EP0910927B1 - Process for coding and decoding stereophonic spectral values - Google Patents

Abstract

A method of coding stereo audio spectral values first carries out grouping of those values in scale factor bands, with which scale factors are associated. Sections are formed next, each comprising at least one scale factor band. The spectral values are coded within at least one section with a code book assigned to the section, out of a plurality of code books each with a code book number assigned to it, the number of the code book used being transmitted as side information to the coded stereo audio spectral values. At least one additional code book number is provided, which does not refer to a code book but shows information relevant to the section to which it is assigned. A method of decoding stereo audio spectral values which are partly coded by the intensity stereo process and which have side information uses the relevant information, showing the additional code book numbers, to cancel the existing coding of the stereo audio spectral values.

Description

The present invention relates to coding and Decoding stereo audio spectral values and in particular on indicating the fact that stereo intensity coding is active.

Modern audio coding processes or decoding processes that work according to the MPEG Layer 3 standard, for example able to control the data rate of digital audio signals for example to compress by a factor of twelve without the Deteriorate noticeably.

In addition to a high coding gain in the individual channels, such as. the left channel L and the right channel R, is in Stereo case also the redundancy and irrelevance of the two channels exploited among each other. Known and already used Methods are the so-called MS stereo method (MS = Middle-side) and the intensity stereo method (IS method).

The MS stereo method known to those skilled in the art essentially uses the redundancy of the two channels with each other , with a sum of the two channels and a difference of the two channels, which is then calculated as modified channel data for the left or right channel be transmitted. The redundancy removed in the encoder between the two channels is again in the decoder added. This means that the MS stereo procedure is exactly reconstructive is.

In contrast, the intensity stereo method mainly uses the stereo irrelevance. Regarding stereo irrelevance is to be said that the spatial perception of the human hearing system from the frequency of the perceived Audio signals depends. At lower frequencies amount and phase information of both stereo signals assessed by the human auditory system, whereby the perception of high frequency components mainly on the analysis of the energy-time envelopes of both channels is justified. So that's the exact phase information of signals in both channels for spatial perception Not relevant. This property of human hearing is used to further reduce the stereo irrelevance of audio signals using the intensity stereo method to use.

Since the stereo intensity method at high frequencies none is able to resolve exact location information, it is therefore possible from an intensity limit frequency determined in the encoder instead of two stereo channels L, R a common energy envelope to transmit for both channels. In addition to of these common energy envelopes are roughly quantized Direction information additionally as side information transfer.

So because when using the intensity stereo coding a channel is only partially transmitted, the bit saving up to 50%. However, it should be noted that the IS method in the decoder is not exactly reconstructive.

In the IS process, which was previously in the standard MPEG layer 3 is used, is a so-called Modus_Erweiterungs_Bit (mode_extension_bit) indicated that the IS method in a block of stereo audio spectral values at all is active, with each block assigned to it Mode_extension_bit.

1 shows a basic illustration of the known IS method. Stereo audio spectral values for a channel L 10 and for a channel R 12 are summed at a summation point 14 in order to obtain an energy envelope I = L _i + R _{i of} the two channels. L _i and R _i here represent the stereo audio spectral values of channel L and channel R in any scale factor band. As already mentioned, the use of the IS method is only permitted above a certain IS cutoff frequency in order not to encode interference in the coded Introduce stereo audio spectral values. Therefore, the left and right channels must be coded separately in a range from 0 Hz to the IS cutoff frequency. The determination of the IS cutoff frequency as such is carried out in a separate algorithm which does not form part of this invention. From this limit frequency, the encoder encodes the sum signal of the left channel 10 and the right channel 12, which is formed at the summation point 14.

In addition to the energy envelope, i.e. the sum signal from left and right channel, for example in the coded left channel can be transmitted are also Scaling information 16 for channel L and scaling information 18 for channel R for decoding necessary. With the intensity stereo method as it is for example implemented in MPEG Layer 2 Transfer scale factors for the left and right channels. At this point, however, it should be noted that the IS method in MPEG Layer 3 for IS-coded stereo audio spectral values Intensity direction information only in right channel are transmitted, with which then like it is set out later, the stereo audio spectral values be decoded again.

