HK1093814B - Method and device for quantizing a data signal - Google Patents

Description

The present invention relates in general to the quantization of information signals and in some embodiments to the quantization of audio signals, such as those used for data compression of audio signals or for audio coding.

The most commonly used audio compression method is MPEG-1 Layer III. In this compression method, the sound values of an audio signal are encoded into a coded signal, which is loss-sensitive. In other words, in compression, the irrelevance and redundancy of the original audio signal are reduced or ideally removed. To achieve this, a psychoacoustic model detects simultaneous and temporal masking, i.e. a time-varying masking threshold is calculated, which depends on the audio signal, or which indicates from which frequency sounds of a certain frequency are perceivable to the human brain.

Err1:Expecting ',' delimiter: line 1 column 488 (char 487)

Err1:Expecting ',' delimiter: line 1 column 467 (char 466)

Err1:Expecting ',' delimiter: line 1 column 358 (char 357)

The above article proposes a block size of 128 sampling values 906 which allows a relatively short delay of 8 ms at 32 kHz sampling rates. As for the detailed implementation, the article further describes that to increase the efficiency of the page information encoding the page information, namely the coefficients x# and a#, is only transmitted if there is sufficient change compared to a previously transmitted set of parameters, i.e. if the change exceeds a certain threshold. The article also describes that the implementation is carried out in such a way that a current parameter substitution is not directly used on all respective block encoding values, namely the coefficients x# and a#, but only if there is a sufficient change compared to a previously transmitted set of parameters, i.e. if the change exceeds a certain threshold.

Although the audio coding scheme described in the above-quoted article already reduces the delay time sufficiently for many applications, a problem with the above scheme is that the need to transmit the masking threshold or transmission function of the coding filter, hereinafter referred to as the prefilter, places a relatively high strain on the transmission channel, although the filter coefficients are only transmitted if a predetermined threshold is exceeded.

Another disadvantage of the above coding schemes is that, because the masking threshold or inverse thereof must be provided by the x# parameter set to be transmitted on the decoder side, a compromise must be made between the lowest possible bitrate or high compression ratio on the one hand and the most accurate approximation or parametrization of the masking threshold or inverse thereof on the other.As can be seen, there are areas where the transmission function of the decoder-side filter, hereinafter also called post-filter, exceeds the masking threshold b. The problem is now aggravated by the fact that the parameterization is only transmitted and interpolated intermittently when there is sufficient change between the parameterizations. An interpolation of the x# filter coefficients, as proposed in the article, alone leads to audible disturbances if the amplification value a# is kept constant from base to base or from new parameterization to new parameterization.Other Another problem with the audio coding scheme shown in Fig. 12 and 13 is that the filtered signal can take an unpredictable form due to frequency selective filtering, in which, in particular due to a random overlap of many individual overwaves, a single or individual audio value of the coded signal adds up to very high values, which, due to their rarity, in turn lead to a worse compression ratio in the subsequent reduction of redundancy.

The present invention is intended to provide a method and device for quantizing an information signal so that a higher data compression of the information signal is possible with only a slight deterioration in the quality of the original information signal.

This task is solved by a procedure as described in claim 9 and a device as described in claim 1.

A quantization of an information signal of a sequence of information values according to the invention involves the frequency selective filtering of the sequence of information values to obtain a sequence of filtered information values and the quantization of the filtered information values to obtain a sequence of quantized information values by means of a quantization step function that maps the filtered information values to the quantized information values and whose path is steeper below a threshold information value than above the threshold information value.

It has been observed that by frequency selective filtering of an audio signal in the resulting filtered information signal artificially generated artifacts are produced in which individual information values, due to a random constructive interference of all or many of the overwaves, take on values significantly higher than the maximum values of the original signal, such as more than twice as high. The core idea of the present invention is that cutting the filtered information signal above a suitable threshold, for example twice as much as the maximum possible upper limit of the original information signal to be filtered, so that the artifact values produced by the frequency selective filtering are removed from the signal but are not properly filtered or the filtered information signal is reduced to a quantum value, whereas cutting the filtered information signal back into a quantum value after filtering results in a reduction in the quality of the information, but not in the quantum value of the filtered information, whereas cutting the information is reduced to a quantum value, or a reduction in the quality of the information filter, where the filtered information is reduced to a value of one bit, but not much less than the value of the filtered information.

According to a preferred embodiment, the information signal is an audio signal in which selective quantization above or below a certain threshold results in a barely audible reduction in audio quality while simultaneously greatly reducing the bit rate.

The quantization step function may alternatively be used to quantize all audio values above the threshold to a highest quantization step or a quantization step function may be used that is flatter above the threshold or has a larger quantization step range above the threshold, so that artificially generated artifacts are more roughly quantized.

