoder 200. The controller 250 may be, for example, adapted to coordinate the receipt of redundancy in the device audio decoder 210, with the aim of obtaining additional information on the transfer factors or information about the previous frame area predictions, etc. In other embodiments, the controller 250 may be adapted to receive the above information on the coefficients of the switch, for example, by obtaining the decoded frames by the adder 240 and holding LP-analysis on the output of the adder 240. The controller 250 may be adapted to coordinate or manage the stage of synthesis predictions 220 and convert the time domain 230 in order to create the above frames overlap, synchronization, analysis of the time domain and the abolition of analysis time domain, etc.

The following is the LPC-based coding region, including the predictors and internal filters, which during start-up of the time required to achieve the state that would ensure accurate synthesis filter. In other words, in embodiments of the device, audio decoding stage 100 analysis coding prediction 110 may be adapted to determine information about the filter coefficients of the synthesis and the frame information region prediction based on the LPC analysis.

In embodiments, the device is udio decode stage 200 synthesis predictions 220 may be adapted to determine a predicted frame using the filter synthesis QTL.

Obviously, the use of rectangular Windows in the beginning of the first LPD (LPD = domain linear prediction) frame and reset the encoding based on the LPD in the zero state, does not provide the perfect execution of these transitions, because there is not enough time in the LPD coding to create a good signal, which will be introduced blocking artifacts.

In versions to manage the transition from not - LPD mode to mode, LPD, you can use overlapping Windows. In other words, in embodiments, the audio encoding device 100, the inverter frequency domain 120 may be adapted to convert a frame of audio samples based on the fast Fourier transform (FFT [fast Fourier transform] = fast Fourier transform), or MDCT (MDCT = Modified Discrete Cosine Transformation). Variants of the device audio decoder 200, the time domain Converter 230 may be adapted to convert the frame spectrum time domain on the basis of inverse FFT (IFFT = inverse FFT), or [based on] the inverse MDCT (IMDCT=inverse MDCT).

When this options can work in non-LPD mode, which can be used as the primary mode conversion, or [options can work] to LPD, which is also used as analysis and synthesis prediction. In General, the options could the t to use overlapping Windows, especially when using MDCT and IMDCT. In other words, in the not-LPD mode can be used overlapping Windows with the temporal overlapping area (TDA = the Overlap in the Time domain). Moreover, when switching from non-LPD mode LPD, the overlap in the time domain in the last not-LPD frame can be compensated. Embodiments may enter a temporary overlay to the original signal before performing the LPD coding, however, the imposition of [aliasing] the time domain may not be compatible with the prediction based on the coding domain of time, such as ACELP (ACELP = Excitation Linear Prediction Algebraic Code Table). Embodiments can introduce artificial smoothing at the beginning of the segment LPD and apply the cancellation of the domain of time as well as for transitions from ACELP to non-LPD. In other words, in variants of the embodiment of the analysis and synthesis of the prediction can be based on ACELP.

In some embodiments, the artificial smoothing is performed on the basis of the signal synthesis instead of the original signal. As the signal synthesis is inaccurate, especially during the startup phase LPD, these embodiments may compensate for the block artifacts by introducing artificial TDA, however, the introduction of artificial TDA may introduce additional error, along with the reduction of artifacts.

Figure 3 illustrates the process of transition in the nome of the embodiments. In the variant of figure 3, it is assumed that the transition process is switched from the non-LPD mode, for example mode MDCT on the LPD mode. As indicated in figure 3, the total length of the window is equal to 2048 samples. On the left part of figure 3 shows the extension of the front MDCT window on all 512 samples. In the processes of MDCT and IMDCT, these 512 samples from the front of the MDCT window will appear with the following 512 samples, which in figure 3 are for MDCT kernel, including the Central 1024 samples in the full window of 2048 samples. Next will be explained in more detail, using processes MDCT and IMDCT in the time domain overlap is not critical, when the previous frame was encoded in a non-LPD mode. This is one of the best advantages of MDCT is determined by smoothing the time domain can be essentially compensated by the corresponding sequential overlapping MDCT Windows.

Now consider the right part of the MDCT window. When the switch LPD revocation of temporary overlay is not automatic, and, starting with the first frame decoded in the LPD mode, the imposition of a time domain to compensate with the previous MDCT frame is not automatically used. Thus, in the overlapping region variants may use artificial smoothing of the time domain, as shown in figure 3 in the area of 128 samples with the center in MDCT Windows kernel i.e. with the center after 1536 samples. In other words, figure 3 assumes that artificial smoothing time domain introduced at the beginning, i.e. In this embodiment, the first 128 samples of the frame mode, LPD, introduced in the end of the last frame MDCT to compensate for the temporary overlay.