The scaling information 16 and 18 is called side information each in addition to the coded spectral values of channel L and channel R transmitted. A decoder delivers on a decoded channel L '20 or on one decoded channel R '22 decoded audio signal values, where the scaling information 16 for the channel R and the Scaling information 18 for channel L with the decoded Stereo audio spectral values of the respective channels an L multiplier 24 or an R multiplier 26 multiplied by the originally encoded stereo audio spectral values to decode again.

Before applying IS coding above a certain one IS cutoff frequency or MS coding below this Cutoff frequency will be the stereo audio spectral values for each channel grouped into so-called scale factor bands. These bands are related to the perceptual properties of the hearing customized. Each of these bands can be equipped with an additional one Factor, the so-called scale factor, which is transmitted as page information for the respective channel and part of the scaling information 16 and the scaling information 18 from FIG. 1. These factors shape one a quantization of noise introduced in such a way that the same considering psychoacoustic considerations "masked" and thus becomes inaudible.

Figure 2a shows a format of the encoded right channel R which for example with an audio coding method MPEG layer 3 is used. All other statements regarding the intensity stereo coding refer to the method according to the standard MPEG Layer 3. In the first line in FIG. 2a are the individual scale factor bands 28 into which the stereo audio spectral values are grouped, schematically shown. The same bandwidth shown in Fig. 2a Scale factor bands only serve the clarity of the Representation and is in practice due to the psychoacoustic Features of the auditory system do not occur.

Coded stereo audio spectral values are in the second line of FIG. 2a sp, which is below an IS cutoff frequency 32 are non-zero, with the stereo audio spectral values in the right channel above the IS cutoff frequency, like already mentioned, set to zero (Zero_Part) nsp (nsp = Zero spectrum).

The third line of FIG. 2a contains part of the page information 34 for the right channel. This part of the side information 34 shown consists, on the one hand, of the scale factors skf for the area below the IS cut-off frequency and of direction information rinfo 36 for the area above the IS cut-off frequency 32. This directional information is also used in the intensity stereo method to ensure a rough spatial resolution of the IS-coded frequency range. This direction information rinfo 36, which is also called intensity positions (is_pos), is therefore transmitted in the right channel instead of the scale factors. It should be noted once again that below the IS cutoff frequency, the scale factors 34 corresponding to the scale factor bands 28 are still present in the right channel. The intensity positions 36 indicate the perceived stereo imaging position (the ratio from left to right) of the signal source within the respective scale factor bands 28. In each scale factor band 28 above the IS cutoff frequency, the decoded values of the transmitted stereo audio spectral values are scaled according to the MPEG Layer 3 method by the following scaling factors k _L for the left channel and k _R for the right channel: k L = is_ratio / (1 + is_ratio) and k R = 1 / (1 + is_ratio) The equation for is_ratio is as follows: is_ratio = tan (is_pos · π / 12)

The value is_pos is a value quantized with 3 bits, whereby only the values from 0 to 6 represent valid position values. From the following two equations, the left and right channels can be calculated back from the I signal (I = L _i + R _i ): R i = I · is_ratio / (1 + is_ratio) = I · k L L i = I · 1 / (1 + is_ratio) = I · k R R _i and L _i represent the intensity stereo decoded stereo audio spectral values. At this point it should be noted that the format of the left channel is analog to the format of the right channel shown in FIG. 2a, but in the left channel above the IS- Cutoff frequency 32, instead of the zero spectrum, the combined spectrum I = L _i + R _i can be found, and furthermore there is no direction information is_pos for the left channel, but normal scale factors. The transition from the quantized sum spectral values not equal to zero to the zero values in the right channel can implicitly indicate the IS cut-off frequency to the decoder in the MPEG Layer 3 standard.

