Method and apparatus for selective signal coding based on core encoder performance. RU patent 2504026.

FIELD: information technology.

SUBSTANCE: in a selective signal encoder, an input signal is first encoded (1004)using a core layer encoder to produce a core layer encoded signal. The core layer encoded signal is decoded (1006) to produce a reconstructed signal, and an error signal is generated (1008) as the difference between the reconstructed signal and the input signal. The reconstructed signal is compared (1010) with the input signal. One of two or more enhancement layer encoders is selected (1014, 1016) depending on the comparison and is used to encode the error signal. The core layer encoded signal, the enhancement layer encoded signal and a selection indicator are output (1018) to a channel (e.g., for transmission or storage).

EFFECT: high-quality speech and audio reproduction at acceptable low data rates.

18 cl, 10 dwg

Transmission of text, image, voice and speech signals via the communication channels, including the Internet, is expanding rapidly as the provision of multimedia services that can host different types of information, such as text, images, and music. Multimedia signals, including speech and music signals require a lot of bandwidth during transmission. Thus, for the transmission of multimedia data, including text, images and audio, it is highly desirable that these data were compressed.

Compression of digital speech and audio signals are well known. Compression is generally required to effectively transmit signals via communication channels or store the compressed signals on digital media devices, such as computer hard disk or solid state memory device.

A fundamental principle of data compression is to eliminate redundant data. Data can be compressed by eliminating redundant information, for example, when the sound is repeated, is the predictable or redundant. Taking into account insensitivity person to high frequencies.

As a rule, compression results in distortions of a signal, and higher compression rates lead to significant distortions. Bit stream is called scalable, when part of the flow may be removed in such a way that the resulting subflow generates another valid stream of bits for some target decoder, and subflow is content with the quality of the recovery, which is lower than the quality of the original full stream of bits, but is high, whereas a smaller number of remaining data. Bit streams that do not provide this property, called siblings streams of bits. Conventional modes scalability are temporal, spatial scalability scalability and quality. Scalability allows you to adjust the compressed signal for best performance in a limited bandwidth channel.

Scalability can be implemented in such a way that provides for several levels of encryption, including baseline and at least one level of expansion, and the corresponding levels are generated to have different permissions.

Although many coding schemes are generalized some coding schemes include signal model. In General, the best signal compression is achieved when the model characterizes the encoded signal. Thus known the choice of coding scheme based on the classification of the signal type. For example, the voice signal can be simulated and coded in a different way than the musical signal. However, the classification of the signal in the General case, is a complex task.

Example of how compression (or coding), which remains very popular for digital coding of speech, known as linear prediction with the coded excitation (CELP), which is one of the family of algorithms of encoding by “analysis by synthesis”. Analysis by synthesis in the General case refers to the process of encoding, where a number of parameters of digital models are used for the synthesis of a set of signals candidates, which are compared with the input signal and assessed for distortion. The set of parameters that provide the most low distortion, then either sent or saved and ultimately is used to restore the evaluation of the original input signal. CELP is a specific method of analysis by synthesis, which uses one or more code books, each of which essentially contains many codes-vectors that is extracted from the codebook in response to the index in the code book.

In modern CELP encoders there is a problem with the maintaining of high-quality reproduction of speech and audio is acceptable low speeds data. This particularly applies to music or other generic audio signals, which do not correspond to CELP voice model with a high degree of accuracy. In this case, the discrepancy of the model can cause a serious deterioration in the quality of audio that may be unacceptable to the end user of the equipment that uses similar methods.

Brief description of drawings

Illustrating the drawings, in which the same reference positions are functionally identical or similar items on individual species, and that, together with detailed descriptions below, are included in the structure and form part of the specification, serve as a further illustration of the different variants of implementation and explain the various principles and benefits of in accordance with the present invention.

Figure 1 - block diagram of coding and decoding system of prior art.

Figure 2 - block diagram of coding and decoding system in accordance with some of the options for carrying out the invention.

Figure 3 is a block diagram of the method of selection of the coding system in accordance with some of the options for carrying out the invention.

