WO2009089728A1 - Method for high frequency band replication, coder and decoder thereof - Google Patents

Abstract

A method for high frequency band replication, which comprises: to filter the audio or voice signal to gain the low frequency sub-band and the high frequency sub-band; to determine the strategy of spectral band replication; to gain the correlation between said low frequency sub-band and high frequency sub-band according to said determined strategy of spectral band replication, and to select the low frequency sub-band as the replication frequency band for the high frequency sub-band according to said correlation, and to output the information of high frequency band replication parameters which includes the correspondence relation of the selected frequency band. The present invention else provides a method for high frequency band replication, which comprises: to receive the information of high frequency band replication parameters which includes the correspondence relation of the selected frequency band, and said correspondence relation of the selected frequency band is the relation between the high frequency sub-band and the corresponding low frequency sub-band thereof, which is determined by the correlation; In the high band, to replicate the high frequency sub-band by the low frequency sub-band according to said information of high frequency band replication parameters which includes the correspondence relation of the selected frequency band. Accordingly, the present invention provides a coder and decoder. The solution of the embodiment of the present invention can fulfill the high frequency replication more exactly.

Description

 High frequency reconstruction method, coding module and decoding module

 This application claims priority to Chinese Patent Application No. 200710305087.3, entitled "High Frequency Reconstruction Method, Encoding Module and Decoding Module", filed on December 27, 2007, the entire contents of which are incorporated by reference. In this application.

Technical field

 The present invention relates to the field of communications technologies, and in particular, to a high frequency reconstruction method, an encoding module, and a decoding module.

Background technique

 In audio and speech processing, high frequency reconstruction is a relatively critical technology. High-frequency compression and recovery technology represented by SBR (Spectral Band Replication) is a better high-frequency reconstruction method. It copies the waveform of the low-frequency band to the high-frequency section and then extracts the extracted frequency. The energy adjustment parameters and the harmonic adjustment parameters repair the copied high frequency band to achieve the purpose of high frequency reconstruction.

 There are two main methods for performing high frequency reconstruction using low frequency band signals in the prior art, please refer to the following introduction:

 Prior art one:

 The low frequency signal of audio or speech is passed through a digital filtering group to obtain a set of low frequency subband signals; and the low frequency subband group is used as a whole block signal to perform high frequency signal reproduction. The method for replicating the entire high frequency band signal is to divide the high frequency band into segments according to the frequency from low to high, each segment having substantially the same bandwidth as the whole low frequency signal; then continuously copying the entire low frequency subband group to the high frequency Each segment of the band. In this way, the entire low-frequency sub-band group will be periodically used in the high-frequency band several times until the entire high-frequency band that needs to be recovered is copied. There are two specific ways: 1) The whole low-frequency sub-band group Translating to the corresponding high frequency band, the method can be referred to FIG. 1 , which is a schematic diagram of the overall translational replication of a low frequency sub-band in the prior art; 2) folding the whole low frequency sub-band group first, that is, inverting the sub-band arrangement order, and then The whole low frequency sub-band group is translated to the corresponding high frequency band. This method can be seen in FIG. 2 , which is a schematic diagram of the overall folding translation of the prior art low frequency sub-band. During the copying process, modes 1) and 2) may be used interchangeably. Thus, the entire low frequency subband group is used periodically until the entire high frequency band that needs to be recovered is copied.

Please refer to FIG. 4, which is an energy waveform diagram of the prior art original audio and its sub-band signals (for straight In comparison, only the waveforms of the first 29 sub-bands are drawn in the figure). Figure 5 is a three dimensional view of the energy waveforms of the various subbands of the prior art original audio. FIG. 6 is an energy waveform diagram of each sub-band signal obtained by performing high-frequency reconstruction in the mode 1 of the prior art, and FIG. 7 is a corresponding three-dimensional diagram of each sub-band energy. 8 is an energy waveform of each sub-band signal obtained by performing high-frequency reconstruction in the mode 2 of the prior art, and FIG. 9 is a corresponding three-dimensional diagram of each sub-band energy. For the energy waveform diagram, the structure of the waveform shown therein is: The lowest waveform is the original audio waveform; the 0th to 8th curves are low frequency subband waveforms, which will be used to replicate the high frequency subband; Between 8 and 9 is the dividing line between high frequency and low frequency; all sub-bands represented upward from the 9th curve are the range of high frequency reconstruction and processing. For the energy three-dimensional map, the audio parameters described in the figure are the energy amplitude, the number of audio frames (30 frames) and the number of sub-bands (29). Corresponding to the processed 29 sub-bands shown in the energy waveform diagram. Among them, the ninth sub-band or more is a high-frequency processing part.

 Prior art 2:

 The low frequency subband is passed through the low pass filter bank to obtain a set of low frequency subbands. Here, as in the prior art 1, the selected low-frequency sub-band group is taken as a whole, and the high-frequency part to be recovered is continuously copied in the whole segment, and the sub-bands in the low-frequency sub-band group are respectively used to recover correspondingly. Some discretely distributed high frequency sub-bands.

 In the high frequency part, if there is a very rich harmonic component, the frequency of the harmonic component is often an integer multiple of its corresponding fundamental frequency. Accordingly, the prior art 2 proposes that if the subband numbers of some subbands of the high frequency portion are integer multiples of natural numbers such as 2, 3, 4, 5, in other words, some high frequency subbands and low frequency subbands There is a correspondence between multiples, then these sub-bands are likely to have rich harmonic components, which need to be restored.

 Please refer to FIG. 3 , which is a schematic diagram of discrete replication of the prior art two low frequency sub-bands. The entire frequency band of the audio signal is divided into thirty sub-bands by sub-band filtering (sub-band numbers are 0, 1, 2 in order)

 31, 32). The low frequency sub-band group includes eight sub-bands such as 0, 1, and 27, and the high-frequency sub-band group to be recovered includes twenty-five sub-bands such as 8, 9, 10, 31 and 32. Among them, the low frequency subband group provides four consecutive sub-bands each time to complete the copy.

The first thing to start is the replication process II. Because the serial numbers of the 4th, 10th, 12th, and 14th subbands in the high frequency subband group are all integer multiples of 2, the 4th, 5th, 6th, and 7th subbands are selected from the low frequency subband group, and sequentially correspond to Restore the four high frequency sub-bands of 8, 10, 12, and 14. Then there is replication process III. The number of the five sub-bands of the 9th, 12th, 15th, 18th, and 21th bands in the high-frequency sub-band group is an integer multiple of 3, but the 12th sub-band has just been copied, and its position also affects the continuity of these sub-bands. Then, the third sub-bands of the third, fifth, sixth, and seventh sub-bands are selected from the low-frequency sub-band group, and the four high-frequency sub-bands of the 9, 15, 18, and 21 are sequentially restored.

 This is followed by the replication process IV. The serial numbers of the 8th, 12th, 16th, 20th, 24th, 28th, etc. subbands in the high frequency subband group are all integer multiples of 4, but the 8th and 12th subbands have been copied, and then from the low frequency subband group. Four sub-bands of 4, 5, 6, and 7 are selected, and the four high-frequency sub-bands of the 16, 20, 24, and 28 are sequentially restored.

 Finally, the copy process. The serial numbers of the 10th, 15th, 20th, 25th, 30th, etc. subbands in the high frequency subband group are all integer multiples of 5, but the 10th, 15th, and 20th subbands have been copied, and only the low frequency subbands are needed. The sixth and seventh sub-bands are selected in the group, and the two high-frequency sub-bands of the 25th and the 30th are sequentially restored.

 Thus, the process of recovering the discretely distributed high frequency sub-bands with successive low frequency sub-band sets is completed. Finally, for the high-frequency sub-bands missed by the above method, the low-frequency sub-bands with similar waveforms are also selected to recover the missing high-frequency sub-bands, thereby completing the replication of all high-frequency sub-bands.

 In the research and practice of the prior art, the inventors have found that the prior art has the following problems: In the prior art, a low frequency sub-band according to the prior art is used as a whole block for periodic translational copying or folding replication. Or double-frequency reproduction according to the prior art 2, both mechanically recovering harmonics, without considering the diversity and variability of the audio voice signal, and additionally copying and copying according to the sub-band number in sequence, due to the low frequency sub-band and high The waveform of the frequency subband is originally different, so the reproduced high frequency sub-band may have a larger waveform difference or peak difference than the original high frequency sub-band, so the reconstructed high-frequency signal is not too high. If the waveform diagram mentioned above is observed, it can be found that the waveform reconstructed by the prior art method has a large difference between the original waveform and the original waveform; the energy waveform diagram mentioned above is observed, and the comparison result shows that the prior art method is lost after reconstruction. Many high frequency harmonics.