In the encoder, the transmitted channel L is thus calculated as the sum of the left and right channels, and the transmitted direction information can be determined using the following equation: is_pos = nint [arctan (√E L / √E R ) · 12 / π]

The function nint [x] represents the function "next integer", where E _L and E _{R are} the energies in the respective scale factor bands of the left and right channels. This formulation of the encoder / decoder leads to an approximate reconstruction of signals in the left and in the right channel.

As already mentioned, known audio coding methods the stereo audio spectral values into the scale factor bands grouped, these tapes attached to the perceptual properties the hearing are adjusted. With the audio coding method according to the standard MPEG layer 3, these are scale factor bands now divided into exactly three regions. In order to Areas with the same signal statistics should now be grouped become. This is due to the redundancy reduction now taking place using the known Huffman coding advantageous. For each of these regions from scale factor bands 28 is now selected one of a plurality of Huffman tables in which the gain through redundancy reduction using Huffman coding using the selected Huffman table on greatest is. This table is encoded in the bit stream Data is displayed using a 5-bit value for each region. There are 30 different tables, the Tables 4 and 14 are not used.

The non-backward compatible NBC encoding method, which is currently in the standardization process different from the standard audio coding process MPEG Layer 3 among other things, that in the bitstream syntax for this Do not just process exactly three regions from scale factor bands are allowed, but that so-called sections or "Sections" can be present in any number and have any number of scale factor bands can. A section is now analogous to the previous one described method in MPEG Layer 3 to achieve a maximum redundancy reduction a corresponding Huffman table assigned from a plurality of such tables, which should then be used for decoding. In extreme cases for example, a section consists of only one single scale factor band. In practice, however, this will be tend not to occur because the necessary page information would be way too big. In the NBC process exist a total of 16 Huffman coding table numbers as 4-bit values be transmitted. This means that one of the twelve existing ones Coding table numbers can be selected.

The object of the present invention is methods for encoding or decoding stereo audio spectral values to create those for coding or decoding relevant information with minimal effort be signaled by side information.

This task is accomplished by a method of encoding stereo audio spectral values according to claim 1 and by a method for decoding partially using the intensity stereo method encoded stereo audio spectral values according to claim 2 solved.

The present invention is based on the finding that that additional coding table numbers that are not for reference used on coding tables, others for one Section can display relevant information. The additional" Coding table numbers are the coding table numbers, that do not refer to coding tables. By a 4-bit coding of twelve different coding table numbers are the numbers 13, 14 and 15 for an assignment with other information freely available to a certain extent. At a preferred embodiment of the present invention two (No. 14 and No. 15) of the three (No. 13, No. 14 and No. 15) used additional coding table numbers, on the one hand to an intensity coding present in a section and on the other hand to the mutual phase position of IS-coded stereo audio spectral values in two stereo channels point out.

The unused additional coding table number 13 can be used to adapt to Huffman coding point out.

Fig. 1

den Signalfluß bei einem Codierungs/Decodierungs-Schema nach dem Intensity-Stereo-Verfahren;

Fig. 2a

ein Format der Daten bei Vorliegen einer Stereo-Intensity-Codierung für den rechten Kanal für den Standard MPEG Layer 3;

Fig. 2b

ein Format der Daten bei Vorliegen einer Stereo-Intensity-Codierung für den rechten Kanal für das MPEG-NBC-Verfahren; und

Fig. 3

ein schematisches Blockschaltbild eines Decodierers, der die vorliegende Erfindung ausführt.

Preferred embodiments of the present invention are explained below with reference to the accompanying drawings. Show it:

Fig. 1

the signal flow in a coding / decoding scheme according to the intensity stereo method;

Fig. 2a

a format of the data in the presence of stereo intensity coding for the right channel for the standard MPEG Layer 3;

Fig. 2b

a format of the data in the presence of stereo intensity coding for the right channel for the MPEG-NBC method; and

Fig. 3

is a schematic block diagram of a decoder which carries out the present invention.