Figure 4-6 - number of graphs showing the for example signals in comparator/selector in accordance with some of the options for carrying out the invention, when you enter the speech signal.

Fig.7.-9 - number of graphs showing the for example signals in comparator/selector in accordance with some of the options for carrying out the invention, when you enter the musical signal.

Figure 10 is a block diagram method for selective coding signal in accordance with some of the options for carrying out the invention.

The specialists in this field should be clear that the elements of the drawings are illustrated for simplicity and clarity, and need not be to scale. For example, the size of some elements of the drawings can be overstated relative to other elements that contribute to a better understanding of options for the implementation of the present invention.

Detailed description

Before a detailed description of the options implemented in accordance with the present invention, it should be noted that the options for implementation are mainly in combinations stages of the method and the device components related to the selective coding signal based on the corresponding model. Accordingly, components and stages of the method are presented, where appropriate, the usual symbols on drawings, showing only those particular did, which are essential for the understanding of options for the implementation of the present invention, in order not to clutter the disclosure of the invention details, which will be obvious to specialists in the field of technology on the basis of the description.

In this document relative terms such as «first» and «second», «upper» and «lower» and etc. can be used exclusively to distinguish one object or action of another object or action, without the necessity of any actual such relationship or order between such objects or actions. The terms “includes”, “contains” or any other variations thereof are intended to cover non-exclusive turn-on, so that process, the way a product or device that contains a list of items includes not only those elements, but can include other elements that are not listed explicitly or inherent in such a process, method, product or device. Element, which is preceded by the word “contains...”, without major restrictions, does not preclude the existence of additional identical elements in the process, the way the product or your device, which include this element.

It should be noted that the options for carrying out the invention described here may include one or more conventional processors and unique saved programming statements that manage one or more processors to implement, in connection with some schemes, some, most, or all, of the functions of selective coding of the signal on the basis of fit of the model described here. Alternatively, some or all of the functions can be implemented through the state machine that does not have saved programming instructions, or in one or more specific integrated circuits (ASIC), in which each function or some combination of certain functions are implemented as custom logic circuits. Of course, it could be used combination of the two approaches. Thus, here are the ways and means to perform these functions. In addition, it is expected that the specialist, the significant efforts and a large number of design options, motivated, for example, disposable time, modern technology and economic considerations, guided by the concepts and principles described here, will be able to create such software manuals and programmes and chips with a minimum amount of experimentation.

Fig. 1 is a block diagram of the integrated system of coding and decoding 100 prior art. Figure 1 source s(n) 102 is input to the encoder, 104 basic level of coding systems. Coder 104 basic level encodes the signal 102 and generates a coded signal 106 basic level. In addition, the source signal 102 is entered in the coder 108 extension level coding system. Coder 108 level expansion also takes the first restored s c (n) 110 as the input. The first restored signal 110 formed by passing the encoded signal 106 basic level through the first decoder 112 basic level. Coder 108 level of expansion is used to encode more information, based on some comparison of signals s(n) (102) and s c (n) (110) and may optionally use the settings from the encoder 104 basic level. In one embodiment, the encoder 108 extension level encodes the error signal, which is the difference between the restored signal 110 and input 102. Coder 108 extension level generates a coded signal 114 extension level. As a coded signal 106 basic level, and coded signal 114 level of expansion is transferred into the channel 116. Channel is the environment, such as the channel of communication and/or storage media.

After passing through the second channel of the restored signal 118 formed by transmitting the received encoded signal 106' basic level through the second decoder 120 basic level. The second decoder 120 basic level performs the same function as the first decoder 112 basic level. If the coded signal 114 level of expansion is also transmitted through a channel 116 and is taken as a signal 114', it can be transmitted by the decoder 122 extension level. Decoder 122 extension level also accepts a second restored signal 118 as input and produces a third of the restored signal 124 as an output. Third, the recovered signal 124 is consistent with the original signal 102 more accurately than the second restored signal 118.