Summary of the invention

 The technical problem to be solved by the embodiments of the present invention is to provide a high frequency reconstruction method, an encoding module, and a decoding module, which can perform high frequency reconstruction more accurately.

 In order to solve the above technical problem, the embodiment provided by the present invention is implemented by the following technical solutions:

An embodiment of the present invention provides a high frequency reconstruction method, including: filtering an audio or voice signal to obtain a low frequency subband and a high frequency subband; determining a frequency band replication strategy; and replicating according to the determined frequency band The strategy acquires the correlation between the low frequency subband and the high frequency subband, selects a low frequency subband as a copy frequency band for the high frequency subband with reference to the determined correlation, and outputs high frequency reconstruction parameter information including a correspondence relationship of the selected frequency band. .

 The embodiment of the present invention provides a high-frequency reconstruction method, including: receiving high-frequency reconstruction parameter information including a correspondence relationship of selected frequency bands, where the correspondence relationship of the selected frequency bands is specifically corresponding to a high-frequency sub-band determined by reference correlation. The relationship of the low frequency sub-bands; the low frequency sub-band is copied as a high frequency sub-band according to the high frequency reconstruction parameter information including the correspondence relationship of the selected frequency bands in the high frequency band.

 An embodiment of the present invention provides an encoding module, including: an analysis filter module, configured to filter audio or voice signals to obtain a low frequency subband and a high frequency subband; and a frequency band selection module, configured to determine a frequency band replication strategy, Determining a correlation between the low frequency sub-band and the high frequency sub-band by using the determined frequency band replication strategy, selecting a low frequency sub-band as a copy frequency band for the high frequency sub-band with reference to the determined correlation, and outputting a correspondence relationship including the selected frequency band High frequency reconstruction parameter information.

 An embodiment of the present invention provides a decoding module, including a high frequency generator module, where the high frequency generator module includes: a receiving unit, configured to receive high frequency reconstruction parameter information including a correspondence relationship of selected frequency bands, where the selected frequency band is The correspondence relationship is specifically a relationship between the high frequency sub-band determined by the reference correlation and the corresponding low frequency sub-band; the reconstruction unit is configured to, in the high frequency band, the low frequency sub-band according to the high frequency reconstruction parameter information including the correspondence relationship of the selected frequency band Copy as a high frequency sub-band.

 The foregoing technical solution can be seen that the scheme of the embodiment of the present invention fully considers the correlation between the low frequency subband and the high frequency subband, and obtains the correlation between the low frequency subband and the high frequency subband according to the determined frequency band replication strategy. The determined correlation is that the high frequency sub-band selects the low frequency sub-band as the copy frequency band, and outputs the high frequency reconstruction parameter information including the correspondence relationship of the selected frequency bands, so that the high frequency reconstruction can be performed based on the high frequency reconstruction parameter information. The energy waveform and the energy three-dimensional image of each sub-band signal obtained after the high-frequency reconstruction can be clearly found that the reconstructed high-frequency portion of the embodiment of the present invention is closer to the original audio signal, and the effect is better than the prior art solution. Therefore, the technical solution of the embodiment of the present invention can perform high frequency reconstruction more accurately.

DRAWINGS

 1 is a schematic diagram of an overall translational replication of a low frequency sub-band of the prior art;

 2 is a schematic diagram of the overall folding translation of a low frequency sub-band of the prior art;

3 is a schematic diagram of discrete replication of a prior art two low frequency sub-band; 4 is an energy waveform diagram of a prior art original audio and its respective sub-band signals;

 5 is a three-dimensional energy diagram of energy waveforms of each sub-band of the prior art original audio;

 6 is an energy waveform diagram of each sub-band signal obtained by performing high-frequency reconstruction in the manner of the prior art 1;

 7 is a three-dimensional diagram of energy of each sub-band obtained by performing high-frequency reconstruction in the manner of the prior art;

 8 is an energy waveform diagram of each sub-band signal obtained by performing high-frequency reconstruction in the manner of the prior art;

 9 is a three-dimensional diagram of energy of each sub-band obtained by performing high-frequency reconstruction in the manner of the prior art;

 10 is a block diagram showing the principle and structure of a high frequency reconstruction according to an embodiment of the present invention;

 11 is a schematic diagram of a high frequency segmentation low frequency matching replication strategy according to an embodiment of the present invention;

 12 is a schematic diagram of a high frequency matching replication strategy for a low frequency dominant frequency band according to an embodiment of the present invention; FIG. 13 is a schematic diagram of a high frequency matching replication strategy for a low frequency band according to an embodiment of the present invention;

 14 is a schematic diagram of an extended replication policy according to an embodiment of the present invention;

 Figure 15 (a) is a structural block diagram of an adaptive band copying mode of an encoding end according to an embodiment of the present invention; Figure 15 (b) is a structural block diagram of a fixed-band copying mode of an encoding end of the embodiment of the present invention; Figure 16 is an adaptive embodiment of the present invention; FIG. 17 is a flow chart of selecting a frequency band replication strategy according to an embodiment of the present invention; FIG.

 18 is a flowchart of optimal frequency band selection in an embodiment of the present invention;

 19 is a flowchart of an algorithm for detecting time-varying characteristics according to an embodiment of the present invention;

 20 is a schematic diagram of a high frequency reconstruction performed by a decoding end according to parameter information of an encoding end according to an embodiment of the present invention;

 21 is a flowchart of an algorithm for decoding a high frequency generator according to an embodiment of the present invention;

 22 is an energy waveform diagram of the original audio signal after the method of the embodiment of the present invention is restored; FIG. 23 is a three-dimensional diagram of the energy of the original audio signal recovered by the method of the embodiment of the present invention; FIG. 24 is a schematic structural diagram of the encoding module according to the embodiment of the present invention; ;

 25 is a schematic structural diagram of an encoding module 2 according to an embodiment of the present invention;

FIG. 26 is a schematic structural diagram of a decoding module according to an embodiment of the present invention. detailed description

 The embodiment of the invention provides a high frequency reconstruction method, which can perform high frequency reconstruction more accurately. Please refer to FIG. 10, which is a schematic diagram and a structural block diagram of a high frequency reconstruction according to an embodiment of the present invention.

 As shown in Fig. 10, the upper part is the module related to high frequency processing in the encoding end, and the lower part is the module related to high frequency processing in the decoding end.

 At the encoding end, the original audio signal is converted by the analysis filter module into subband signals distributed in different frequency bands, including low frequency subbands and high frequency subbands, and the low frequency subbands are encoded by the core encoder and transmitted to the decoding end. The low frequency sub-bands are also processed to obtain parameter information that guides the high frequency reconstruction. Specifically, at the encoding end, the low frequency sub-band passes through the analysis and detection module, and the obtained detection result is input to the frequency band selection module for guiding the analysis strategy of the frequency band selection module; the frequency band selection module selects according to the guidance information of the analysis and detection module. A suitable replication strategy, and using the maximum correlation criterion to select a matching low frequency subband for each or each high frequency subband, and the envelope parameter extraction module performs envelope parameter extraction, and finally outputs the parameter information of the high frequency reconstruction.

 At the decoding end, the core decoder decodes and recovers the low-frequency sub-band signal by using the encoded information of the received low-frequency sub-band, and then copies the high-frequency sub-band according to the parameter information of the high-frequency reconstruction from the encoding end, and then performs envelope adjustment to obtain the reconstructed high. The frequency subband; the last low frequency subband and high frequency subband signals are recovered by the synthesis filter to recover the full band audio and speech signals.

 The frequency band replication strategy of the embodiment of the present invention is first described below:

 Conventional conventional band replication methods typically select a range of low frequency bands as the fundamental frequency and then replicate to the high frequency portion using some fixed rule. For example, translational replication, multiplier relationship replication, and folding replication, such replication methods limit the accuracy of high frequency reconstruction. Because although the high frequency and low frequency components have a certain correlation, the translation and folding replication cannot ensure the correspondence of this correlation, and the correlation between the low frequency subbands for copying and the copied high frequency subbands may not be very good, even Poor correlation, which will introduce noise or change the sound quality; the frequency doubling method utilizes harmonic periodicity, but not every sub-band is a harmonic component, and the correlation is not good, which will introduce noise or change the sound quality. In addition to the diversity of sound sources in speech and audio signals, it is more likely to cause misuse of the copying band by some fixed copying method.