The following are examples of preferred embodiments of the present invention, which are described in more detail in the accompanying drawings: Fig. 1a block diagram of an audio encoder according to an example of the present invention;Fig. 2a flowchart illustrating the operation of the audio encoder of Fig. 1 at the data input;Fig. 3a flowchart illustrating the operation of the audio encoder of Fig. 1 with regard to the evaluation of the incoming audio signal by a psychoacoustic model;Fig. 4a flowchart illustrating the operation of the audio encoder of Fig. 1 with regard to the application of the parameters obtained from the psychoacoustic model to the incoming audio signal;Fig. 5a schematic diagram illustrating the audio signal obtained from the evaluation of the results of the analysis of the audio signal;Figure 4 shows a schematic diagram illustrating the structure of the encoded signal;Figure 6 shows a flow diagram illustrating the operation of the audio coding device in Figure 1 from the final processing to the encoded signal;Figure 7 shows a diagram illustrating an example of an execution of a quantization step function;Figure 7 shows another example of an execution of a quantization step function;Figure 8 shows a block-switch image of an audio coding device capable of decoding an audio by decoding the decoded signal.Figure 1 shows an example of an execution of the coding device in accordance with Figure 9 of the present invention.8 at the data input;Fig. 10a flowchart illustrating the operation of the Decoder of Fig. 8 in terms of the interstoring of the pre-decoded, quantized, filtered audio data and the processing of the audio blocks without associated page information;Fig. 11a flowchart illustrating the operation of the Decoder of Fig. 8 in terms of the actual re-filtering;Fig. 12a schematic diagram illustrating a conventional audio decoder scheme with a short delay time; andFig. 13a diagram illustrating the transmission spectrum of an audio signal, a metric threshold and the decoder function of the post-filter.

Figure 1 shows an audio encoder according to an example of the present invention. The audio encoder, generally designated by 10, comprises first a data input 12 on which it receives the audio signal to be encoded, which, as will be explained in greater detail later with reference to Figure 5a, consists of a sequence of audio values or sampling values, and a data output on which the encoded signal is output, the information content of which is discussed in greater detail with reference to Figure 5b.

The audio encoder 10 of Fig. 1 is divided into an irrelevance reduction part 16 and a redundancy reduction part 18. The irrelevance reduction part 16 comprises a device 20 for detecting a miter threshold, a device 22 for calculating an amplification value, a device 24 for calculating a parametrization, a base comparison device 26, a quantizer 28 and a parameterizable prefilter 30, and an input FIFO (first-in-first-out) buffer 32, a buffer or memory 38 and a multiplier or multiplier device 40. The redundancy reduction part 18 comprises a bitrate controller 34 and a bitrate compressor 36.

In particular, data input 12 is connected to a data input of the device 20 for detecting a miter threshold and a data input of the input buffer 32. A data output of the device 20 for detecting a miter threshold is connected to an input of the device 24 for calculating a parametrization and to a data input of the device 22 for calculating an amplification value to transmit to the same determined miter threshold. The devices 22 and 24 calculate based on the miter threshold parametrization or amplifier and are connected to the comparison device 26 to transmit these results to the same station.The parameterized pre-filter 30 is switched between a data output of the input buffer 32 and a data input of the buffer 38. The multiplier 40 is switched between a data output of the buffer 38 and the quantizer 28. The quantizer 28 forwards to the redundancy reduction 18 scaled or multiplied, if necessary, but in any case quantized, filtered audio input values, and more precisely to a data leak of the compressor 34.The bitrate control is connected via a control link to a control input of the multiplier 40 to ensure that the quantized filtered audio values received from the prefilter 30 are multiplied by the multiplier 40 with an appropriate multiplier, as discussed in more detail below. The bitrate control 36 is switched between an output of the compressor 34 and the output 14 of the audio codec 10 to determine the multiplier suitable for the audio multiplier 40. Each value quantum multiplied by the multiplier 40 is first set to one of the calibration factors,However, buffer 38 continues to store any filtered audio value to give bitrate control 36 an opportunity to change the multiplier for another run of a block of audio values, as described below. If such a change is not displayed by bitrate control 36, buffer 38 may release the memory taken up by that block.

The structure of the audio coding device in Fig. 1 has been described above, and its functioning is described below by reference to Fig. 2-7b.

As can be seen from Figure 2, when the audio signal reaches data input 12, it has already been received by audio signal sampling 50 from an analogue audio signal. The audio signal sampling is carried out at a predetermined sampling frequency, which is usually between 32 and 48 kHz. Consequently, at data input 12 there is an audio signal consisting of a series of sampling or audio values. Although, as will be shown in the following description, the coding of the audio signal is not block-based, the values at the audio input table are first summed up into 52 audio blocks in one step. The summing of audio blocks is carried out in a single audio sample, it follows that only two consecutive summing of values are possible, as shown in the following example.

Fig. 5a shows the sequence of samples with 54, each sample being illustrated by a rectangle 56. The samples are numbered for illustration purposes, but for the sake of clarity only some of the samples of the sequence 54 are shown. As shown by the brackets above the sequence 54, the present example shows 128 consecutive samples being combined into one block, with the 128 samples immediately following forming the next block.

While the audio blocks grouped in step 52 in the device 20 are processed in the device 20 to determine a miter threshold, the input buffer 32 buffers or interpolates the input audio values 54 until the parametrizable pre-filter 30 has received 26 input parameters from the base comparison device to perform pre-filtering as described below.