A method of encoding stereo audio spectral values and the method for decoding partially in the intensity stereo method encoded stereo audio spectral values according to a first embodiment of the present Invention use a novel presence signaling the intensity stereo coding within one Section. According to the present invention are also 16 coding table numbers available. In contrast to the stand however, only the first 12 coding table numbers correspond to the technology (No. 1 to No. 12) real coding tables. With the help of the last and the penultimate coding table number it is now signaled that within the section, to which this coding table number is assigned, the Stereo intensity method is used.

2b shows a format of the data for the right channel R in the presence of stereo intensity coding, the MPEG2-NBC method is used. The difference to Fig. 2a, or to the MPEG Layer 3 method, consists in that a user now has the flexibility, even above the IS cutoff frequency 32 is an intensity stereo coding of the Selective stereo audio spectral values for each section on or off. This is the IS cutoff frequency compared to MPEG Layer 3 actually not a real one Cutoff frequency more, because with the NBC method also above IS cutoff frequency, the IS coding is switched on or off again can be. This was not possible with Layer 3, i.e. the stereo audio spectral values above the IS cutoff frequency had IS coding for a section in any case, right up to the upper end of the spectral range IS encoded. The new NBC process must now not for the entire spectral range above the IS limit activate the IS coding, but allows the same also switching off the IS coding, so this signals is. Since according to the bitstream syntax for a section anyway a coding table number must be transmitted, multiply themselves in the described signaling according to the invention not even the page information ("overhead").

That in a section with IS coding for the right channel transferred scale factors are now also analog the state of the art the direction information 36, wherein these values themselves also a difference and Huffman coding be subjected. Standing in the right channel as already mentioned, in the scale factor bands the are not IS coded, no stereo audio spectral values, but a zero spectrum. The left channel contains encoded in IS Sections the sum signal of the left and the right channel. However, the sum signal is standardized in such a way that its energy within the respective scale factor bands after the IS decoding the energy of the left Channel corresponds. Therefore, the left channel in the case of a also used IS coding in the decoding device unchanged and does not have to be scaled back can be determined separately. The stereo audio spectral values the right channel can now from the stereo audio spectral values using the left channel of the direction information is_pos 36 that in the page information of the right channel are calculated become.

As described at the beginning, the stereo intensity method results according to the prior art, two coherent signals for the left or right channel, which are only in its amplitude, i.e. Intensity, depending on the Distinguish directional information is_pos 36 (equations (4) and (5)).

In the present invention, since the presence of the stereo intensity coding is signaled by means of two "unreal" coding table numbers, a phase relationship of the two channels to one another can be included. If the channels have the same phase position, the recalculation rule according to the invention to be carried out in the decoder is as follows: R i = 0.5 ^ (0.25is_pos (sfb)) L i , while in the case of an opposite phase, the spectrum is multiplied by -1, which results in the following equation for the calculation of the right channel: R i = (-1) * 0.5 ^ (0.25 * is_pos (sfb)) * Li. In the two previous equations, R _i denotes the back-calculated, ie decoded, stereo audio spectral values of the right channel. sfb denotes the scale factor band 28 to which the direction information is_pos 36 is assigned. L _i denotes the stereo audio spectral values of the left channel, which are adopted unchanged in the decoder.

The coding table number 15 now indicates whether the first retroactive accounting rule to be used while the coding table number 14 indicates that the second retroactive accounting rule to be used, i.e. that the two channels are in phase opposition. It is obvious to experts that the expressions in-phase and in-phase in the sense this application are widely used. For example a phase discriminator can be provided, from one determined phase discriminator output value, for example Can be 90 °, determines that the signals are out of phase are the same with a phase difference of less 90 ° are considered to be in phase.