The frequency band replication strategy proposed by the embodiment of the present invention fully considers the correlation between the low frequency subband and the high frequency subband, and is also suitable for short-term characteristics and time-varying characteristics of audio and voice signals, and has flexible frequency band selection. The function ensures that the frequency bands used for copying and copying have an optimal correlation. The following embodiments of the present invention propose the following three frequency band replication strategies and their extension strategies:

 (1) High-frequency segmentation low-frequency matching replication strategy (strategy 1):

 This strategy divides the high frequency band components into multiple copy bands, and the division of the copy bands can be performed in different ways. For example, the copy band can be divided according to the Bark scale band, and the sub-bands are grouped into a copy band and the like according to different resolutions in the Bark proportional band. The resolution indicates the number of subbands included in the copy band. The smaller the number, the higher the resolution, and the lower the number, the lower the resolution. When the copy band is divided within the Bark band, the resolution decreases as the frequency increases. The copy band division can also equally divide the high frequency component into a plurality of copy bands at the same resolution, and then select the most relevant low band for each high band for copying.

 As shown in FIG. 11, it is a schematic diagram of a high frequency segmentation low frequency matching replication strategy according to an embodiment of the present invention. B0 is the end subband of the low frequency subband signal, B1 is the end subband of the high frequency processed signal, and bl, b2, b3 are the boundaries of the complex band division. After dividing the copy frequency band, the strategy selects the most relevant low frequency sub-band for each copy band for copying, and the low-frequency sub-band can be reused as long as it has the greatest correlation with the copied copy band.

 This replication strategy makes full use of the correlation between high frequency and low frequency, and is suitable for the case where the spectral envelope is relatively stable. Both high and low frequencies have good spectral envelope energy, because the high frequency components at this time have important audio components. Especially in the high-frequency band near the low-frequency part, if the high-frequency reproduction produces distortion, noise will be introduced and the sound quality will be affected. The segment-selective copying strategy will select the most relevant low-frequency band for each high-band to copy, ensuring the inter-band. Correspondence can be used to avoid distortion caused by misuse of the frequency band.

 Compared with the prior art, the copying strategy has the difference that: the prior art continuously and repeatedly copies the high frequency signal by using the whole low frequency signal, and copies the low frequency subband with a large difference when the correlation between the high and low frequency signals is poor. The high-frequency sub-band introduces a large distortion, and the copying strategy uses the original high-frequency sub-band signal as a reference to select the low-frequency sub-band with the greatest correlation to copy, which can effectively avoid the band misuse problem.

 (2) Low frequency dominant frequency band high frequency matching replication strategy (strategy 2):

The strategy first selects the frequency band in which the energy is concentrated in the low frequency signal as the dominant frequency band, and then selects a high frequency signal segment that is more correlated with the low frequency signal of the segment, and performs the frequency band in which the energy is concentrated in the selected low frequency signal segment in the high frequency signal segment. The copy is used as a high frequency sub-band; for the high frequency signal segments, the remaining small frequency bands that have not been copied are respectively selected for the closest low frequency band to be copied. The method first has good harmonics The frequency band is processed, and then the nearest low-frequency sub-band is selected for the scattered band between them. In the frequency band selection, the highest-frequency harmonics with the best correlation are selected first, and then the most relevant low-frequency sub-bands are selected by the non-harmonic high-frequency sub-bands.

 As shown in FIG. 12, it is a schematic diagram of a high frequency matching replication strategy for a low frequency dominant frequency band according to an embodiment of the present invention. Firstly, the low-frequency signal of the energy concentration is selected, and then the high-frequency part with good correlation is selected according to the selected low-frequency signal, and the part is copied, for example, the high-band signals i and j are the copy frequency bands selected by the low-frequency signal, and then the i and The scattered high frequency band other than j selects the appropriate low frequency band for copying.

 The high frequency matching replication strategy of the low frequency dominant frequency band utilizes the frequency domain harmonic characteristics of the signal to select high frequency harmonics of different orders for the fundamental frequency signals in the low frequency, and is suitable for speech and audio signals with good harmonic characteristics. According to the general nature of harmonic transformation, the interval of harmonics appearing in the high frequency part will gradually decrease, and the frequency range of harmonic coverage will gradually increase. Therefore, the conventional harmonic copying by the frequency doubling replication method will be generated. High-frequency distortion, and the embodiment of the present invention can accurately determine the position of the high-frequency harmonic using the maximum correlation determination method, and also select the low-frequency sub-band for copying using the maximum correlation method for the scattered sub-band between the high-frequency harmonics. Ensure the integrity of the harmonics and compensate for the phenomenon of high frequency harmonic widening.

 Compared with the prior art, the copying strategy has the difference that the prior art adopts the frequency doubling copying method, and uses the low frequency subband signal to copy to the high frequency in multiples, and the formed high frequency harmonics contain harmonics of different low frequency signals. The wave destroys the continuity of the harmonics, and the copying strategy continuously replaces the fundamental frequency signal with the high frequency harmonics to ensure harmonic continuity without causing high frequency distortion.

 (3) High frequency matching replication strategy in low frequency band (strategy 3):

 This strategy treats the entire low-band as a reference-band signal and then selectively replicates the harmonics in the high-band according to the optimal matching principle.

 As shown in FIG. 13, it is a schematic diagram of a low frequency band high frequency matching replication strategy according to an embodiment of the present invention. The high frequency bands i and j are selected harmonic components more related to the low frequency signal, and the entire low frequency band is copied at this position, and the scattered subband between them still uses the maximum correlation copy band selection method to select the low frequency subband copy. .

Such a harmonic selection replication strategy is suitable for an audio signal with a relatively smooth spectral envelope and a high-frequency energy drop, and a low-frequency energy. Such an audio signal generally has an index of high-frequency harmonic energy that increases with an order of magnitude. Decrease, because the high-frequency energy is small, it can be regarded as a mixture of harmonics and noise, so the entire low-frequency signal can be selectively copied to the high-frequency, but the high-frequency sub-band replication accuracy near the low frequency band at a low code rate Very important, you should make detailed selection of relevant bands.

 Compared with the prior art, the copying strategy has the following differences: the prior art uses a low-frequency signal to continuously copy high-frequency signals, and the copying strategy considers the low-frequency as a whole and selects and low-frequency signals from the high-frequency components. The most relevant high-frequency harmonics, at which the entire low-frequency band is reproduced, as i and j in Figure 13, allowing a transition band between the harmonics, and then selecting the appropriate low-frequency for the transition band using the optimal band selection method. Brings a copy, which prevents harmonic shifts.

 (4) Extended strategies for band replication strategies (Policy 1, Strategy 2, and Strategy 3):

 This method of expansion strategy is to use the lower frequency high frequency subband band previously obtained by high frequency copying for higher frequency band reproduction.

 Because the low frequency signal may not cover a complete harmonic under low bit rate, the band selection range is extended, and the low frequency signal and a small number of high frequency subbands adjacent to the low frequency signal are regarded as an integral part, and then strategy 1 and strategy 2 are used. Or the method in Strategy 3 determines the replication band. The most relevant low frequency sub-band replica is selected for the small number of high frequency sub-bands added. When performing the band selection detection, first, the range of the extended band (ie, the high frequency sub-band as the copy source at the time of recovery) is determined according to the encoding bit rate and the harmonic integrity relationship, and each extended band is selected for copying using the maximum correlation criterion. The low frequency subband, since the extended band requires the highest reconstruction accuracy, the band selection uses the highest band resolution (ie, a single subband is the copy band); then the extended band and the low frequency subband are combined as a copy source, and then Strategy 1, Strategy 2 or Strategy 3 is used to select the replication band for the high frequency subband.

 As shown in FIG. 14, it is a schematic diagram of an extended replication policy according to an embodiment of the present invention. After the reproduction of the frequency band 1 is re-established, the lower frequency bands of the low frequency band and the copy frequency band 1 are combined into a continuous frequency band, and used for signal reconstruction of the complex frequency band 2 and above.

 This copying strategy is suitable for use at low bitrates because the low frequency band handled by the core codec is shorter at low bitrates, may not cover all fundamental frequency overtones, and is more harmonic in the mid-range than low-frequency overtones. Close to the high frequency overtone characteristics, the reconstructed signal can be used for higher frequency band reproduction after ensuring that the lower frequency high frequency signal is reconstructed with higher resolution. Through this method of copying, harmonics can be more completely depicted, which is conducive to expanding the range of high-frequency reconstruction.

 The high frequency reconstruction method of the embodiment of the present invention is described in further detail below.