As shown in Fig. 3, the device 20 for detecting a miter threshold begins processing immediately after sufficient audio values have been received at the data input 12 to form an audio block or the next audio block, which is monitored by the device 20 by a check at step 60. If an editable complete audio block is not yet available, the device 20 waits. If a complete audio block is available to be processed, the device 20 for detecting a miter threshold at step 62 is calculated on the basis of a suitable psychoacoustic model of a miter in step 62. To illustrate the miter threshold, a mural wave is used. In particular, Fig. 12 refers to the miter wave curve, which is based on a psychological model of a real-world phenomenon, such as a frequency variation in the frequency of a sound wave, which can be obtained by means of a classical noise mask, and which can not be transmitted in any frequency range.

In a subsequent step 64, device 24 and device 22 calculate an amplification value a or a set of parameters from N parameters x ((i) (i = 1, ..., N) from the calculated miter threshold M ((f) (where f indicates the frequency). The parameterization x ((i) that device 24 calculates in step 64 is for the parameterizable prefilter 30, which is used, for example, in an adaptive filter structure, as used in LPC coding. For example, if s (n) = n 0, ... 127 the 128 instantaneous values of the audio audio lock and s (n) = 128 the resulting filter values are then satisfied, for example, the following equation is fulfilled: $\mathrm{sʹ}\left(n\right)=s\left(n\right)-\sum _{k=1}^K{a}_k^{t}s\left(n,-,k\right),$ Other where K is the filter order and $a_k^t$ The filter coefficients are k = 1, ..., K and the index t is to illustrate that the filter coefficients change with successive audio blocks. $a_k^t$ The transmission function H (f) of the pre-filter 30 is approximately equal to the inverse of the masking threshold M (f), i.e. $H\left(f,,t\right)\approx \frac{1}{\left(M,\left(f,,t\right)\right)}$ Other The first is the use of the pre-filter 30 as the adaptive filter described above, whereby the dependence on t is used to illustrate that the masking threshold M (f) changes for different audio blocks. $a_k^t$ obtained as follows: the inverse discrete Fourier transform of M (f, t) ∈ 2 over the frequency for the block at time t gives the objective autocorrelation function $r_{\mathrm{mm}}^t\left(i\right)\mathrm{.}$ Then they will . $a_k^t$ obtained by solving the linear system of equations: $\sum _{k=0}^{K-1}{r}_{\mathrm{mm}}^t\left(\left(k,-,i\right)\right)\mathrm{.}{a}_k^t=r_{\mathrm{mm}}^t\left(i,-,1\right),0\le i<K\mathrm{.}$ Other However, in order to avoid instabilities in the linear interpolation between the parametrizations described below, it is preferable to use a lattice structure for filter 30 whereby the filter coefficients for the lattice structure are re-parametricated into reflection coefficients.

Thus, while the device 24 calculates a parametrization for the parameterizable pre-filter 30 such that its transmission function is equal to the inverse of the masking threshold, the device 22 calculates a noise power limit based on the noise threshold, namely a limit that specifies what noise power the quantizer 28 may introduce into the audio signal filtered by the pre-filter 30 so that the quantization noise quantum after the return or post-filtering on the decoder side is still below the noise threshold (f) or on the same quantum. The device 22 calculates this noise power limit as the area below the noise noise noise volume quadrant, i.e. as ΣMf. The noise power quantum is calculated by the quantization of the noise noise noise frequency and is determined by the noise noise noise noise limit expressed by the noise noise noise noise frequency.

As shown in the previous description, the device 22 calculates the NOx limit in addition to the amplification value a. Although it is possible for the reference reference device 26 to recalculate the NOX limit from the amplification value a received by the device 22, it is also possible for the reference reference device 26 to transmit the determined NOX limit in addition to the amplification value a.

After the calculation of the amplification value and the parameterization, the reference reference device 26 then checks in step 66 whether the parameterization just calculated differs by more than one predetermined threshold from the current parameterization, which was last passed on to the parameterizable pre-filter.If this is not the case and the parametrization just calculated does not differ by more than the predetermined threshold from the current one, the bench comparator (26) passes to the pre-filter 30 in step 72 only the current bench parametrization instead of the just calculated parametrization, i.e. the one that last produced a positive result in step 66, i.e. by more than a predetermined threshold from a previous bench parametrization.

In the event that the parametrization just calculated does not differ from the current base parametrization and the prefilter 30 therefore receives back in step 72 the base parametrization already obtained for at least the last audio block, the prefilter 30 applies this base parametrization to all the sampling values of this audio block in the FIFO 32 as described in more detail below, thus extracting this current block from the FIFO 32 and the quantizer 28 receives a resulting audio block of prefiltered audio values.

Fig. 4 shows the operation of the parameterizable prefilter 30 in the case that it receives the parameterization and the amplification just calculated because they differ sufficiently from the current base parameterization. As described with reference to Fig. 3, therefore, not each of the successive audio blocks is processed according to Fig. 4, but only to audio blocks where the associated parameterization differs sufficiently from the current base parameterization. The other audio blocks are filtered, as described, by the fact that the respective current base parameterization and the associated current base parameterization have been sufficiently overestimated.

Once such a transfer has taken place, the parameterizable pre-filter 30 starts processing the current audio block of audio values which is currently in the intermediate memory 32, i.e. the one for which the parameterization has just been calculated. For example, Fig. 5a shows that all audio values 56 before the audio value with the number 0 have already been processed and therefore passed through the memory 32.