In the described first exemplary embodiment, therefore for a section that consists of at least one scale factor band consists of the coding table numbers 14 or 15 the Phase relationship of the two channels to each other can be determined. The Side information through IS and phase signaling are 8 bits for a section that consist of four bits for the section length and four bits for compile the coding table number 14 or 15. Should now encode an audio signal that is in scale factor bands its stereo audio spectral values make frequent changes in phase has, so according to the first embodiment with each reversal of the phase position of the scale factor band a new section ("section") for scale factor band be started. A signal with a frequently changing So phasing creates a lot of sections since everyone Section by the coding table number assigned to it only either in-phase or out-of-phase of its stereo audio spectral values can display in the two channels. An unfavorable signal therefore becomes a large number of sections and thus a large amount of page information to lead.

A second embodiment of the present invention allows a phase factor coding in scale factor band a section in which the intensity coding is active. By this method according to the second embodiment the present invention thus succeeds using an MS mask, which is described below, a scale factor band Phase position coding without an enlargement the number of sections and without an additional one Extra effort.

It is obvious to those skilled in the art that the middle-side procedure and the intensity stereo method in one Exclude the scale factor band from each other. These two procedures are therefore orthogonal.

Is an MS coding of stereo audio spectral values in one Bit stream is used, so a signaling bit in according to the page information, the globally turn on the MS coding. Setting this bit states that an MS bit mask is transmitted with which it MS coding is possible selectively for each scale factor band (scfbd) on or off. For each scale factor band is a bit reserved in the MS bit mask, which is why the length of the bit mask corresponds to the number of scale factor bands.

In the scale factor bands in which IS is active, the MS scale factor information is not necessary because the MS coding must not be activated here. The MS bit mask can be in this area can be used for other signaling. It is therefore possible to use the MS bit mask to detail the Display IS coding. In accordance with the first The information is exemplary embodiment in IS coding regarding the phase relationship of the channels in a section by means of the code table numbers 14 and 15. The coding table numbers also indicate that in a Section the IS coding is active at all.

In deviation from the first embodiment, the second embodiment of the present invention MS bitmask used to scale scale bands in a section with different phases. The MS bit mask now serves, in relation to the coding table number, which signals that an IS coding in one phase is active, the phase position of the individual scale factor bands in this section. Is a bit not set for a scale factor band in the MS bit mask (i.e. zero), they are identified by the coding table number for the section in which the scale factor band is located, Keep phase information displayed while with a set (i.e. one) bit in the MS bit mask for the scale factor band is given by the coding table number for the section where the scale factor band is located displayed phase position of the two channels is inverted. in the The principle is therefore an EXCLUSIVE-OR operation between the one indicated by the code table number Phase position and the MS bit mask.

Specifically, the phase relationships of the two stereo channels L and R calculated from the coding table number and MS bit mask in a scale factor band located in a section in which the IS coding is used are as follows: Coding table number (for a section) 15 15 14 14 MS - bit mask (for a scale scale band) 0 1 0 1 Phase position of L and R 0 180 ° 180 ° 0 ° Retroactive accounting regulation Eq. 7 Eq. 8th Eq. 8th Eq. 7

The described second exemplary embodiment of the present invention thus allows scale factor bands with stereo audio spectral values with different phase positions to occur in one section, as a result of which fewer sections than in the first exemplary embodiment have to be formed for coding. This means that less page information also has to be transmitted.

In deviation from the embodiment described above can also use the additional coding table numbers information relevant to a section is displayed become.

Further information relevant to a section can be found for example an indication of the use of an adaptive Be Huffman coding in one section. At a adaptive Huffman coding can depend on the signal statistics an adapted Huffman table can be generated. The coding table number 13 has the coding device not to use any of the twelve fixed Huffman tables, but to use an adapted Huffman table that is not known a priori to the decoder. Then this is from Advantage if the signal statistics are not in a section ideally with one of the twelve predefined coding tables encoded, i.e. can be compressed. The coding is no longer restricted to the twelve fixed Huffman tables, but can optimally match the signal statistics Create and use the adapted table. The information via the adaptive coding table are used as additional page information transfer.

A decoding device needs this additional page information, to get out of the same ones when coding used Huffman adapted table to calculate the Huffman-encoded stereo audio spectral values are correct again to be able to decode.