There are two types of high frequency reconstruction methods in the embodiments of the present invention, namely, an adaptive band copy mode and a fixed band copy mode: (1) Adaptive band copying mode: As shown in FIG. 15(a), it is a structural block diagram of an encoding end adaptive band copying mode according to an embodiment of the present invention. In this way, the characteristics of the audio signal are detected by the energy language analysis and estimation method, and the guidance information is output according to the detection result for guiding the selection of the replication strategy, thereby guiding the optimal frequency band selection. Since the characteristics of the speech and audio signals are usually the same in a certain period of time, that is, the quasi-stationary characteristic, there is no need to re-band the band selection, so the introduction of the time-varying characteristic detection only needs to re-band the band when the time-varying characteristic variable is greater than the tolerance. select.

 (2) Fixed-band copying mode: As shown in Fig. 15 (b), it is a structural block diagram of the fixed-band copying mode of the encoding end of the embodiment of the present invention. In this way, as long as the fixed copy mode is selected in advance according to actual needs, that is, one of the above-mentioned proposed band copy policies (such as policy 1, policy 2, policy 3 or their extended policies) is determined in the entire audio. It remains unchanged during processing and is implemented in conjunction with appropriate optimal frequency band selection. Because the fixed-band copy mode does not need to guide the selection of the band copy strategy according to the result of the short-term characteristic analysis module, the copy policy is specified by setting parameters, so the short-term characteristic analysis module is not needed.

 It should also be noted that in the adaptive band copy mode and the fixed band copy mode, time-varying characteristic detection is not necessary.

 The high frequency reconstruction method of the adaptive band copy method will be described below.

 Please refer to FIG. 16, which is a flowchart of a high frequency reconstruction method in an adaptive band replication mode according to an embodiment of the present invention, including the steps:

 Step 1601: Perform short-term characteristic analysis on the sub-band signal obtained by the analysis filter module; Step 1602: perform a frequency band replication strategy selection according to the result of the short-term characteristic analysis;

 Step 1603: Perform optimal frequency band selection according to the selected frequency band replication policy.

 Step 1604: Perform frequency band replication according to the optimal frequency band.

 The steps are described in detail below.

 Step 1601: Perform short-term characteristic analysis on the sub-band signal obtained by the analysis filter module. For the original audio signal, convert the sub-band signal distributed in different frequency bands by the analysis filter module, and perform short-term characteristic analysis on the sub-band signal. .

Short-term characterization is a preparation for selecting the appropriate band replication strategy. The audio or speech signal is first subjected to time-frequency transform, and then the energy distribution of the harmonic condition, the low frequency part and the high frequency part is analyzed, and the analyzed parameter result determines the band replication strategy. There are many implementation algorithms for short-term characteristic analysis, and embodiments of the present invention use one of the algorithms, but are not limited thereto.

Let the low frequency sub-band sample point be. ^ , /). Where "represents the low frequency sub-band number, ί ≤ η, which is the first sub-band number of the high-frequency processing; / represents the sample point in the sub-band, 0 ≤ / < 32. Let the high-frequency sub-band sample be ^^ ( W), where: represents the high frequency subband, k ₀ ≤ k ≤ k _e , which is the end subband of the high frequency processing.

 1. Calculate the energy of each subband in the low frequency part, as shown in the following equation:

E(") = f ² . _w (", /)

 /=0

2. Calculate the energy mean of the entire low frequency part, as shown in the following equation:

 3. Calculate the energy of each sub-band in the high-frequency part, as shown in the following equation:

Ε{^ = ^Χ _Η ² _ίφ { Ι)

 1=0

4. Calculate the energy average of the entire high frequency part, as shown in the following equation:

 - After the above calculation, the short-term characteristic analysis ends and the obtained analysis parameters will be applied to the band copy strategy selection section.

 Step 1602: Perform a frequency band replication policy selection according to a result of short-term characteristic analysis;

 Four frequency band replication strategies have been mentioned above, and based on the results of the short-term characteristic analysis, a strategy is selected for replication. After the band replication policy is determined, the replication policy flag and the replication policy information are output to guide the subsequent optimal band selection.

 Please refer to FIG. 17, which is a flowchart of selecting a frequency band replication policy according to an embodiment of the present invention.

Comparing the energy values of the low frequency sub-bands obtained by the short-term characteristic analysis with the energy average values of the low-frequency parts, and comparing the average values of the high-frequency partial energy with the low-frequency partial energy average. According to the comparison, when the energy value of the partial low frequency sub-band is much lower than the mean value, and the energy of some continuous low-frequency sub-bands is close to the mean or above the mean value, then strategy 2 is selected; if the energy of each sub-band of the low frequency is close, the low-frequency part The energy curve is continuous and gentle, and the difference between the average value of the high-frequency part energy and the low-frequency part energy is larger, then choose strategy 3; otherwise, select Choose strategy 1.

 The judgment process consisting of Strategy 2, Strategy 3 and Strategy 1 is the main body of policy selection, and the band replication expansion strategy is used as an auxiliary strategy, mainly in the process of recovering high frequency, for the case where the low frequency band is relatively narrow, and the expansion is used for copying. The width of the low frequency band improves the integrity of the fundamental frequency. When the coding rate is low and the number of low frequency subbands is limited, the band replica expansion strategy can make the selection of the low frequency subbands of the high frequency portion not too restrictive. It bundles several mid-high frequency sub-bands and low-frequency parts that would otherwise require band replication to form a new low-frequency part for most of the high-frequency sub-bands to choose and extract their corresponding copy parameters. At the same time, for some of the high-frequency sub-bands in the new low-frequency part, the best-matched low-frequency sub-bands are selected from the original low-frequency parts, and the copy parameters are extracted in turn.

 The Band Replication Extension Policy is an extension of Policy 2, Policy 3, and Policy 1. When the extend-flag flag is output in the band copy policy selection process, the band copy extension policy is used. Therefore, when there is an extend_flag output, the selected policy 2, policy 3 or policy 1 will become an extended policy 2, an extended strategy 3 or an extended strategy 1 accordingly.

 The specific process in Figure 17 is as follows:

 Step 1701: The time-frequency transform is completed, and a QMF (Quadature Mirror Filter) sub-band is input;

 Step 1702, determining whether the input sub-band is in a low coding rate mode, if yes, proceeding to step 1703, if no, proceeding to step 1705;

 Step 1703, expanding the range of the low frequency part participating in the copy, forming a new low frequency part, proceeding to step 1704;

 Step 1704, the output flag extend_flag, for the band copy expansion strategy, proceeds to step 1705; Step 1705, determines whether there is a low frequency subband with low energy, if not, proceeds to step 1708, and if yes, proceeds to step 1706;

 Comparing the energy of each low-frequency sub-band with the mean value of the low-frequency part, if there is a certain low-frequency sub-band energy that satisfies the following formula: Ε^ δ Ε^ , indicating that there is a sub-band energy drop in the low-frequency sub-band In the event that the fundamental frequency energy distribution is not continuous, proceed to step 1706, otherwise proceed to step 1708. The value range is 0< 4 < 1, and the value range is the empirical value obtained by observing the waveform corresponding to the copy strategy. The value can be set according to requirements.

Step 1706, searching for a low frequency subband with higher energy and continuous distribution and determining a selection strategy 2; This step mainly looks for a subband interval in which the energy distribution of the low frequency portion is continuous as the fundamental frequency portion of the strategy 2. The judgment algorithm is as follows:

If (") _>( 3⁄4*^.„, E(n + l)>S ₂ *E _Low , ...... , E(n + ql)>S ₂ *E _Low ,

E{n + q)<S ₂ *E _Low , (where ? ^≥1 , ^l≤n<k 。 ^,, the value range is 0 <<< 1, the value range is the waveform corresponding to the copy strategy If the empirical value obtained by observation can be set according to requirements, then it is decided to adopt strategy 2, and record the sub-band number n and the sub-band interval number q;

 Step 1707: Output the flag flag corresponding to the policy 2, and output the subband sequence number n and the subband interval number q. It should be noted that if the extend_flag is output at the same time, the current policy is the extended policy 2.

 Step 1708, determining whether the high frequency energy average is too low compared with the low frequency energy average, if not, proceeding to step 1709, and if yes, proceeding to step 1710;

 If the low-frequency sub-band energy and the low-frequency energy mean ^^ are compared by step 1705, the following formula is satisfied: £(")> *Ε^, the emphasis of the analysis is shifted to the energy relationship between the low-frequency and high-frequency parts.

Compare the ^£ and the value, where the value range is 0<<1, which is the empirical value observed for the relevant waveform. The value can be set as required.

When able to satisfy ^£ ^ ^{≤ /1} *. Then decide to adopt strategy 3, proceed to step 1710, otherwise, when ^{£ >/1} * ^£ ^, decide to adopt strategy 1, and proceed to step 1709;

 Step 1709: Output the corresponding flag of the policy 1 Flag; It should be noted that if the extend_flag is output at the same time, the current policy is the extended policy 1.