Err1:Expecting ',' delimiter: line 1 column 541 (char 540)

However, the parametrization calculated for the block 1 still in FIFO 32 differed from the parametrization x0 (i) by more than the threshold specified in the example example in Fig. 5a and was therefore transmitted to the pre-filter 30 as parametrization x1 (i) in step 68 together with the amplification value a1 (step 70) and, where appropriate, the associated noise limit, the indices of a and x in Fig. 5a being an index of the bases as they are used in the interpolation to be investigated later on with regard to the abdominal values 128-255 in block 1 by a pulley 82 and the following steps 80 in Fig. 4 The cliff face would be used to start processing the audible at step 80 hence the number of cliffs.

At the time of transmission of the parameter set a1, x1, the memory 32 therefore contains only the audio values 128 - 255, i.e. the current audio block after the last audio block processed by the prefilter 30, 0. Now, after the transfer of base parameters x1 (i) has been determined in step 80, the prefilter 30 determines in step 84 the noise power limit q1 corresponding to the amplification value a1. This can be done by the base comparator 26 transmitting this value to the prefilter 30, or by re-calculating this value by the prefilter 30, as described in the previous reference to step 64.

Err1:Expecting ',' delimiter: line 1 column 244 (char 243)

The parameterizable pre-filter 30 performs in step 88 the interpolation of the filter coefficients x0, x1 between the two bases in the form of a linear interpolation to obtain the interpolated filter coefficient at the sampling position j, i.e. x(tj) ((i) with i = 1...N).

Then, at step 90, the parameterizable pre-filter 30 performs an interpolation between the noise limit q1 and q0 to obtain an interpolated noise limit at the sampling position j, i.e. q(tj).

In step 92 the parameterizable prefilter 30 then calculates the amplification value for the sampling position j on the basis of the interpolated noise limit and the quantization noise power and preferably also the interpolated filter coefficients, i.e. depending on the root. $\frac{\mathrm{Quantisierungsrauschleistung}}{q\left(t_j\right)},$ referring to the explanations given in step 64 of Figure 3.

In step 94, the pre-filter 30 shall then apply the calculated gain and the interpolated filter coefficients to the sampling value at sampling position j to obtain a filtered sampling value for that sampling position, namely s'(tj.

In step 96 the parameterizable prefilter 30 checks whether the sampling position j has reached the current support, i.e. support 1, in the case of Fig. 5a the sampling position 255, i.e. the sampling value for which the parameterization plus amplification value transmitted to the parameterizable prefilter 30 is to be applied directly, i.e. without interpolation. If this is not the case, the parameterizable prefilter 30 increases or increments the index j by 1, repeating steps 88-96. However, if the check in step 96 is positive, in step 100 the parameterizable prefilter is to be moved from the support direction 26 to the last comparable amplification value of the intermediate amplification direction 80 (i.e. without interpolation) and the next analogue filter is to be processed immediately after the next analogue block 26 is set up, but in case of the latter, the sampling process may be carried out at the next audioblock, without interruption, in the case of the next analogue block 26 or in the case of the next analogue block 26 (i.e.g. the next analogue block 26), and if necessary, at the next audioblock, at the next analogue sound level.

Before describing the further process of processing the filtered sample values s' with reference to Fig. 5, the purpose and background of the process described in Fig. 3 and 4 are described below. The purpose of filtering is to filter the audio signal at input 12 with an adaptive filter whose transmission function is constantly optimally adapted to the inverse of the miter threshold, which also changes in time. This is because the decoder side of the re-filtering is by an adaptive filter whose transmission function is constantly adapted to the miter threshold, introduced by a quantization of the filtered audio noise, i.e. in the form of a constant quantum free noise, i.e. the quantum form of the miter.

The purpose of this is to quantitatively reset the amplification noise which is introduced into the filtered audio signal by the subsequent quantization described in detail and which is quantitatively adjusted to the shape of the miter wave by the re-filtering decoder side so that the miter wave power is not exceeded. The objective is to reduce the quantitatively-adjusted noise by the number of pre-value sounds, which is proportional to the quantitatively-adjusted value of the miter wave. This is also done by the use of the decoder side, which is proportional to the miter wave power.

In other words, the prefilter effect can be thought of as a normalization of the signal to its masking threshold, so that the level of quantization interference or quantization noise can be kept constant in both time and frequency. Since the audio signal is in the time range, therefore, quantization can be done gradually with a uniform constant quantization, as described below. This ideally removes any irrelevance from the audio signal, and a lossless compression scheme can be used to remove the remaining redundancy in the filtered and quantized audio signal as described below.

It is also to be emphasized once more clearly from Fig. 5a that although the filter coefficients and amplification values a0, a1, x0, x1 used must be available on the decoder side as side information, the transmission effort is reduced by not simply re-using new filter coefficients and new amplification values for each block. Rather, a threshold check 66 is carried out in order to transmit the parameterizations as side information only if there is a sufficient change in parameterization, and otherwise the side information or interpolation is not transmitted.

The following Figure 6 describes the further processing of the pre-filtered signal, which essentially involves quantization and redundancy reduction: first, the filtered sound values emitted by the parameterizable pre-filter 30 are stored in buffer 38 and simultaneously passed from buffer 38 to the quantum multiplier 40, where they are again, as it is their first quantum flow, initially unchanged, namely with a scaling factor of one, by the quantum multiplier 40 to the quantum multiplier 28.