In a preferred embodiment, MDCT is used for obtaining the critical sample to move from the encoding operation in one area to the encoding operation in another different area, i.e. is carried out in the embodiments of the Converter region 120 and/or the time domain Converter 230. However, in all other converters [MDCT] can also be applied. Since, however, MDCT is the preferred option, MDCT will be discussed in more detail on figa and fig.4b.

On figa shows the box 470, which is a growing area on the left and decreasing the plot on the right, where you can divide the window into four parts: A, B, C, and D. the Box 470 has, as can be seen from the figure, shows the situation only with the imposition of the plots at 50% area overlap/add. In particular, the first part with samples from zero to N corresponds to the second part of the previous box 469, and the other half located between samples from N to 2N in the box 470 overlaps with the first area box 471, which in the shown embodiment is a window i+1, and the window 470 javljaetsja number i.

Operations MDCT can be seen as cascading operations: window, rollup, operations subsequent conversion and, in particular, with subsequent DCT (DCT = discrete cosine transform), which uses DCT operation type IV (DCT-IV). In particular, the operation of convolution is obtained by calculating the first part of N/2 folding unit as-c_R-d, and calculate the second part of N/2 folding output samples, and a-b_Rwhere R is the inverse operator. Thus, the results of the convolution operation is represented in N output values, while it was obtained 2N input values.

The corresponding operation of the scanner in the decoding device illustrated in equation form on figa.

As a rule, MDCT operation results in the form (aa , b, c, d) exactly the same values at the output, and DCT-IV with the (-CR-d, a-b_Rthat is shown on figa.

Accordingly, using the scan results IMDCT output operation of the scanner is transmitted to the output of the reverse DCT-IV. Thus, the time overlap is determined by performing a convolution operation in the encoding device. Then, the result of the window operation and the operation of convolution is converted into the frequency domain using the DCT conversion unit-IV, which requires N input values.

In device is iste decoding, N input values are converted back into the time domain using the DCT-IV operations, and the output of this operation is the inverse transformation, thus, turns into the operation of the scanner to obtain the 2N output, which, however, are smoothed output values.

To avoid aliasing, which was introduced on the operations of convolution and which still remains after the operation of the scanner, the operation of the overlap/convolution can cancel overlap in the time domain.

Therefore, when the result of the operation of the sweep is added to the previous result IMDCT in overlapping areas, reverse the cancellation conditions are obtained simply from the equation in the lower part figa, for example, b and d, thus restoring the original data.

In order to get TDAC for MDCT window, there is a requirement known as "Princen-Bradley-state, which means that the window coefficients is increased by 2 for the respective samples, which are combined in the time domain compensator overlay so that the result is in block (1) for each sample.

On figa shows the sequence of Windows, which, for example, is used in AAC-MDCT (AAS = Superior Audio Coding), for long or short Windows, figure 4, b illustrates the various window functions, which have, in addition to uchastka the overlay also the plot without anti-aliasing.

Figure 4, b shows the analysis function window 472 having a null area a1 and d2, plot overlay 472a, 472c, and land without overlays 472c.

Plot overlay 472c length of c2, d1 has a corresponding area applying subsequent box 473, marked 473b. Accordingly, the box 473 additionally includes the area without overlap, 473a. From the comparison of figure 4, b figa clear that due to the fact that due to the fact that there is zero plots a1, d1, box 472, or c1 to box 473, both Windows have the plot without overlap, and the window function in the area applying steeper than Figo. In this regard, the site overlay 472a corresponds to the L_kplot without overlap 472 C corresponds to the plot of M_kand the plot overlay 472b corresponds to R_kon fig.4b.

When the convolution operation is applied to the block of samples placed in the box 472, it turns out that the situation depicted in fig.4b. The left plot is extended to the first N/4 folding samples. The second part of the length N/2 samples are free from overlap, since the convolution operation is applied to the parts of a window that has a zero value, and the last N/4 samples, again, are added. In connection with the operation of convolution, the number of output values of the convolution operation is N, and at the entrance was 2N values, although, actually, N/2 values in this embodiment were installed in the ul due to operation in the window using the window 472.