Coded signal 114 level expansion includes additional information that allows you to restore the signal 102 more accurately than the second restored signal 118. That is, this is advanced (improved) recovery.

One of the advantages of this integrated system of encoding is that a particular channel 116 is not able to consistently support the requirement for bandwidth associated with the algorithms of the coding of high quality audio. Built-in encoder, however, allows taking part of a stream of bits (for example, only a stream of bits of base level) from the channel 116 for the formation of, for example, only a basic audio output when the stream of bits extension level is lost or distorted. However, there are compromises in quality between the embedded and encoders, and between the various goals optimization built-in encoding. That is, the encoding extension level higher quality can contribute to a better balance between the base level and expansion, and reduce the overall speed of data transmission to improve the characteristics of transmission (for example, reducing the overload), which may reduce the frequency of occurrence of packet errors for levels of expansion.

Although many coding schemes are generalized some coding schemes include signal model. In General, the best signal compression is achieved when the model is coded signal. Thus, it is known that the encoding scheme is selected after the classification type of the signal. For example, the voice signal can be simulated and coded in a different way than the musical signal. However, signal classification, as a rule, is a difficult task.

Figure 2 is a block diagram of the system 200 encoding and decoding in accordance with some of the options for carrying out the invention. According to Figure 2 source signal 102 is entered in the coder 104 basic level coding system. Original signal 102 may be voice/audio signal or another type of signal. Coder 104 basic level encodes the signal 102 and generates a coded signal 106 basic level. The first restored signal 110 formed by passing the encoded signal 106 basic level through the first decoder 112 basic level. Original signal 102 and the first restored signal 110 are compared in module 202 comparator/selector. Module 202 comparator/selector compares the source signal 102 with the first restored signal 110 and, based on a comparison, forms the signal selector 204, which chooses which of coders 206 extension level to use. Although only two of the encoder extension level is shown on the drawing, it should be clear that can be used many coders extension level. Module 202 comparator/selector 202 choose the encoder level extensions that are most likely to form the best restored signal.

Although decoder 112 expansion is shown as a host coded signal 106 basic level, accordingly, sent to the channel 116, the physical connection between items 104 and 106 may allow for a more efficient implementation, so that the General elements of the processing and/or status can be shared, and thus, it would not have regeneration or duplication.

Each coder 206 extension level takes the original signal 102 and the first electric signal as input (or a signal such as a differential signal received from these signals), and the selected encoder generates a coded signal 208 extension level. In one embodiment, the encoder 206 extension level encodes the error signal, which is the difference between the restored signal 110 and input 102. Coded signal 208 extension level contains additional information based on comparing the signals s(n) (102) and s c (n) (110). In addition, it can use the settings from the decoder 104 basic level. Coded signal 106 basic level, a coded signal 208 extension level and signal selector 204 all are transferred to the channel 116. Channel is the environment, such as the channel of communication and/or storage media.

After passing through the second channel of the restored signal 118 formed by transmitting the received encoded signal 106' basic level through the second decoder 120 basic level. The second decoder 120 basic level performs the same function as the first decoder 112 basic level. If the coded signal 208 level of expansion is transmitted through a channel 116 and is taken as a signal 208', it can be transmitted by the decoder 210 extension level. Decoder 210 extension level also accepts a second restored signal 118 and adopted selector signal 204' as input and produces a third of the restored signal 212 as output. Work decoder 210 extension level depending on the decision gate signal 204'. Third, the recovered signal 212 is consistent with the original signal 102 more accurately than the second restored signal 118.

Coded signal 208 level of expansion to include additional information, therefore the third restored signal 212 consistent with the signal 102 more accurately than the second restored signal 118.