The quantized filtered sampling values are denoted by σ' in Fig. 7a. The quantization step function is preferably a quantization step function with a constant step range below the threshold value, i.e. the jump to the next quantization step always takes place after a constant interval along the input values S'. In the implementation, the step range is set to the threshold value such that the number of quantization steps is preferably a power of 2. Compared to the flow representation of the incoming filtered sampling values s', the wave is smaller, so that a maximum value of the representation range of the flow commodity reaches the sword.

The reason for the threshold is that it has been observed that the filtered audio signal emitted by the prefilter 30 has isolated audio values which add up to very large values due to an unfavourable accumulation of surface waves; furthermore, it has been observed that cutting these values, as achieved by the quantization step function shown in Figure 7a, results in a high data reduction but only a slight impairment of the audio quality; rather, these isolated places in the filtered audio are artificially created by the frequency selective filtering in the parameterized filter 30, so that cutting the same places only slightly affects the audio quality.

A somewhat more concrete example of the quantization step function shown in Fig. 7a would be one that rounds all filtered sampling values s' to the nearest integer up to the threshold value and then quantizes all filtered sampling values above that to the highest quantization step, such as 256.

Another example of a possible quantization step function would be the one shown in Fig. 7b. Up to the threshold, the quantization step function of Fig. 7b is similar to that of Fig. 7a. However, instead of abruptly flattening for sampling values s' above the threshold, the quantization step function continues with a steepness smaller than the steepness in the area below the threshold. In other words, above the threshold, the quantization step scale is larger. This achieves a similar effect to the quantization function of Fig. 7a, but with a much greater effort due to the different step scales of the quantization steps above and below the s' of the wave and on the other hand, because high-quality audio is not fully quantized, but only with a much greater quantization step scale.

As described above, the decoder side must have not only the quantized and filtered audio values σ' but also the input parameters for the pre-filter 30 which have been used to filter these values, namely the base parameterization including an indication of the corresponding amplification value. In step 114 the compressor 34 therefore makes a first compression attempt and compresses page information including the amplification values a0 and a1 at the bases, e.g. 127 and 255, and the filter coefficients x0 and x1 at the base and the quantized filtered values σ' in a pre-filtered signal. The compressor 34 is a pre-filter or Codifier 34 without loss of processing and adapter.

The memory 38 through which the sound values σ' are passed is used as a buffer for an appropriate block size, with which the compressor 34 processes the quantized, filtered and, if necessary, scaled sound values σ' output by the quantizer 28 as described below.

As mentioned above, for the first compression attempt, the bitrate control 36 controlled the multiplier 40 with a multiplier of 1 so that the filtered audio values from the prefilter 30 reached the quantizer 28 unchanged and from there the compressor 34 as quantized, filtered audio values. The compressor 34 monitors in a further step 116 whether a certain compression block size, i.e. a certain number of quantized, sampled audio values, has been encoded into the provisional encoded signal or whether further quantized, filtered audio values σ' are to be encoded into the current provisional encoded signal.However, if the compression block size is reached, in step 118 the bitrate control 36 checks whether the bit amount required for compression is greater than a bit amount required by a desired bit rate. If this is not the case, in step 120 the bitrate control 36 checks whether the bit amount required is less than the bit amount required by the desired bit rate. If this is the case, the bitrate control 36 adds the encoded signal in step 122 with fill bits until the bit amount required by the desired bit rate is reached.Finally, the compressor shall pass the compressor block of filtered audio values σ' in a form multiplied by the multiplier 40 by a factor of 1 to the quantizer 28 to repeat steps 110 to 118 until the bit rate required by the desired bitrate is reached, as indicated by a dashed step 125.

However, if the check in step 118 shows that the required bit rate is greater than the required bit rate, the bit rate control 36 changes the multiplier for the multiplier 40 to a factor between 0 and 1 only, which it does in step 126. After step 126, the bit rate control 36 ensures that the memory 38 re-issues the last compression block of filtered audio values σ' underlying the compression, which is then multiplied by the quantity factor set in step 126 and fed back to the quantizer 28, after which steps 110 to 118 are repeated and the previously coded signal is discarded.

It is noted that, if steps 110 to 116 are repeated in step 114, the factor used in step 126 (or step 125) is incorporated into the coded signal.

The point of the procedure in step 126 is that the factor increases the effective step range of the quantizer 28; this means that the resulting quantization noise is uniformly above the masking threshold, leading to audible interference or audible noise, but resulting in a reduced bitrate. If, after repeating steps 110 to 116 in step 118, it is again found that the required bitrate is greater than that prescribed by the further desired bitrate, the factor is reduced in step 126, and so on.

When the data is finally output as a coded signal at step 124, the next compression block is performed from the subsequent quantized filtered audio values σ'.

It should be noted that a pre-set value other than 1, for example 1, could be used for the multiplication factor.