Next, the DCT-IV is applied to the result of the convolution operation, but, importantly, the plot overlay 472, which in the transition from one mode of encoding to another encoding mode is processed by a method differing from that area without overlap, although both parts belong to the same block of samples and, more importantly, are entered in the same block of transform operations.

In addition, fig.4b shows the sequence of values in the Windows 472, 473, 474, where the box 473 is a window of transition from a situation, when there are plot without overlap to a situation where there are only areas of overlap. It turns out asymmetrically shaped window function. The right plot window 473 similar to the right area of the window in the sequence window on figa, while the left plot is the plot without overlap, and the corresponding zero plot (C1). Thus, figure 4, b shows the transition from MDCT-TCX in AAS, when AAS is performed using a fully overlapping Windows or, on the contrary, [figure] shows the transition from the AAS in MDCT-TCX when the window 474 contains TLC block of data with full overlap, which is a regular operation for MDCT-TCX on the one hand, and MDCT-AAC on the other hand, therefore, there is no reason to switch from one mode to another. Thus, the box 473 can be called "stopped by the window, which, moreover, has come predpochtitelney characteristic due to the fact, that the length of the window coincides with the length of at least one adjacent the window so that the overall structure of the unit or the border of the raster is saved when the unit is set to the same number of window coefficients, i.e, for example, 2N samples per figa or fig.4b. Hereinafter, methods of artificial overlap in the time domain and the abolition of overlapping in the time domain will be described in detail. Figure 5 shows the block diagram, which can be used in the embodiment, which contains a chain of signal processing. Figures 6a through 6g and 7a through 7g illustrate the sampling signals, and figures 6a through 6g illustrate the principle of the process of the abolition of overlapping in the time domain assuming that you are using the original signal, and figures 7a through 7g illustrate the sampling signals, which are determined on the basis of the assumption that the first frame LPD is obtained after a full reboot and without any adaptation.

In other words, figure 5 illustrates the embodiment of the process of introducing artificial overlap in the time domain and the abolition of overlapping in the time domain for the first frame in the LPD mode in case of transition from non-LPD mode in LPD mode. Figure 5 shows that the first window is used for the current frame LPD in block 510. As shown in figa, 6b, and figa, 7b, the window corresponds to the disappearance of the corresponding signals. As shown in the small g is the Afik above block window 510 in figure 5, it is assumed that the window is applied to the L_ksamples. The operation in box 510 corresponds to the convolution operation 520, resulting in L_k/2 samples. The result of the operation of convolution is shown in figs and 7c. It is seen that due to the reduction in the number of samples that has zero area, extended to L_k/2 samples in the beginning of the corresponding signals.

Window operation unit 510 and the addition in block 520 can be summarized as overlapping in the time domain, which is inserted through the MDCT. However, a subsequent convolution effects arise when the reverse conversion using the IMDCT. The effects caused by the IMDCT shown in figure 5 blocks 530 and 540, which can be summarised in a reverse overlap in the time domain. As shown in figure 5, when this is done the scan in block 530, which leads to a doubling of the number of samples, i.e. the result will be a Lk samples. Appropriate signals presented on fig.6d and 7d.

From fig.6d and 7d shows that the number of samples was doubled, and the time was set to overlay. The operation of the scanner 530 is invoked another window operation 540, by passing signals. The results of the second window operation 540 presents on file and 7e. Finally, for artificial [specified] time overlay of the signals shown fige and 7e, is the imposition of the addition to the previous frame, zakodirana is conducted in a non-LPD mode, [frame] which is shown by block 550 figure 5, and the corresponding signals are represented on fig.6f and 7f.

In other words, in embodiments the device audio decoder 200 and the adder 240 may be adapted to perform the functions of block 550 figure 5.

The resulting signals are shown in Figg and 7g. Summing up, in both cases, the left part of the respective frames are processed in the window that is shown in figa, 6b, 7a, and 7b. Then the left part of the window is formed, as shown in figs and 7C. After deployment, see the 6d and 7d, use different window operation, see file and 7e. On Fig.6f and 7f shows the frame of the current process, having the form of a previous non-LPD frame, and Figg and 7g represent the results after the operation overlay and summation. From figures 6a through 6g shows that high quality restoration can be achieved in the embodiments with the use of artificial TDA for LPD frames and the use of overlapping and convolution with the previous frame. However, in the second case, i.e. in the case shown in the figures 7a through 7g, the recovery is not perfect. As mentioned above, it is assumed that in the second case, the LPD mode has been completely reset, i.e. all state and memory with LPC synthesis were set to zero. The result of the synthesis signal was not accurate, since the first samples. Case artificial TDA-added overlap the results of convolution leads to distortions and artifacts more than perfect recovery, cf. figd and 7g.