Figure 3 is a block diagram method for the selection of the coding system in accordance with some of the options for carrying out the invention. In particular, figure 3 describes the comparator/selector in the embodiment of the invention. After a block 302 beginning of the input signal (102 figure 2) and the recovered signal (110 figure 2) is converted, if necessary, in the selected area of the signal. Signals temporary area can be used without conversion, or in the 304 signals can be converted in the spectral range, such as frequency region of the modified discrete cosine transform () or area, and may also be processed by other additional elements, such as perceptual weighting certain frequency or temporal characteristics of the signals. The converted input (or a temporary area) indicated as S(k) for the spectral component k, restored and converted signal (or a temporary area) is denoted as S c (k) for the spectral component k. For each component k in the selected set of components (which can be of some or all of the components) energy E_tot in all components of the S c (k) of the recovered signal is compared with the energy E_err components that are greater than (for example, by a factor)than the corresponding component S(k) of the original input signal.

While the components of the input and the restored signal may significantly vary in amplitude, a significant increase in amplitude component of the recovered signal indicate poorly simulated input signal. As such, a component of the recovered signal lesser amplitude can be compensated by the specified encoding extension level, while the component of the recovered signal of greater amplitude (i.e. poorly modeled) may be better for alternative ways of encoding extension level. One such alternative method of encoding extension level can use decreased energy of some components of the restored signal before encoding extension level, so audible noise or distortion resulting from the mismatch signal model of the basic level are reduced.

According to figure 3 cycle component is initialized in the block 306, where component k initialized, and measures energy E_tot and E_err are initialized to zero. In the final unit 308, a check is made to determine whether the absolute value of the component of the restored signal is considerably greater than the corresponding component of the input signal. If it significantly more, as indicated by the positive branch from the casting unit 308, the component is added to the energy of the error E_err in the block of 310, and processing jumps to the 312. In the block 312 component of the recovered signal is added to the full value of energy E_tot. In the final block 314 component value is incremented, and a check is performed to determine if all components have been processed. If not, as shown negative branch of solving block 314, processing returns to unit 308. Otherwise, as shown positive branch from solving block 316, the loop ends, and the total cumulative energy is compared in the final block 316. If the energy of the error E_err significantly lower than the total energy E_tot, as shown negative branch of solving block 316, in block 318 level is an extension of type 1. Otherwise, as shown positive branch from solving block 316, in block 320 level is an extension of type 2. Processing of this block, the input signal is terminated in a block 322.

For specialists in a given field of technology, it should be obvious that other measures the signal energy that can be used, for example, the absolute value of the component to some degree. For example, the energy component Sc(k) can be estimated as |Sc(K)| p , and the energy component, S(k) can be estimated as |Sc(K)| p , where P is a number greater than zero.

Can be added stage of hysteresis, so that the type of extension level change only if a certain number of blocks signal are of the same type. For example, if you are using type 1 encoder, type 2 is selected until two consecutive block does not show the use of type 2.

Figure 4-6 is a series of graphs showing the for example, the results for the speech signal. Schedule 402 figure 4 shows the energy E_tot reconstructed signal. Energy is calculated for 20 millisecond frame, so that the graph shows the change in energy of a signal for 10 seconds. Schedule 502 figure 5 shows the ratio of the energy of the error E_err to the total energy E_tot over the same period. Threshold Thresh2 shown as a dotted line 504. The speech signal in frames where the ratio is greater than the threshold does not adequately modeled by the encoder. However, for most of frames threshold is not exceeded. Schedule 602 6 shows the selector or decisive signal for the same period of time. In this example, a value of 0 indicates that the selected coder-level extension of type 1, and a value of 1 specifies that the selected coder level type extensions 2. Isolated shots where the ratio is higher than the threshold, are ignored, and the selection is changed only when two successive frames indicate the same choice. For example, the encoder extension level type 1 is selected for the frame 141, even if the ratio exceeds the threshold.

Fig.7.-9 show the appropriate number of graphs of a musical signal. Schedule 702 figure 7 shows the energy E_tot input signal. Again, energy is calculated for 20 millisecond frame, so that the graph shows the change of the input energy for 10 second interval. Plot 802 on Fig.8 shows the ratio of the energy of the error E_err to the total energy E_tot over the same period. Threshold Thresh2 shown as a dotted line 504. The musical signal in frames where the ratio is greater than the threshold does not adequately simulated by the encoder. This is the case for the majority of personnel, since the reference encoder designed for speech signals. Schedule 902 figure 9 shows the selector or decisive signal for the same period of time. Again, a value of 0 indicates that the selected coder-level extension of type 1, and a value of 1 specifies that the selected coder level type extensions 2. Thus, the encoder extension level type 2 is selected for most of the time. However, in the frames where the reference encoder works well for music signal is selected coder-level extension of type 1.