Fig. 5b again illustrates the resulting coded signal, which is generally displayed with 130. The coded signal includes side information and main data in between. The side information includes, as already mentioned, information from which the value of the amplification value and the value of the filter coefficients can be derived for specific audio blocks, namely audio blocks in which a significant change in the filter coefficients has occurred as a result of audio blocks.Within the coded signal, the side information is preferably arranged in such a way that the side information on filter coefficients and associated amplification value or noise limit is placed before the main data on the audio block of quantized filtered audio values σ' from which these filter coefficients with associated amplification value or noise limit have been derived, i.e. the side information a0, x0i) after block -1 and the side information a1, x1 (i) after block 1.The audio values σ'(t0) - σ'(t127) have been obtained or decoded solely by means of the page information 132 as mentioned above in relation to Fig. 5a, while the audio values σ'(t128) - σ'(t255) have been obtained or decoded only by interpolation by means of the page information as support values at the base with the decoded page number 127 and the 134 information as support values instead of the decoded page information 255 as mentioned above.

For example, in Figure 5b, the side information block 132 contains only the amplification value a0 and filter coefficients x0 with respect to the base at time t-1. In the side information block 132 these values are derived from the block itself. However, the side information block 134 does not contain any more the page information about the hourly power at time t255 from the block itself. Most of the side information block 134 contains only information about the difference between the amplification value a0 and the filter coefficients x0 at the base at time t-1.

This way of incorporating the page information into the page information blocks 132 and 134 has the advantage of allowing a higher compression rate, because although the page information is transmitted only if possible after a sufficient change in the filter coefficients to the filter coefficients of a previous base, the effort of differentiation on the coding side and summation on the decoding side is worthwhile, since the resulting differences are small despite the query at step 66 and thus allow for advantages in entropy coding.

After the previous example of an audio encoder has been described, the following example of an audio encoder is described which is capable of decoding the encoded signal generated by the audio encoder 10 of Figure 1 into a decoded, playable or reprocessed audio signal.

The decoder, generally designated 210 is shown in Figure 8 and comprises a decompressor 212, a FIFO storage 214, a multiplier 216 and a parameterizable post filter 218. Decompressor 212, FIFO storage 214, multiplier 216 and parameterizable post filter 218 are in this order switched between a data 220 and a data output 222 of the decoder 210, receiving the encoded signal at data 220 and transmitting the decoded audio input at data output 222, which is only distinguished by the quantum amplification of the original audio signal generated by the audiocompressor 28 in the audio 10 at the data input 128. The decodifier 212 is connected to a further data input and a further multi-parameter amplification at the input 218.

As shown in Figure 9, the decompressor 212 decompresses the compressed signal at step 224 at data input 220 to arrive at the quantized filtered audio data, namely the σ' scan values and the corresponding page information in the page information blocks 132, 134 which show the filter coefficients and amplification values or, instead of the amplification values, the noise power limits at the bases.

As shown in Fig. 10, in step 226 the decompressor 212 checks the decompressed signal in the order of its arrival for the presence of side information with filter coefficients, in its completed form without differential reference to a previous side information block. In other words, the decompressor 212 searches for the first side information block 132. Once the decompressor 212 has been found, the quantized, associated filtered audio values σ' are stored in step 228 in FIFO memory 214. During step 228 a complete audio value of decompressed, quantum filtered audio σ' has been stored, so that a decompressed audio value is immediately followed by a post-block, thus obtaining information about the amplified audio in step 228 and the amplified information contained in step 226 and subsequent steps in the filter.

In step 230 the decompressor 212 monitors the decompressed signal for the appearance of some kind of page information block, namely with absolute filter coefficients or filter coefficient differences to a previous page information block. For example, in the example of Fig. 5b, the decompressor 212 would detect the appearance of page information block 134 at step 230 upon detection of page information block 132 at step 226.

As soon as the page information block 134 occurs, the decompressor 212 calculates in step 232 by adding the difference values in page information block 134 to the parameter values in page information block 132 the parameter values at the base 1, i.e. a1, x1 (i). However, step 232 is omitted if the current page information block is an in-house closed page information block without differences, which can be arranged every second, for example, as described above.To ensure that the waiting time for the decoder 210 is not too long, page information blocks 132, in which the single values, i.e. the difference values are absolutely unrelated to another page information block, are arranged in such a small number of decoded blocks that the time between the transmissions of the corresponding audio information blocks is not altered, for example, when the number of pages 132 is coded in the most appropriate way, or when the number of pages 132 is arranged in such a way that the number of decoded blocks is not altered in the most appropriate way, for example, when the transmission of the information is in the most efficient mode, or when the audio information blocks are arranged in such a way that the number of decoded blocks is not altered in the order to be transmitted between the corresponding pages 132 or 132 of the page information block.

As shown in Fig. 11, after a page information block for a new base has been reached, and in particular after step 226 or 232, a scan index j is initialised to 0 in step 234. This value corresponds to the scan position of the first scan value in the audio block currently remaining in FIFO 214 to which the current page information refers. Step 234 is performed by the parameterizable mail filter 218. The mail filter 218 then performs a noise limit calculation at the new base in one of the steps 236, which corresponds to step 84 in Fig. 4 and can, if necessary, be used to calculate the noise limit.For example, when the smoke limit is transferred to the bases in addition to the gain values, in subsequent steps 238 and 240 the post filter 218 then performs interpolations with respect to the filter coefficients and the noise limit corresponding to the interpolations 88 and 90 of Figure 4. The subsequent calculation of the gain value for the sampling position based on the interpolated noise limit and the interpolated filter coefficients from steps 238 and 240 in step 242 corresponds to step 92 of Figure 4.This step differs from step 94 in Fig. 4 in that the interpolated filter coefficients are applied to the quantized filtered sampling values σ' in such a way that the transmission function of the parameterizable post filter is not the inverse of the miter threshold but the miter threshold itself.