On figa-6g and 8a-8g show a comparison of the use case of the initial signal for the artificial imposition of the temporary area and the abolition of the artificial imposition of the temporary storage area, [compared] with another use case for a start signal LPD, however, refer to figures 8a through 8g, it is assumed that the initial LPD takes more time than required on the drawings 7a through 7g. Figures 6a through 6g and 8a through 8g illustrate graphs of the selected signal that has been subjected to the same operations that have already been explained in figure 5.

From the comparison figd and 8g, it is seen that the distortion and artifacts introduced in the signal shown in Figg are more significant than Figg. The signal shown in Figg, contains a lot of distortion over a relatively long period of time. For comparison, figd shows perfect reconstruction [recovery] when applied to the original signal cancellation overlap in the time domain.

Embodiments of the present invention can accelerate the start-up period, for example, encoders based on the LPD compared with the embodiment of the analysis stage of the coding of the prediction stage 110 and synthesis predictions 220, respectively. Embodiments can update all necessary state and memory to bring the synthesized signal as close as possible to the oripa the national signal, to minimize distortion, as shown in Figg and 8g. In addition, in the embodiment can be enabled large overlap and periods of convolution, which is possible because of the improved injection time overlap in the time domain and the abolition of overlapping in the time domain.

As has been described above, the use of rectangular Windows in the beginning of the first or of the current frame LPD and reset the encoding based on the LPD in the zero state, is not the ideal option for navigation. Distortion and artifacts can occur, since there may not be enough time remaining for LPD encoder to create a good signal. Similar considerations are valid for setting the internal state variables of the encoder for any given initial values, as the steady state of the encoder depends on many properties of the signal, and the start time of any fixed, but fixed initial state can be long.

In variants of the embodiment of the audio encoding device 100, the controller 140 can be adapted for determining information on coefficients for a filter of the synthesis and the frame information forecasting switching on the basis of the LPC analysis. In other words, the options can use a rectangular window and reset the internal state of the encoder LPD. In some embodiments, Cody is owsik may include information about memory filter and/or adaptive code table, using ACELP, the synthesis of samples from the previous, not-LPD frames in the encoded frames and ensuring their decoding. In other words, the embodiment of the audio encoder 100 can decode the previous not-LPD frames, perform LPC analysis, and apply filter LPC analysis for non-LPD signal synthesis and to provide information to decode.

As noted above, the controller 140 may be adapted to determine the information about the ratio of the switch so that this information can represent a frame of audio samples, overlapping the previous frame.

In embodiments, audio codec, and 100 may be adapted to encode such information in the coefficients of the switch by using the encoder redundancy reduction 150. In one variant embodiment, the reset procedure can be improved by transfer or by including additional information on the LPC parameter, calculated from the previous frame in the bit stream. An additional set of LPC coefficients we will call LPC0.

In one embodiment, the encoder can operate in native mode, LPD coding using four LPC filter, namely LPC1 on LPC4, which are estimated and are defined precisely for each frame. In a variant, in the transitions from non-LPD coding coding LPD, additional LPC Phil is Tr marked as LPC0, which corresponds to the LPC analysis center at the end of the previous frame, [filter] can also be exactly determined or estimated. In other words, in the embodiment, frames, audio samples, overlapping the previous frame may have a center in the end of the previous frame.

In embodiments, the device audio decoder 200, the decoder receiving redundancy 210 may be adapted to decode the information on the ratio of switching from the encoded frames. Accordingly phase synthesis predictions 220 may be adapted to determine the switching frame prediction, which is applied to the previous frame. In another embodiment, when switching frame prediction, he can have the center at the end of the previous frame.

In embodiments, the LPC filter corresponding to the end-LPD segment or frame, i.e. LPCO, can be used for interpolation of LPC coefficients or to calculate the response in the absence of an input signal in the case ACELP.