 Step 1710: Output the corresponding flag of the policy 3 Flag; It should be noted that if the extend_flag is output at the same time, the current policy is the extended policy 3.

 Step 1603: Perform optimal frequency band selection according to the selected frequency band replication policy.

 The optimal frequency band selection module uses the maximum correlation as a standard to flexibly search for the optimal matching frequency band for copying of a reference frequency band, ensuring the correlation of frequency band replication, so that the copied high frequency signal does not need excessive adjustment, and approaches the original signal.

According to the determination of the determined replication policy and replication policy information (including the corresponding initialization band table), the optimal high-low frequency signal correspondence is selected. The band replication strategy guides the optimal band selection, determines whether the band selection is to select a low frequency signal with a high frequency signal, or select a high frequency harmonic with a low frequency signal, for example, In the first round, the optimal frequency band is selected to select the best low frequency signal for copying for each high frequency copy band signal, and in strategy 2, the optimal frequency band selection first selects the high frequency that can be copied for the base frequency signal. harmonic. The initialization band table guides the estimated bandwidth of the optimal band selection, as well as the selected band range.

 The optimal frequency band selection is to compare the high and low frequency signal correlation and the similarity degree of the high and low frequency signal envelope characteristics, and then determine the optimal high and low frequency signal matching relationship based on the obtained two parameters. In order to avoid the calculation difference caused by the signal energy amplitude when comparing the correlation and the envelope characteristics, the signal is normalized according to the initial band table range before the estimation, so that the matching signal is selected to analyze the similarity of the signal characteristics, and the energy difference can be Make adjustments when the signal is reconstructed.

 Regardless of whether the reference band selected by the optimal band is a high frequency signal or a low frequency signal, the algorithms used for optimal band selection are the same. For convenience of explanation, the following strategy 1 is taken as an example, and a single subrepresentative representing the maximum frequency resolution is taken. The band is the copy band, indicating the general algorithm for optimal band selection.

 Referring to FIG. 18, it is a flowchart of an optimal frequency band selection according to an embodiment of the present invention, including the following steps: Step 1801: divide a copy frequency band and an alternate frequency band according to an initialized frequency band table in the frequency band copy policy information, and input an input sub-band signal according to an initialized frequency band table. Perform high and low frequency division;

Let the number of copy bands be represented by nb, and set the low-frequency sub-band samples with ^ (", where 1 ≤ "< refers to the low-frequency sub-band number, which is the first sub-band number of high-frequency processing, 0 ≤ / < ³² Indicates the sample points in the sub-band, and sets the high-frequency sub-band samples as ^ ^, ^Ζ ), where ^≤ indicates the high-frequency sub-band, which is the end sub-band of the high-frequency processing.

Step 1802: Perform normalization processing on the high and low frequency signals in the length of the copy frequency band according to the division of the copy frequency band in the initialization band table. Here, it is assumed that the copy frequency band is equal to the transform subband, and the normalization process is as follows: 丽(0) ^X m ^{k !,} step 1803, calculated for each frequency band or a high frequency band with the correlation function may be low frequency or low frequency band; offset considering the case of sample, in order to obtain the best approximation of the waveform of the high-band low frequency band of the low frequency band After the sample is offset, the correlation function is calculated. The formula is as follows: r _k ^m [n] = _j X _High (k,l)X _Low (n - m) , m = 0,1,2 „ , _A ^ „ , ^ „ _r ,

^ , where is the number of offset samples, ^r k ["] Represents the correlation function value of the high and low frequency bands after the sample offset. If the time-frequency transform uses a complex transform form, the sub-band samples are complex-valued, and the real part can be analyzed.

In the offset correlation function with a certain high frequency band, the correlation value is selected for each low frequency band.

The above calculation is performed on all the high frequency bands, and the maximum correlation matrix ax WW composed of ["] is obtained, and ^R max WW records the maximum correlation values of all the high and low frequency bands. Step 1804, estimating the variation characteristics of the high and low frequency band envelopes, and calculating the high frequency Band and low band envelope variation characteristics;

 The envelope characteristic estimation method is to treat the sample point in the length of the replication frequency band as a sample, calculate its autocorrelation function within the second order, and then obtain the high and low frequency envelope by comparing the mean square error of the autocorrelation function of the high and low frequency bands. Difference in characteristics.

 First, calculate the second-order autocorrelation function of the high and low frequency components according to the length of the replication band:

, where Ή and the autocorrelation function representing the high and low bands, respectively, represent the autocorrelation interval. Then, calculate the envelope difference between the high and low frequency subbands: e (k, n) = _Pl ∑ (r [l] -

[/]) + _A ∑ ( [/] - r _n ² [l]

, where A and A are two coefficients, A + A = 1 , ^e ^, " ) represents the envelope difference between the high and low bands. Step 1805, comprehensively comparing ^ and selecting an optimal low frequency for each copy band With copying; obviously, the larger the value of ΓΜ, the better the correlation between high and low frequency bands, and the smaller the value of ^, "), the more similar the envelope representing the high and low frequency bands. To find the most suitable matching relationship between high and low frequency bands, uniformity must be formed. The parameters are compared, so there are the following transformations:

Where "sum is the weight coefficient, which is the formed band selection coefficient.

Is the mean of the array ^] and W, α = β. Step 1806, generating a band selection table indicating an optimal frequency band required for copying.

For each high frequency sub-band, the value that maximizes the %(") value is selected. When the band copy expansion strategy is selected, the reference band is not only the low frequency sub-band, but may be all the bands before the detection band. A band selection table _W K - +1] is generated indicating the optimum frequency band required for copying.

 Step 1603: After the optimal frequency band selection is performed according to the selected frequency band replication policy, the optimal frequency band may be used all the time, and the time-varying characteristic detection may be further performed, and the frequency band is reselected according to the detection result.

 The following describes the time-varying feature detection:

Audio and speech signals generally have the same characteristics (ie quasi-stationary characteristics) over a period of time, so the same high-frequency duplication strategy may be used in successive frames, with the same copy strategy, based on the audio signal Time periodicity, the same copy band selection table can be used in consecutive frames, and it is not necessary to perform optimal band selection every frame. Once the band selection table is determined, it can be used continuously for multiple frames, which can save calculation amount and transmission bit rate. It can also guarantee the continuity between frames. In order to judge whether or not the band selection table of the previous frame can be used, a time-varying characteristic detecting module is introduced. The function of time-varying characteristic detection is to determine whether the current frame can use the band selection table of the previous frame. If it is detected that the difference in audio characteristics between adjacent frames is greater than the threshold, the band selection table is refreshed and the frequency band is reselected; otherwise, the frequency band The selection table remains unchanged. The time-varying characteristic detecting method is to estimate the audio characteristic change of the low-frequency signal of the current frame and the previous frame, and the envelope difference comparison method may be specifically used. If the envelope difference is small, according to the correlation between high and low frequencies, the difference of the high frequency signal will be 4艮, and the band selection table generated by the previous frame can be used; if the current low frequency signal and the previous frame low frequency signal The envelope difference is within the tolerance range, but there is a frequency offset. If the frequency offset is greater than 5% of the critical band, then the optimal frequency band needs to be re-selected, and the band selection table is refreshed because the two sets of overtones are in accordance with the pitch shift theory. The frequency difference within the defined critical band is between 5% and 50%, and the two sets of overtones are dysfunctional, producing auditory perceptible differences.

Referring to FIG. 19, it is a flowchart of an algorithm for time-varying characteristic detection according to an embodiment of the present invention, including the following steps: Step 1901: Calculate a low-frequency sub-band energy mean square error ATM of a current frame and a previous frame;

, which represents the low frequency subband energy of the current frame, ^£ » represents the low frequency subband energy of the previous frame, step 1902, determines whether the mean square error of the low frequency subband energy is less than the threshold of the decision is Thr, and if yes, proceeds to step 1903, and if not, enters Step 1906; Step 1903, estimating an offset frequency of the low frequency band in which the energy is concentrated; selecting a frequency band or a frequency band with the highest energy, and setting a center frequency thereof,

, where Λ represents the lower and upper bounds of the highest energy band, respectively, and calculates the frequency offset ^Δ / = _ ;

Step 1904, determining whether the offset frequency ^Δ / is less than 5% of the current critical band bandwidth, and if so, proceeds to step 1905, and if not, proceeds to step 1906;

 Step 1905: A band selection table generated by using a previous frame is used;

 Step 1906: Perform optimal band selection again.