If the post filter 218 has not yet reached the current base with the sampling position j, which is checked in step 246, it increases the sampling position index j in step 248 and starts steps 238 to 246 again. Only when the base is reached does it apply the amplification value and filter coefficients of the new base to the sampling value at the base, namely in step 250. Again, as in step 218, the application involves, instead of a multiplication, a division by the amplification value and a filtering with a transmission function equal to the resonant wave and not inverted to the latter.

As mentioned above, filtering and applying the amplification value in steps 218 and 224 will adjust the noise added by the quantization in step 110 and 112 respectively to the mithaudible threshold in both form and height.

It should be noted that if the quantized filtered audio values have been subjected to further multiplication in step 126 due to bitrate control before encoding into the encoded signal, this factor can also be taken into account in steps 218 and 224.

With regard to Figures 3, 4, 6 and 9-11, it should be noted that the same flowcharts show the operation of the encoder in Figure 1 or the decoder in Figure 8 and that each of the steps shown in these flowcharts by a block is implemented in a corresponding device as described in the previous one. The implementation of the individual steps can be implemented in hardware, as an ASIC circuit, or in software, as subroutines. In particular, the explanations recorded in these figures show roughly which operation the respective step corresponding to the respective block refers to, while the arrows between the blocks are arranged in the order in which the encoder or decoder is operated.

For example, it is not necessary that a parametrization and an amplification value or a noise power limit, as determined for a particular audio block, be considered directly valid for a particular audio value, as in the previous example, the last audio value of each audio block, i.e. the 128th value in that audio block, so that this audio block can be interpolated; rather, it is possible to relate these support parameters to a support point located temporarily between the reference point n = 0, 64 and the noise power limit, as in the previous example, so that an interpolation between the mean and the maximum values of each audio block, or the maximum values of the audio blocks, is not necessary.

It should also be noted that the above example of implementation was an audio coding scheme designed to produce an encoded signal with a controlled bitrate, but bitrate control is not required in every application, so steps 116 to 122 and 126 and 125 respectively may be omitted.

With regard to the compression scheme mentioned in step 114, for the sake of completeness, reference is also made to the document by Schuller et al. described in the introduction to the description, and in particular to point IV, the contents of which are hereby incorporated by reference to the reduction of redundancy by lossless coding.

Although the present invention was described in the foregoing with reference to a special audio coding scheme which allows for short delay times, the present invention is certainly applicable to other audio coding schemes, for example an audio coding scheme where the coded signal consists of the quantized filtered audio values itself without any redundancy reduction being carried out would be conceivable, but it would also be conceivable to perform frequency selective filtering in a different way from that described above, namely by coding on the one side with a transmission function equal to the interference of the meter and on the other side with a transmission function equal to the interference of the meter.

Furthermore, individual aspects of the above embodiments can also be omitted, for example by reducing the compression ratio, it is also possible to transmit the page information for each audio block, to omit interpolation and/or to always transmit the parameters in the page information in self-contained page information blocks and not as differences relating to previous page information blocks.

Furthermore, the present invention is not limited to audio signals, but is also applicable to other information signals, such as video signals consisting of a sequence of frames, i.e. a sequence of pixel arrays.

In any case, the above audio coding scheme provides a way of limiting the bitrate in an audio coder with very low delay time. The bitrate peaks that arise in coding depending on the audio signal are avoided by limiting the output range of the prefilter. Although it is therefore consistent with the nature of the audio signals to be transmitted that they lead to different bitrates for transmission, namely more complex audio signals to higher bitrates and less complex ones to lower bitrates, an upper limit for the bitrate of transmission can always be observed, which often exists, for example, in overhead wireless transmission media.

In the above examples, the encoder consisted of a pre-filter that shaped the audio signal appropriately, a quantizer with a quantization step height, followed by an entropy encoder, which generated values also called indices.

Although it has been shown in the preceding example that the threshold for quantization is always constant and the quantization step function is always constant, i.e. the artifacts produced in the filtered audio signal are always quantized or cut with gross quantization, which could potentially quantum deteriorate the audio quality, this measure should be used only if it requires the complexity of the audio signal, i.e. if the bitrate required for coding exceeds a desired bitrate. In this case, in addition to the functions in Figures 7a and 7b, such a function could be used for example on the entire range of the audio signal, possibly with a constant and precise amplification of the filter.to use either the constant-rate quantization step quantization function or one of the quantization step functions of Fig. 7a or 7b, so that the signal could be communicated to the quantizer, in case of slight audio quality deterioration, to perform the quantization step reduction above the threshold or cutting above the threshold. Alternatively, the threshold could also be gradually reduced. In this case, the sword wave reduction could be performed instead of the factor reduction of step 126.In a second step, the filtered audio values would then be quantified with the quantization level function, which has a flatter path above the audio threshold.