As mentioned above, this LPC filter can be estimated by the direct method, i.e. the [filter] is calculated on the basis of the input signal is sampled by the encoder and transmitted to the decoder. In other embodiments, the LPC filter can be estimated inverse method, i.e. the decoder based on the synthesized signal. Direct assessment which may use additional bitrates [speed passing bits of information], but can also provide more efficient and reliable start-up period. In other words, in other embodiments, the controller 250 in the embodiment of the device audio decoder 200 can be adapted for the analysis of the previous frame and the information on the previous frame in the form of coefficients for filter synthesis and/or information about the previous frame in the frame region predictions. In addition, the controller 250 may be adapted to provide information about the previous frame in the form of coefficients for stage synthesis predictions 220, i.e. the coefficients of the switch. The controller 250 may also give information about the previous frame in the frame area predictions for the preparation stage of the synthesis predictions 220.

In embodiments, when the audio encoding device 100 provides information about the magnitude of switching, the number of bits in the bit stream may increase slightly. Analysis of the decoder may not increase the number of bits in the bitstream. However, the analysis in the decoding device can have additional complications. Thus, in embodiments, the resolution LPC analysis may be improved by reducing the spectral dynamic range, i.e. the frames of the signal can undergo preliminary processing through the filter compensation predicate the rd. Reverse low frequency distortion can be used in the embodiment of the decoding device 200, and the audio encoding device 100 for receiving the excitation signal or frame area predictions required for encoding subsequent frames. All these filters can give the response in the absence of the input signal, i.e. the signal at the output of the filter due to the influence of the current input, which is not the previous inputs, i.e. under the condition that information about the state of the filter is set to zero after reset. Generally, when an LPD mode of coding is working properly, information about the state of the filter is updated in the final state after filtration of the previous frame.

In the embodiments to set the state of the internal filter, LPD is encoded in such a way that the first LPD frame all filters and predictors are initialized to operate in an optimal or improved modes for the first frame, or information about the ratio of the switch /[or] the coefficients can be represented by the audio encoding device 100, or in the decoding device 200 can be carried out additional processing.

As a rule, filters and predictors for analysis, implemented in the audio encoding device 100 for use in the analysis stage of coding prediction 110, different from the filters and predictors, used for synthesis in the device audio decoder 200.

For this analysis, as, for example, the stage of analysis coding prediction 110, all or at least one of these filters you can apply the corresponding original samples of the previous frame to update the memory. Figa shows an embodiment of a filter structure used for analysis. The first filter is a filter to compensate for pre-emphasis 1002, and can be used to improve the resolution filter LPC analysis 1006, i.e. stage of analysis coding prediction 110. In embodiments, the filter LPC analysis 1006 can accurately calculate or estimate the short-term filter coefficients using, for example high-frequency filtering of the speech samples within the analysis window. In other words, in embodiments, the controller 140 may be adapted to determine information about the coefficient switching on the basis of the high-frequency filtering of the decoded frame spectrum of the previous frame. Similarly, assuming that the analysis is carried out using the embodiment of the device audio decoder 200, the controller 250 can be adapted for analysis of the high-frequency filter of the previous frame.

As shown in figa, filter LP analysis 1006 is preceded by a filter assess perceptions 1004. In embodiments, the filter estimates in the acceptance 1004 can be used when searching for code tables in the analysis-synthesis. The filter can use masking noise properties of the formant, such as, for example, the resonances of the vocal [voice] tracts, by assessing the reduction of errors in areas close to the formant frequencies and increase in areas distant from them. In embodiments, the encoder reduce redundancy 150 can be applied to encode based on the code tables, adapted to the respective frame region prediction/frames. Accordingly, the decoder introducing redundancy 210 may be adapted to decode based on the code table that is adapted to the sampling frames.

Fig.9b illustrates a block diagram of the signal processing in the case of synthesis. In the case of synthesis, in embodiments, all or at least one of the filters you can apply the corresponding synthesized sample of the previous frame to update the memory. In embodiments, the device audio decoder 200, it can be just as directly accessible synthesis of the previous non-LPD frames. However, in the embodiment, the audio encoding device 100, the synthesis can not be done by default and, accordingly, the synthesized sample may not be available. Thus, in embodiments, the audio encoding device 100, the controller 140 may be adapted to decode the previous not-LPD frame. After decoding the non-LPD frame at about the both variants, i.e. devices audio coding 100 and 200, the synthesis of the previous frame can be carried out in accordance with figb in block 1012. In addition, the output of filter synthesis LP 1012 can be entered into the inverse filter estimate perception 1014, after which the filter is applied to the pre-emphasis compensation 1016. In embodiments adapted code table can be used and is filled with synthetic samples from the previous frame. In other embodiments, adapted code table can contain the excitation vectors that are adapted for each subframe. Adapted code table can be obtained from the long-term filter state. Delay values can be used as an index in an adapted code table. In variants of the embodiment for filling adapted code table, the excitation signal or the differential signal may, as a result, to be computed by filtering the sample weighted signal by using the inverse filter weighing [evaluation] with zeroed memory. In particular, the excitation may be necessary in the encoding device 100 in order to update the long-term predictor of memory.