In the test for 22803 frames of the speech signal level coder extension of type 2 has been selected only 227 personnel, that is, only 1% of the time. In the test for 29644 frames of the music signal level coder extension of type 2 has been selected in 16145 frames, that is 54% of the time. In other frames reference encoder works well for music signal, and for the speech signal is selected coder extension level. Thus, the comparator/selector switch is not a classifier speech/music signal. This is in contrast to previous schemes that seek to classify input as speech or music, and then choose the encoding scheme accordingly. Proposed approach consists in the choice of encoder level of extension depending on the performance of the encoder basic level.

Figure 10 is a block diagram illustrating the work of the embedded coder in accordance with some of the options for carrying out the invention. The flowchart shows the method used for encoding one frame signal data. Frame length is selected depending on the temporal characteristics of the signal. For example, 20 MS frame can be used for speech signals. After a block 1002 beginning 10 input signal is encoded in the block 1004 using the encoder base level for the formation of the encoded signal at the basic level. In the block 1006 coded signal basic level decoded for the formation of the recovered signal. In this variant of the implementation of the error signal is generated in the block 1008 as the difference between the restored signal and the input signal. The recovered signal is compared with the input signal in the unit 1010, and in the final block 1012 determined whether the agreed restored signal to the input signal. If negotiation good, as shown positive branch from solving block 1012, encoder extension level type 1 is used for coding of the error signal in the block 1014. If approval is not good, as shown negative branch of solving block 1012, encoder extension level type 2 is used for coding of the error signal in the block 1016. In the block 1018 coded signal basic level, a coded signal level enlargement and the selection pointer is displayed in the channel (for example, for transport or storage). Processing of the frame ends in a block 1020.

In this embodiment, the encoder extension level responds to signal the error, however, in the alternative to the encoder extension level responds to the input signal, and additionally, if necessary, one or more signal from the encoder to the baseline and/or decoder basic level. In another embodiment, the alternative signal errors, such as weighted difference between the input signal and the restored signal. For example, some of the frequencies of the recovered signal can be attenuated to signal errors. The resulting error signal can be referred to as the weighted error signal.

In another alternative implementation of the encoder and decoder basic level could also include other levels of expansion, and the comparator according to the present invention may receive, as input, the output of one of the previous levels of expansion, as reconstructed signal. In addition, there may be levels of expansion, subsequent concerning the above-mentioned levels of expansion, which may or may not switch from a comparison. For example, built a coding system can contain five levels. Basic level (L1) and second level (L2) may establish the restored signal S c (k). The recovered signal S c (k) and the input signal S(k) can be used to select ways of encoding extension level at levels 3 and 4 (L3, L4). Finally, the level of 5 (5) may include only one way of coding extension level.

The encoder may choose between two or more encoders level of development depending on the comparison between the restored signal and the input signal.

Encoder and decoder can be implemented at the programmed processor processor or, for example, the specialized (oriented on application) integrated circuit.

The preceding description of the described specific embodiments of the invention. However, specialist in the art must be understood that various modifications and changes can be made without deviating from the volume of the present invention, as set out in the claims. Accordingly, the description and drawings should be considered illustrative sense and not as a limitation, all such changes should be included in the scope of the present invention. The benefits, advantages, solving problems and any(s) of the item(s)that may determine that any benefit, advantage or decision occur, or more pronounced, should not be viewed as critical, required or essential characteristics or features of any or all of the claims. Invention is determined solely by the claims, including all changes made during the consideration of this application, and all equivalents claims, which followed the issuance of a patent.