 It should be noted that, in the case of the fixed-band copying mode, the short-term characteristic analysis of the sub-band signals obtained by the analysis filter module is not required, and the short-term characteristics are not required. The result of the analysis is selected for the band replication strategy, and the flow of the optimal band selection process and the time-varying characteristic detection is the same.

Step 1604: Perform frequency band replication according to the optimal frequency band. After the decoding end obtains the optimal frequency band, the frequency band copying can be performed according to the optimal frequency band. Referring to FIG. 20, it is a schematic diagram of the high frequency reconstruction performed by the decoding end according to the parameter information of the encoding end according to the embodiment of the present invention.

 Compared with the prior art decoding end, the decoding end of the embodiment of the present invention does not change the function and cooperation relationship of most modules, but only the "high frequency sub-band copying strategy" of the high frequency generator module is modified. In the SBR code stream input to the high frequency generator, three parameters are added according to the high frequency reconstruction guide parameter information, namely "new algorithm use flag", "band selection table replacement flag" and "band selection table".

 The "band selection table" is a relatively important parameter, which records the copy correspondence between the high frequency sub-band and the low frequency sub-band when the high frequency sub-band is recovered for each frame signal.

 The "new algorithm uses flags" determines whether high-frequency reconstruction is guided by a new algorithm at the encoding end or high-frequency reconstruction using a standard SBR method. The new algorithm referred to herein refers to the algorithm used in the process of finally determining the high frequency reconstruction parameters at the encoding end by the embodiment of the present invention described above. If the flag is "1", the high frequency is reconstructed according to the new algorithm; if the flag is "0", the high frequency is reconstructed according to the standard SBR method. With this setup, the interface can be reserved for compatibility between the new algorithm and the standard SBR method in subsequent studies.

 The "band selection table replacement flag" determines how the current signal obtains the "band selection table" when the high frequency sub-band is restored. If the flag bit is "0", the current signal directly extends the high-low frequency sub-band correspondence of the previous frame signal to guide the high-frequency copy; if the flag bit is "1", the refresh is read according to the read in the SBR code stream. The "band selection table" parameter completes the high frequency copy. The main role of the "band selection table replacement flag" is to reduce the amount of data transmitted to the decoder. It should be noted that when the "band selection table replacement flag" is 0, the "band selection table" parameter will not be included in the transmitted SBR code stream.

 The code stream information received by the decoder is described in detail below.

 See Table 1, which describes the header file HeadFile data structure, which is used to initialize the settings when the decoder starts working.

 Variable name Description Data type sbr_amp_res Quantization ladder, as a guide when not encoding Char (character type)

 Sbr_start—frequency SBR field start QMF subband selection pointer, select corresponding element in the band selection Int (integer) table StartTable

Sbr_ stop—frequency SBR field terminates QMF subband selection pointer, selects Int in frequency band Select the corresponding element in the table StopTable

 Xover—band main frequency representative pointer Int

 Header— extra— 1 Indicates if there is additional information, value 0 or 1 Int

 Header— extra— 2 Indicates if there is additional information, value 0 or 1 Int

 FreqScale SBR band contains the maximum number of QMF subbands used to calculate Int

 Main band selection table

 alterScale The highest frequency SBR band contains the number of QMF subbands that can be increased, using Int

 Computing main band selection table

 Noise—the number of noise bands Int

 Sbr_limiter— band The number of bands that are limited when calculating the gain, used to calculate the limit frequency Int

 Belt selection table

 Sbr_limiter—gains P The maximum gain in the band Int

 Sbr — interpol — freq Indicates whether band rewriting is applied, default value 1 Int

 Sbr— smoothing— length Whether to apply filtering Int

 Table 1 HeaderFile data structure table

 ENV—DATA is the data structure that describes the SBR information for each frame. The parameters "New Algorithm Usage Flag", "Band Selection Table Replacement Flag" and "Band Selection Table" are added to the ENV DATA structure describing the SBR information.

 Define a structure variable specifically for storing "new algorithm usage flag", "band selection table replacement flag" and "band selection table, three parameter information, as follows:

 Struct INDVEC

 { char flag — 1;

 Char flag— 2;

 Char FreTable[28];

 } index Vector;

 The values of the "new algorithm use flag" and "band selection table replacement flag" are both "0" or "1", so two character variables "flag-1" and "flag-2" are set to describe respectively. "New algorithm uses flag", "band selection table replacement flag".

The "band selection table" stores the corresponding serial numbers of the high frequency sub-band and the low frequency sub-band that need to be restored, and is stored. In the array FreTable[28]. Among them, different encoding modes have different numbers of high frequency sub-bands to be recovered. For the highest bit rate encoding mode, the number of high frequency sub-bands that need to be recovered is 28. As the encoding bit rate decreases, the number of high frequency sub-bands that need to be recovered also decreases accordingly.

 The position of the structure variable indexVector in the ENV-DATA structure is shown in Table 2.

 Table 2 ENV-DATA Structure Definition Table

 The algorithm flow of the high frequency generator module is described below. Referring to FIG. 21, it is a flow chart of the decoder high frequency generator module algorithm according to the embodiment of the present invention, including the steps:

Step 2101: Receive "new algorithm use flag", "band selection table replacement flag", and "band selection" Table";

 Step 2102, judging "new algorithm use flag", if it is 0, proceeds to step 2103, if it is 1, proceeds to step 2104;

 Step 2103, decoding according to a standard SBR method;

 Step 2104, judging "band selection table replacement flag", if it is 0, proceeds to step 2105, if it is 1, proceeds to step 2106;

 Step 2105: If the flag bit is "0", the current signal directly extends the high-low frequency sub-band correspondence of the signal of the previous frame, and guides the high-frequency copy;

 It should be noted that after determining the high-low frequency sub-band correspondence of each frame signal, the band selection table of the current frame is backed up in the buffer. The next frame signal is called on the band selection table in the buffer, if needed.

 Step 2106: If the flag bit is "1", the high frequency copy is guided according to the "band selection table" parameter read in the SBR code stream;

 Step 2107: Perform preliminary high frequency copying.

 After the initial replication of the high-frequency sub-band, it will enter the module of envelope adjustment and harmonic component addition to complete the high-frequency replication.

 The high frequency reconstruction method of the embodiment of the invention more accurately realizes the reconstruction of the high frequency signal. Referring to FIG. 22, it is an energy waveform diagram of the original audio signal after the method of the embodiment of the present invention is restored; FIG. 23 is a three-dimensional diagram of the energy of the original audio signal after the method of the embodiment of the present invention is restored. By comparing these two figures with the prior art figures, it can be found that the high frequency reconstruction effect of the embodiment of the present invention is better than the prior art. Therefore, the method of the embodiment of the present invention can reconstruct the high frequency signal more accurately by using the information of a few low frequency subbands, and is also beneficial for compressing the audio information, can greatly improve the compression efficiency of the audio and speech encoder, and improve the audio quality; Reduce distortion and noise caused by bit rate audio and speech signal compression coding. And for different audio characteristics, a variety of corresponding frequency band replication strategies are proposed, which can provide adaptive high-frequency reconstruction methods for various audio and voice signals, and improve the flexibility of audio and voice signal processing.

 The above description is accompanied by a detailed description of the high frequency reconstruction method of the embodiment of the present invention. Correspondingly, the embodiment of the present invention provides an encoding module and a decoding module.

 Referring to FIG. 24, it is a schematic structural diagram of an encoding module according to an embodiment of the present invention.

The encoding module includes: an analysis filter module 241 and a band selection module 242. The analysis filter module 241 is configured to filter the audio or voice signal to obtain a low frequency sub-band and a high frequency sub-band.

 a frequency band selection module 242, configured to determine a frequency band replication policy, obtain a correlation between the low frequency sub-band and a high frequency sub-band according to the determined frequency band replication policy, and select a low frequency sub-frequency for the high frequency sub-band with reference to the determined correlation As the copy frequency band, for example, a low frequency sub-band having a high correlation is selected as a high frequency sub-band as an optimum copy band, and high-frequency reconstruction parameter information including a correspondence relationship of the selected band is output.

 The encoding module further includes: a short-term characteristic analysis module 243 for performing short-term characteristic analysis on the audio or voice signal.

 The band selection module 242 includes: a replication policy selection module 2421, an optimal band selection module 2422.