Embodiments of the present invention can provide the advantage that the restart procedure, the filter can be improved or ush the Ren by providing additional parameters and/or download the internal memory device encoding or decoding the samples of the previous frame, encoded by the encoder on the basis of the conversion.

The embodiment can provide an advantage in accelerating the start of the procedure the basic LPC encoding by updating all or part of the respective memory blocks, resulting synthesized signal may be closer to the original [source] the signal than when using conventional concepts, particularly through the use of a full reset. In addition, options may use large overlap and more open and thereby allow more efficient use of the abolition of the temporary overlay. Embodiments may have the advantage that the unsteadiness of the phase of the encoder speech can be reduced, and the resulting artifacts during the transition from the encoder based on the transformation of the speech encoder can also be reduced.

Depending on the specific requirements to implement the proposed method, the methods of the invention can be implemented in hardware or in software. The implementation may be performed using digital media, in particular, DVDs, CD with electronically readable control signals stored on them, which interact (or are able to work together) with a programmable computer system so that you meet the compliance is adequate methods.

Therefore, the present invention is a program product with program code stored on a machine-readable carrier. When the computer program product runs on a computer, the program code executes one of the methods. In other words, methods of the invention are a computer program having a program code for performing at least one of the methods of the invention when the computer program runs on a computer.

Although the previous version of the invention has been shown and described in detail with reference to specific embodiments, for specialists in this area should be clear that they can be made of various other changes in form and detail without departing from the essence and content of his presentation. It should be understood that various changes may be made in adapting to various variants, without departing from the General concept described here and presented in the provisions that follow.

1. The audio encoding device (100)used to encode the frames presented in the form of samples of the audio signal to obtain encoded frames in which the frame consists of a set of audio samples in the time domain, comprising the step of analysis when coding with prediction (110) for determining information on coefficients of filter synthesis and the frame information region prediction based on a frame of audio samples; Converter frequency domain (120) for converting a frame of audio samples in the frequency domain and obtain a spectrum of the frame; the transmitter coding (130) to determine how the coding region: encoded whether the data of the current frame based on the information about the filter coefficients of the synthesis and the frame information region predictions, or data based on the spectrum of the frame; a controller (140) for determining information about the coefficient switching from the transformation to the field of prediction, when the transmitter coding determines that the encoded data of the current frame based on the filter coefficients of the synthesis and the frame information region predictions, and the transmitter coding specifies when the encoded data of the previous frame have been encoded on the basis of the previous spectrum of the frame obtained by the conversion in the frequency domain; and an encoder redundancy reduction (150) for encoding the frame information region predictions, information on odds, information on odds of switching and/or spectrum of the frame, and information about the coefficient switching includes information that allows you to perform the initialization stage of the synthesis prediction, and the controller (140) is adapted to determine information about the PE ratio is clucene based on an LPC analysis of the previous frame, the controller (140) is adapted to determine the information about the ratio of switch-based high-frequency filtering version of the decoded spectrum of the previous frame.

2. The audio encoding device (100) according to claim 1, in which the step of analyzing the coding of the prediction (110) adapted to determine information about the filter coefficients of the synthesis and the frame information region prediction based coding linear prediction LPC analysis and the inverter frequency region (120), and a Converter adapted to convert a frame of audio samples based on the fast Fourier transform (FFT) or modified discrete cosine transform (MDCT).

3. The audio encoding device (100) according to claim 1, in which the controller (140) is adapted to determine the information about the ratio of the switch when the transmitter coding determines that the encoded data of the current frame based on information about the ratio, and the controller (140) is adapted for determining information on coefficients for a filter synthesis and information about switching frame region prediction based on the LPC analysis.

4. The audio encoding device (100) according to claim 1, where the controller (140) is adapted to determine information about the coefficient switching, the rate of switching is pramudya samples overlay of previous frame.