In particular, it is noted that, depending on the circumstances, the quantization scheme of the invention may also be implemented in software. The implementation may be on a digital storage medium, in particular a floppy disk or a CD with electronically readable control signals, which can interact with a programmable computer system in such a way that the corresponding procedure is performed. In general, the invention thus also consists of a computer program product with program code stored on a machine-readable medium for performing the invention procedure, if the computer program product runs on a computer. In other words, the invention may thus be implemented as a program with a program program program for performing the procedure, if the computer code runs on a computer program.

In particular, the above steps in the flow diagram blocks may be implemented individually or in several in subprogram routines, or alternatively a device of the invention may be implemented in the form of an integrated circuit, in which these blocks are implemented, for example, as individual circuit components of an ASIC.

In particular, it is noted that, depending on the circumstances, the scheme of the invention may also be implemented in software. The implementation may be on a digital storage medium, in particular a floppy disk or a CD with electronically readable control signals, which can interact with a programmable computer system in such a way that the corresponding procedure is performed. In general, the invention thus also consists of a computer product program with program code stored on a machine-readable medium to perform the invention procedure, if the computer product runs on a computer. In other words, the invention can thus be realized as a computer program with program code to perform the procedure, if the computer runs on a computer.

Claims (10)

1. A device for quantizing an information signal of a sequence of information values, the information signal being an audio signal and the information values being audio values, comprising:

   means (20) for determining a first listening threshold for a block of audio values of a sequence of audio values;

   means (24) for calculating a version of a parameterization of a parameterizable filter so that the transfer function thereof roughly corresponds to the inverse of the magnitude of the first listening threshold;

   means (30) for frequency-selective filtering the sequence of audio values to obtain a sequence of filtered audio values;

   means (28) for quantizing the filtered audio values to obtain a sequence of quantized audio values by means of a quantizing step function which maps the filtered audio values to the quantized audio values and the course of which is steeper below a threshold information value than above the threshold information value;

   wherein the means (30) for frequency-selective filtering comprises:

   means for filtering a predetermined block of audio values of the sequence of audio values with the parameterizable filter using a predetermined parameterization depending in a predetermined manner on the version of the parameterization to obtain a block of the filtered audio values.
2. The device according to claim 1, wherein the means for determining a listening threshold is formed to further determine another second listening threshold for another second block of audio values, and the means for calculating is formed to calculate a version of another second parameterization of the parameterizable filter such that the transfer function thereof roughly corresponds to the inverse of the magnitude of the second listening threshold, wherein the means for frequency-selective filtering comprises:

   means for interpolating between the version of the first parameterization and the version of the second parameterization to obtain a version of an interpolated parameterization for a predetermined audio value of the predetermined block of audio values; and

   means for applying the version of the interpolated parameterization to the predetermined audio value of the predetermined block of audio values.
3. The device according to claim 2, further comprising means (22) for determining a first noise power limit depending on the first masking threshold and a second noise power limit depending on the second masking threshold, and wherein the means for filtering comprises means (90) for interpolating between the first noise power limit and the second noise power limit to obtain an interpolated noise power limit for a predetermined audio value of the predetermined block of audio values, means (92) for determining an intermediate scaling value depending on a quantizing noise power caused by quantizing according to a predetermined quantizing rule and the interpolated noise power limit, and means (94) for applying the intermediate scaling value to the predetermined audio value to obtain a scaled filtered audio value.
4. The device according to claim 3, wherein the means for interpolating between the first noise power limit and the second noise power limit performs a linear interpolation.
5. The device according to claims 3 or 4, wherein the means for determining the intermediate scaling value comprises means for calculating the root of the quotient of the quantizing noise divided by the interpolated noise power limit.
6. The device according to one of the preceding claims, wherein the means for quantizing is formed to perform quantizing responsive to a control signal.
7. The device according to one of the preceding claims, further comprising lossless compression means for compressing the filtered audio values into a compressed audio stream, wherein the compression means is formed to control a bit rate of the compressed audio stream and to send the control signal to the means for quantizing in the case that the bit rate is greater than a control value.
8. The device according to one of the preceding claims, wherein the quantizing step function has a flat course above the threshold information value such that filtered audio values greater than the threshold information value are quantized to a maximum quantizing step value.
9. A method for quantizing an information signal of a sequence of information values, the information signal being an audio signal and the information values being audio values, comprising the steps of:

   frequency-selective filtering the sequence of audio values to obtain a sequence of filtered audio values;

   quantizing the filtered audio values to obtain a sequence of quantized audio values by means of a quantizing step function which maps the filtered audio values to the quantized audio values and the course of which is steeper below a threshold information value than above the threshold information value;

   determining a listening threshold for a block of audio values; and

   calculating a version of a parameterization of a parameterizable filter such that the transfer function thereof roughly corresponds to the inverse of the magnitude of the first listening threshold,

   wherein the step of frequency-selective filtering further comprises the step of:

   filtering a predetermined block of audio values of the sequence of audio values with the parameterizable filter using a predetermined parameterization which in a predetermined manner depends on the version of the parameterization to obtain a block of the filtered audio values.
10. A computer program having a program code adapted to perform the method according to claim 9 when the computer program runs on a computer.