 The replication policy selection module 2421 is configured to select a different frequency band replication policy according to the analysis result of the short-term characteristic analysis module 243;

 An optimal frequency band selection module 2422, configured to acquire a correlation between the low frequency sub-band and a high frequency sub-band according to the determined frequency band replication policy, and select a low frequency sub-band as a high frequency sub-band as a reference with reference to the determined correlation The frequency band, for example, a high frequency sub-band selects a low frequency sub-band having a large correlation as an optimal copy frequency band, and outputs high frequency reconstruction parameter information including a correspondence relationship of the selected frequency bands.

 The encoding module further includes: a time-varying characteristic detecting module 244, configured to perform time-varying characteristic detection on the filtered audio or voice signal; correspondingly, the optimal frequency band selecting module 2422 is further detected according to the time-varying characteristic detecting module 244. As a result, the low frequency sub-band is selected, for example, the optimal copy band is selected.

The short-term characteristic analysis module 243 performs short-term characteristic analysis on the audio or voice signal, specifically: calculating a low-frequency partial energy average of the audio or voice signal, a high-frequency partial energy average, a low-frequency partial sub-band energy value, and a high frequency. Part of each sub-band energy value; the replication strategy selection module 2421 correspondingly selects different frequency band replication strategies according to the analysis result of the short-term characteristic analysis module 243, specifically: comparing the energy values of the sub-bands in the low-frequency part with the average value of the low-frequency parts; The sub-band energy value of the partial sub-band of the low-frequency part is less than or equal to the first weighting value of the low-frequency part energy average, and the selected strategy is: selecting the low-frequency sub-band of the energy concentration, and selecting the selected low-frequency sub-band of the energy concentration The high frequency band of the large portion replicates the low frequency sub-band; if the sub-band energy value of the low frequency partial sub-band is greater than the first weighting value of the low-frequency partial energy mean, the second weighting of the high-frequency partial energy mean and the low-frequency partial energy mean is further performed. The value is compared; if the high-frequency partial energy mean is less than or equal to the second weight of the low-frequency partial energy mean Election The selected strategy is: selecting the entire low frequency sub-band, and copying the low frequency sub-band in the selected high frequency band having a large correlation with the entire low frequency sub-band; if the high-frequency partial energy average is greater than the second weighting value of the low-frequency partial energy average, selecting The strategy is as follows: The high frequency is divided into multiple copy frequency bands, and the low frequency sub-bands with large correlation are selected for each copy frequency band for copying.

 The strategy selected by the replication policy selection module 2421 further includes: further copying, by copying, a high frequency sub-band adjacent to the low frequency sub-band together with the selected low frequency sub-band as a copy source, and the high frequency sub-band of the adjacent low frequency sub-band Select the low-frequency sub-band copy with high correlation, which is equivalent to the expansion strategy of each of the above strategies.

 Referring to FIG. 25, it is a schematic structural diagram of an encoding module 2 according to an embodiment of the present invention.

 The encoding module includes: an analysis filter module 241, and a band selection module 242.

 The analysis filter module 241 is configured to receive an audio or voice signal and perform filtering processing to obtain a low frequency subband and a high frequency subband.

 a frequency band selection module 242, configured to determine a frequency band replication policy, obtain a correlation between the low frequency sub-band and a high frequency sub-band according to the determined frequency band replication policy, and select a low frequency sub-frequency for the high frequency sub-band with reference to the determined correlation As the copy frequency band, for example, a low frequency sub-band having a high correlation is selected as a high frequency sub-band as an optimum frequency band, and high-frequency reconstruction parameter information including a correspondence relationship of the selected frequency bands is output.

 The band selection module 242 includes: a copy policy setting module 2423, an optimal band selection module 2422.

 The copy policy setting module 2423 is configured to determine a unique band copying policy according to the preset parameters. The band replication policy is one of the policies described in connection with FIG. 24 or its corresponding extension policy.

 An optimal frequency band selection module 2422, configured to acquire a correlation between the low frequency sub-band and a high frequency sub-band according to the determined frequency band replication policy, and select a low frequency sub-band as a high frequency sub-band as a reference with reference to the determined correlation The frequency band, for example, a high frequency sub-band selects a low frequency sub-band having a large correlation as an optimum frequency band, and outputs high frequency reconstruction parameter information including a correspondence relationship of the selected frequency bands.

 The encoding module further includes: a time-varying characteristic detecting module 244, configured to perform time-varying characteristic detection on the filtered audio or voice signal; correspondingly, the optimal frequency band selecting module further selects according to a result detected by the time-varying characteristic detecting module The low frequency subband, for example, selects the optimal copy band.

 Referring to FIG. 26, it is a schematic structural diagram of a decoding module according to an embodiment of the present invention.

The decoding module includes a high frequency generator module 261, and the high frequency generator module 261 includes: The receiving unit 2611 and the reconstructing unit 2612.

 The receiving unit 2611 is configured to receive high-frequency reconstruction parameter information including a correspondence relationship of the selected frequency bands, where the correspondence relationship of the selected frequency bands is specifically a relationship between the high-frequency sub-band determined by the reference correlation and the corresponding low-frequency sub-band, for example, correlation The large low frequency sub-band corresponds to the high frequency sub-band.

 The reconstruction unit 2612 is configured to copy the low frequency sub-band as a high frequency sub-band according to the high frequency reconstruction parameter information including the correspondence relationship of the selected frequency bands in the high frequency band.

 The parameter information received by the receiving unit 2611 further includes a value of the new algorithm using the flag and a band selection table replacement flag; the reconstruction unit 2612 determines an algorithm used by the copying process according to the value of the new algorithm using the flag, according to the The band selection table replacement flag determines a band selection table used in the copying process, and the low frequency sub-band in the correspondence is copied as a high frequency sub-band in the high band according to the determined algorithm and the band selection table.

 In summary, the solution of the embodiment of the present invention fully considers the correlation between the low frequency subband and the high frequency subband, and obtains the correlation between the low frequency subband and the high frequency subband according to the determined frequency band replication strategy, and refers to the determination. The correlation is that the high frequency sub-band selects the low frequency sub-band as the copy frequency band, and outputs the high frequency reconstruction parameter information including the correspondence relationship of the selected frequency bands, so that the high frequency reconstruction can be performed based on the high frequency reconstruction parameter information. The energy waveform and the energy three-dimensional image of each sub-band signal obtained after the high-frequency reconstruction can be clearly found that the reconstructed high-frequency portion of the embodiment of the present invention is closer to the original audio signal, and the effect is better than the prior art solution. Therefore, the technical solution of the embodiment of the present invention can perform high frequency reconstruction more accurately.

 Further, the solution of the embodiment of the present invention may include an adaptive band copy mode and a fixed band copy mode, and has a flexible band selection function.

 Further, the technical solution of the embodiment of the present invention may further increase the time-varying characteristic detection of the audio or voice signal, and perform adjustment according to the detection result.

 The high-frequency reconstruction method, the coding module, and the decoding module provided by the embodiments of the present invention are described in detail. For those skilled in the art, according to the embodiments of the present invention, the specific implementation manners and application ranges are There are variations, and the description should not be construed as limiting the invention.

Claims

Rights request

 A high frequency reconstruction method, characterized in that it comprises:

 Filtering the audio or speech signal to obtain a low frequency sub-band and a high frequency sub-band;

 Determine the band replication strategy;

 Acquiring a correlation between the low frequency sub-band and a high frequency sub-band according to the determined frequency band replication policy, selecting a low frequency sub-band as a copy frequency band for the high frequency sub-band with reference to the determined correlation, and outputting a correspondence including the selected frequency band High frequency reconstruction parameter information of the relationship.

 2. The high frequency reconstruction method according to claim 1, wherein:

 The determining the frequency band replication strategy is specifically:

 A unique band replication policy is determined based on pre-set parameters.

 3. The high frequency reconstruction method according to claim 1, wherein:

 The determining the frequency band replication strategy is specifically:

 Performing short-term characteristic analysis on the audio or voice signal;

 The different frequency band replication strategies are selected according to the results of the short-term characteristic analysis.

 4. The high frequency reconstruction method according to claim 2 or 3, characterized in that:

 After filtering the audio or voice signal, the method further includes:

 Time-varying characteristic detection is performed on the filtered audio or speech signal; when the low-frequency sub-band is selected as the replica frequency band for the high-frequency sub-band, the low-frequency sub-band is selected as a result of further combining the time-varying characteristic detection.