3 shows a simplified block diagram of a decoder, which is the method for decoding according to the present Can carry out invention. Partly in the intensity stereo process encoded audio spectral values are inverse Quantizers 38 and 40 fed, the inverse Quantizer the quantization introduced during coding undo. Then the dequantized Stereo audio spectral values in an MS decoder 42. This MS decoder 42 does the one introduced in the encoder Undo mid-side encoding. An IS decoder 44 now uses the retroactive accounting rules described above (7) and (8) to restore the original stereo audio spectral values also for the IS-coded scale factor bands to obtain. Respective reverse transformation devices for the left or right channel now lead one Conversion of the stereo audio spectral values into stereo audio time values L (t), R (t) through. It’s obvious to professionals that the reverse transformation devices 46 and 48 for example can be performed by an inverse MDCT.

Claims (10)

1. A method of coding stereo audio spectral values, comprising the following steps:

   grouping the stereo audio spectral values in scale factor bands (28) with which scale factors are associated;

   forming sections, each comprising at least one scale factor band (28);

   coding the stereo audio spectral values within at least one section with a code book, allocated to the at least one section, out of a plurality of code books to each of which a number is assigned, the number of the code book used being transmitted as side information to the coded stereo audio spectral values,

   characterised in that at least one additional code book number is provided, which does not refer to a code book but shows information relevant to the section to which it is assigned, and one section has either a code book number or the at least one additional code book number assigned to it, without affecting the amount of side information.
2. A method of decoding coded stereo audio spectral values which have side information, comprising the following steps:

   detecting a code book number on the basis of the side information for each section of the coded stereo audio spectral values; and

   decoding the stereo audio spectral values of a section, the code book number of which refers to a corresponding code book, using that code book;

   characterised by the following step:
   decoding the stereo audio spectral values of another section with a code book number which does not refer to a code book but shows information relevant to the section to which it is assigned, in accordance with the information shown.
3. A method according to claim I or 2,
   wherein at least one additional code book number refers to coding of the stereo audio spectral values of the associated section by the intensity stereo process.
4. A method according to any of the preceding claims,
   wherein at least one additional code book number refers to adaptive Huffman coding of the stereo audio spectral values of the associated section.
5. A method according to any of the preceding claims,
   wherein the at least one additional code book number, for a section coded by the intensity stereo process, also indicates a phase relationship between two stereo channels.
6. A method according to claim 5,
   wherein one of two additional code book numbers indicates an identical phase position of the two stereo channels, in which case the following calculating-back formula applies for intensity decoding: Ri = 0.5 ^ (0.25 · is_pos(sfb)) · Li, where R_i are the stereo audio spectral values of a right (R) channel, is _pos represents intensity direction information for the scale factor band sfb present, and L_i are the stereo audio spectral values of a left (L) channel.
7. A method according to claim 5 or 6,
   wherein one of two additional code book numbers indicates an opposing phase position of the two stereo channels, in which case the following calculating-back formula applies for intensity decoding: Ri = (-1) · 0.5 ^ (0.25 · is_pos(sfb)) · Li. where R_i are the stereo audio spectral values of a right (R) channel, is _pos represents intensity direction information for the scale factor band sfb present, and L_i are the stereo audio spectral values of a left (L) channel.
8. A method according to any of the preceding claims,
   wherein the intensity stereo process forms a standardised sum signal of the stereo audio spectral values of the left and right channel, in a left channel, while in the right channel the spectrum is zero and intensity direction information is coded as side information.
9. A method according to any of the preceding claims,
   wherein a bit mask which has a bit for each scale factor is used, and a bit on the mask for a scale factor band is gated with the additional code book number in a section to which one of the additional code book numbers is assigned, to determine a phase relationship for two stereo channels.
10. A method according to claim 9,
    wherein the bit mask is an MS bit mask, and the additional code book numbers are linked with the MS bit mask scale factor bandwise by means of an EXCLUSIVE-OR gate.

Images