1. The encoding of the input signal, containing: the encoding of the input signal using the encoder base level for the formation of the encoded signal basic level; decoding of the encoded signal baseline for the formation of the recovered signal; a comparison of the reconstructed signal to the input signal, and the stage of the comparison includes the assessment of energy _ the recovered signal, which contains errors, the definition that the ratio of S(k)/S c (k) component S(k) of the input signal to the S c (k) of the recovered signal exceeds the threshold value, and the sum of the energies of these component S c (k) of the recovered signal when the ratio of S(k)/S c (k) component S(k) exceeds the threshold value; the choice of encoder extension level of many coders extension level depending on a comparison between the restored signal and the input signal; and the generation of the encoded signal level extensions with the selected encoder level extension, and a coded signal level expansion depends on the input signal.

2. The method according to claim 1, further comprising: the generation of the error signal as the difference between the restored signal and the input signal, and the generation of the encoded signal level expansion includes coding of the error signal.

3. The method of claim 2, in which the error signal contains a balanced difference between the restored signal and the input signal.

4. The method according to claim 1, wherein a comparison of the reconstructed signal input signal includes: assessment of energy E_tot as the accumulation of energies of all components of the recovered signal; assessment of energy _ as the accumulation of energies of all components of the recovered signal that contain errors; and comparison of the energy E_tot with energy _.

5. The method according to claim 4, additionally contains: the transformation of the recovered signal for the formation of the components of the recovered signal, and the conversion is selected from the group consisting of the Fourier transform, the modified discrete cosine transform (MDCT) and wavelet transform.

10. The method according to claim 1, additionally contains the output of the encoded signal a more basic level, the encoded signal level extensions and pointer selected encoder extension level in the channel.

11. Selective encoder signal, containing: the encoder basic level, which takes the input audio signal subject coding, and generates a coded signal basic level; decoder basic level, which takes a coded signal to the basic level in as input and generates the recovered signal; and many coders level extensions, each of which is selected for coding of the error signal, to form a coded signal level enlargement error signal contains the difference between the input signal and the restored signal; and comparator module/selector that selects encoder extension level of many coders extension level depending on a comparison of the input signal and the encoded signal basic level, this module comparator/selector evaluates the energy E_err reconstructed signal, which contains errors, defines the relation S(k)/S c (k) component S (k) of the input signal to the S c (k) of the recovered signal when the ratio of S(k)/S c (k) component S(k) exceeds the threshold, and, additionally, the input signal is encoded as a coded signal basic level, a coded signal level enlargement and the selection of the selected encoder extension level.

12. Selective encoder signal according to paragraph 11, in which the encoder basic level includes speech coder.

13. Selective encoder signal according to paragraph 11, in which the module comparator/selector: evaluates the energy E_tot as the accumulation of energies of all components of the recovered signal; evaluates the energy _ as the accumulation of energies of all components of the recovered signal that contain errors; and compares the energy E_tot with energy _.

14. Selective encoder signal according to paragraph 13, in which the module comparator/selector compares the energy E_tot with energy _ by comparing the ratio of energies E_err/E_tot threshold value.

15. Selective encoder signal according to paragraph 13, in which the components of the recovered signal and components of the input signal are calculated using the conversion, selected from the group consisting of the Fourier transform, the modified discrete cosine transform () and wavelet transform.

16. Selective decoder signal that contains a processor that includes instructions to decode the original signal, which is encoded as a coded signal basic level, a coded signal level extensions and a pointer to the selected encoder level enlargement decoder contains: decoder basic level, which takes a coded signal to the basic level in as input and generates the first restored signal; and decoder level of empowerment that receives the signal select to select the encoder level of empowerment that decodes the encoded signal level extensions to the formation of the second reconstructed signal.

17. Selective decoder signal to article 16, in which the second restored signal includes the error signal and in which the original signal is recovered in the form of the sum of the recovered signal and signal errors.

18. Selective decoder signal to article 16, in which the decoder extension level responds to the first and second recovered signals and coded signal level of expansion, and where the second restored signal is an estimate of the initial signal.