 5. The high frequency reconstruction method according to claim 3, wherein:

 Performing short-term characteristic analysis on the audio or voice signal, and selecting a different frequency band replication strategy according to the result of performing short-term characteristic analysis is specifically:

 Determining the average value of the low-frequency part of the audio or speech signal, the mean value of the high-frequency part, the energy value of each sub-band in the low-frequency part, and the energy value of each sub-band in the high-frequency part;

 Comparing the energy values of the subbands in the low frequency portion with the first weighting values of the energy average of the low frequency portions; if the subband energy value of the partial subbands in the low frequency portion is less than or equal to the first weighting value of the energy average of the low frequency portion, the selected strategy is : selecting a low frequency sub-band of the energy concentration, and copying the low frequency sub-band in a selected high frequency band having a high correlation with the low frequency sub-band of the energy concentration;

If the subband energy value of the low frequency partial subband is greater than the first weighting value of the low frequency partial energy average, The first step is to compare the high-frequency partial energy average with the second weighted value of the low-frequency partial energy average; if the high-frequency partial energy average is less than or equal to the second weighted value of the low-frequency partial energy mean, the selected strategy is: selecting the entire low-frequency sub-band, Copying the low frequency sub-band at a selected high frequency band having a large correlation with the entire low frequency sub-band;

 If the high-frequency partial energy average is greater than the second weighted value of the low-frequency partial energy mean, the selected strategy is: dividing the high-frequency into a plurality of replicated frequency bands, and selecting a highly correlated low-frequency sub-band for each replicated frequency band for copying.

 6. The high frequency reconstruction method according to claim 5, wherein:

 The selected strategy further includes:

 When copying, the high frequency sub-band adjacent to the low frequency sub-band is further used as a copy source together with the selected low frequency sub-band, and the high frequency sub-band adjacent to the low frequency sub-band is selected to be copied with the low frequency sub-band having a large correlation.

 7. The high frequency reconstruction method according to claim 4, wherein:

 Performing time-varying characteristic detection on the filtered audio or speech signal, wherein the high-frequency sub-band selects the low-frequency sub-band as the copy frequency band is further combined with the time-varying characteristic detection result, and the low-frequency sub-band is selected as follows:

 Determining the mean square error of the low frequency subband energy of the current frame and the previous frame. If the mean square error of the energy is less than the decision threshold, and the offset frequency of the low frequency subband of the energy concentration is greater than a preset value, then the copy band selection is performed again. Otherwise the previously determined copy band is used.

 8. The high frequency reconstruction method according to claim 1, wherein:

 And obtaining, according to the determined frequency band replication strategy, the correlation between the low frequency sub-band and the high frequency sub-band, and referring to the determined correlation, the high frequency sub-band selecting the low frequency sub-band as the replication frequency band is specifically:

 Determining a correlation function value of the high frequency sub-band and the low frequency sub-band;

 Determining an envelope difference value between the high frequency sub-band and the low frequency sub-band according to the autocorrelation function values of the high frequency sub-band and the low frequency sub-band;

 The low frequency sub-band having a high correlation of the high frequency sub-band is selected as the copy frequency band according to the correlation function value and the envelope difference value.

 9. A high frequency reconstruction method, characterized in that:

Receiving high-frequency reconstruction parameter information including a correspondence relationship of the selected frequency bands, where the correspondence relationship of the selected frequency bands is specifically a relationship between the high-frequency sub-bands determined by the reference correlation and the corresponding low-frequency sub-bands; The low frequency sub-band is copied as a high frequency sub-band according to the high frequency reconstruction parameter information including the correspondence relationship of the selected frequency bands at the high frequency band.

 10. The high frequency reconstruction method according to claim 9, wherein:

 The parameter information further includes a new algorithm using the flag and a band selection table replacement flag;

 The replicating the low frequency subband as the high frequency subband according to the high frequency reconstruction parameter information including the correspondence relationship of the selected frequency bands in the high frequency band is specifically:

 Determining the algorithm used by the replication process based on the value of the flag using the new algorithm,

 A band selection table used by the copying process is determined based on the value of the band selection table replacement flag; the low frequency sub-band in the correspondence is copied as a high frequency sub-band in the high band according to the determined algorithm and the band selection table.

 11. An encoding module, comprising:

 An analysis filter module for filtering an audio or voice signal to obtain a low frequency sub-band and a high frequency sub-band;

 a frequency band selection module, configured to determine a frequency band replication policy, obtain a correlation between the low frequency sub-band and a high frequency sub-band according to the determined frequency band replication policy, and select a low frequency sub-band for the high frequency sub-band with reference to the determined correlation As the copy band, the high frequency reconstruction parameter information including the correspondence of the selected bands is output.

 The coding module according to claim 11, wherein the frequency band selection module comprises:

 a replication policy setting module, configured to determine a unique frequency band replication policy according to a preset parameter; an optimal frequency band selection module, configured to acquire, according to the determined frequency band replication policy, a correlation between the low frequency subband and a high frequency subband, Referring to the determined correlation, a low frequency sub-band is selected as a copy frequency band for the high frequency sub-band, and high frequency reconstruction parameter information including a correspondence relationship of the selected frequency band is output.

 13. The encoding module of claim 11 wherein:

 The encoding module further includes:

 a short-term characteristic analysis module, performing short-term characteristic analysis on the audio or voice signal; the frequency band selection module includes:

a replication policy selection module, configured to select a different frequency band replication strategy according to a result of the short-term characteristic analysis module analysis; An optimal frequency band selection module, configured to acquire a correlation between the low frequency sub-band and a high frequency sub-band according to the determined frequency band replication strategy, and refer to the determined correlation as a high frequency sub-band to select a phase low frequency sub-band as a replica a frequency band, and outputs high frequency reconstruction parameter information including a correspondence relationship of the selected frequency bands.

 The encoding module according to claim 12 or 13, wherein the encoding module further comprises:

 a time-varying characteristic detecting module, configured to perform time-varying characteristic detection on the filtered audio or voice signal;

 The optimal band selection module also selects a low frequency sub-band based on the result of the time varying characteristic detection module detection.

 15. The encoding module of claim 13 wherein:

 The short-term characteristic analysis module performs short-term characteristic analysis on the audio or voice signal, specifically: determining an average energy value of a low-frequency part of the audio or voice signal, an average value of the energy of the high-frequency part, an energy value of each sub-band of the low-frequency part, and a high-frequency part. Energy value of each sub-band;

 The copying strategy selection module correspondingly selects different frequency band replication strategies according to the results of the short-term characteristic analysis module analysis: comparing the energy values of the sub-bands of the low-frequency part with the first weighting values of the energy average of the low-frequency parts;

 If the subband energy value of the partial subband of the low frequency portion is less than or equal to the first weighting value of the low frequency partial energy mean, the selected strategy is: selecting the low frequency subband of the energy concentration, and selecting the low frequency subgroup with the energy concentration Copying the low frequency sub-band with a high frequency band with high correlation;

 If the subband energy value of the low frequency partial subband is greater than the first weighting value of the low frequency partial energy mean, the higher frequency partial energy mean is further compared with the second weighted value of the low frequency partial energy mean;

 If the high-frequency partial energy mean is less than or equal to the second weighted value of the low-frequency partial energy mean, the selected strategy is: selecting the entire low-frequency sub-band, and copying the low-frequency sub-band in the selected high frequency band having a large correlation with the entire low-frequency sub-band ;

 If the high-frequency partial energy average is greater than the second weighted value of the low-frequency partial energy mean, the selected strategy is: dividing the high-frequency into a plurality of replicated frequency bands, and selecting a highly correlated low-frequency sub-band for each replicated frequency band for copying.

 16. The encoding module of claim 14 wherein:

The policy selected by the replication policy selection module further includes: further moving to the low frequency when copying The high frequency sub-band of the sub-band is used as a copy source together with the selected low-frequency sub-band, and the high-frequency sub-band of the adjacent low-frequency sub-band is selected to be copied with a low-frequency sub-band having a large correlation.

 17. A decoding module, comprising a high frequency generator module, wherein the high frequency generator module comprises:

 a receiving unit, configured to receive high frequency reconstruction parameter information including a correspondence relationship of the selected frequency bands, where the corresponding relationship of the selected frequency bands is specifically a relationship between the high frequency sub-band determined by the reference correlation and the corresponding low frequency sub-band;

 And a reconstruction unit configured to copy the low frequency sub-band as a high frequency sub-band according to the high frequency reconstruction parameter information including the correspondence relationship of the selected frequency bands in the high frequency band.

 18. The decoding module of claim 17, wherein:

 The parameter information received by the receiving unit further includes a new algorithm using a flag and a band selection table replacement flag;

 The reconstruction unit determines an algorithm used by the copying process according to the value of the new algorithm using the flag, and determines a frequency band selection table used by the copying process according to the value of the frequency band selection table replacement flag, according to the determined algorithm and frequency band in the high frequency band. The selection table copies the low frequency sub-bands in the correspondence as high frequency sub-bands.