5. The audio encoding device (100) of claim 4, in which the frame of samples that is applied to the previous frame has a center at the end of the previous frame.

6. The encoding of the frames presented in the form of samples of the audio signal to obtain encoded frames, and the frame includes a number of samples in the time domain, comprising the steps of identifying information about the filter coefficients of the synthesis and the frame information region prediction-based frame of samples; converting a frame of audio samples in the frequency domain to obtain a spectrum of the frame; a decision based on whether the encoded data for a frame on the information on the coefficients and on the frame information field prediction or data based on the spectrum of the frame; determining information about the coefficient switching, when it was decided that the encoded data of the current frame based for information on the coefficients and the frame information region prediction when encoding data of the previous frame on the basis of the spectrum of the previous frame obtained by the conversion in the frequency domain; and encoding the frame information region predictions, information on odds, information on odds of switching and/or spectrum of the frame, and information about the coefficient switching includes information that allows in order to cialisabout phase synthesis prediction, and the definition information on the ratio of switching is based on the LPC analysis of the previous frame, and the controller (140) is adapted to determine the information about the ratio of switch-based high-frequency filtering version of the decoded spectrum of the previous frame.

7. Device audio decoder (200) for decoding encoded frames to obtain the frames presented in the form of samples of the audio signal, and the frame consists of several samples in the time domain, including a decoder receiving redundancy (210) for decoding encoded frames, and information about the frame region predictions, information on coefficients for a filter of the synthesis and/or spectrum of the frame; synthesis prediction (220) to determine frame prediction audio samples on the basis of information about the coefficients for filter synthesis and the frame information region prediction; time domain Converter (230) to convert the spectrum of the frame to a temporary the area to receive the converted frame from the frame spectrum; adder (240) for combining the converted frame and frame prediction to obtain a frame is presented in the form of samples of the audio signal; and a controller (250) to control the switching process, the switching process is carried out, the EU and the previous frame based on the converted frame, and the current frame based on the frame prediction, the controller (250) is configured to obtain a coefficient switching for the preparation of the initialization stage of the synthesis prediction (220), by estimating the LPC filter corresponding to the end of the previous frame so that the phase of the synthesis prediction (20) is initialized, when the switching process.

8. Device audio decoder (200) of claim 7, in which the decoder receiving redundancy (210) adapted to decode the information on the ratio of switching from the encoded frames.

9. Device audio decoder (200) according to claim 7, where the stage of synthesis prediction (220) is adapted to determine a frame prediction on the basis of the LPC synthesis and/or the time domain Converter (230), and it is adapted to convert the spectrum of the frame in the time domain on the basis of inverse FFT or inverse MDCT.

10. Device audio decoder (200) according to claim 7, where the controller (250) adapted for the analysis of the previous frame and the information of the previous frame using the coefficients for filter synthesis and information of the previous frame using the frame region predictions, and the controller (250) is adapted to provide information of the previous frame using the coefficients of the synthesis step forecast of the Oia (220) to provide information of the previous frame as a ratio of the switch and/or controller (250), moreover, the controller adapted to provide information about the previous frame using the frame area predictions for stage synthesis prediction (220).

11. Device audio decoder (200) according to claim 7, in which stage of the synthesis prediction (220) adapted to determine a frame prediction switch, the middle of which is located at the end of the previous frame.

12. Device audio decoder (200) according to claim 7, in which the controller (250) adapted for analysis using high-frequency filtering version of the previous frame.

13. The method of decoding encoded frames to obtain the frames presented in the form of samples of the audio signal, and the frame consists of several samples in the time domain, which includes stages of decoding encoded frames to obtain information about the frame region predictions, as well as information on coefficients for a filter of the synthesis and/or spectrum of the frame; the frame predictions of audio samples based on the information about the coefficients for filter synthesis and the frame information region prediction; transformation of the spectrum of a frame in the time domain to obtain a frame prediction of the spectrum of the frame; combining frame conversion and frame prediction to obtain the frame presented in the form of samples the sound of the first signal; and process control switching of the switching process carried out, if the previous frame based on the frame conversion, and the current frame based on the frame prediction; obtaining coefficient switching to initialize, through assessment in the LPC filter corresponding to the end of the previous frame so that the stage of synthesis of the prediction is initialized when the process switch.

14. The computer-readable storage medium stored thereon a computer program having a program code for performing the method of claim 6, when the computer program runs on a computer or processor.