JP4550595B2 - Audio encoding device - Google Patents

Description

  The present invention involves processing for obtaining spectral data by performing time-frequency conversion of audio signals in units of sub-blocks including a plurality of audio signals in one frame, and grouping the sub-blocks included in one frame into the same group. The present invention relates to an audio encoding device for sharing side information among a plurality of sub-blocks.

  Non-Patent Document 1 stipulates encoding by AAC (Advanced Audio Coding) as follows.

  A frame including a transient signal such as an attack signal is processed as a short block. A frame determined to be a short block is divided into 8 sub-blocks and processed for each sub-block.

  However, not all eight sub-blocks contain transient signals. Therefore, grouping is performed so that sub-blocks that do not include transient signals are collected. The compression rate is increased by sharing side information such as a scale factor within the group.

By the way, the sub-block is obtained by dividing the audio signal in the time domain by each of eight windows W0 to W7 as shown in FIG. 3, and individually performing MDCT (Modified Discrete Cosine Transform) on the signals in each section. The position of the transient signal is detected in the audio signal in the time domain. Then, the grouping is performed in consideration of which of the periods P0 to P7 shown in FIG. Specifically, when the position of the transient signal is the position shown in FIG. 3, the position belongs to the period P2, so that a group including only the sub-block corresponding to the window W2 is formed and the other sub-blocks are collected. Group them as follows.
"3rd Generation Partnership Project; Technical Specification Group Service and System Aspects; General Audio Codec audio processing functions; Enhanced aacPlus general audio codec; Encoder Specification AAC part (Release 6)", 3GPP TS 26.403 V6.0.0, 3rd Generation Partnership Project, 2004 September

  As shown in FIG. 3, in the windows W0 to W7, adjacent ones overlap each other by 50%. For this reason, the influence of the transient signal appears in each of the two sub-blocks. In the example of FIG. 3, the influence of the transient signal appears in each of the second sub-block corresponding to the window W2 and the third sub-block corresponding to the window W3. Then, at the position of the transient signal in FIG. 3, the window function of the window W3 is larger than the window function of the window W2, so that the influence of the transient signal appears more in the third subblock than in the second subblock. .

  However, as described above, according to the method defined in Non-Patent Document 1, a group including only the second sub-block is created and grouped so as to collect other sub-blocks, and is based on proper grouping. Encoding was not always possible.

  The present invention has been made in consideration of such circumstances, and an object of the present invention is to perform audio encoding that can improve the compression ratio by properly grouping the signal by correctly reflecting the influence of a transient signal. To provide an apparatus.

  The audio encoding apparatus according to the first aspect of the present invention includes a process of obtaining spectral data by performing time-frequency conversion on an audio signal in units of sub-blocks constituting one frame, and includes the sub signal included in one frame. In an audio encoding device that encodes the audio signal using an audio encoding method for grouping blocks to share side information in the plurality of sub-blocks included in the same group, based on the spectral data, Calculating means for calculating the flatness of a spectrum for each sub-block, selecting means for selecting, as a single block, the one with the highest flatness calculated from the sub-blocks included in the one frame; The sub-block included in one frame is grouped with the single block. Comprising a flop, and a grouping means for performing the grouping to make a group including a consecutive at least two sub-blocks of said sub-blocks other than the single block.

  The audio encoding device according to the second aspect of the present invention includes a process of obtaining spectral data by performing time-frequency conversion on an audio signal in units of sub-blocks constituting one frame, and includes the sub-signals included in one frame. In an audio encoding device that encodes the audio signal using an audio encoding method for grouping blocks to share side information in the plurality of sub-blocks included in the same group, based on the spectral data, Calculating means for calculating the flatness of a spectrum for each sub-block, selecting means for selecting a sub-block whose flatness calculated by the calculating means exceeds a predetermined threshold as a single block, and each of the single blocks The independent group is included in the one frame. Making a group comprising at least two sub-blocks consecutive one of said sub-blocks other than German block and a grouping means.

  According to the present invention, it is possible to improve the compression ratio by properly reflecting the influence of a transient signal and appropriately performing grouping.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram of an audio encoding apparatus (hereinafter referred to as an encoding apparatus) according to this embodiment.

  This encoding apparatus performs an encoding process on an input PCM signal and outputs an encoded bit stream. This encoding apparatus includes a block cutout unit 1, a psychoacoustic model unit 2, a filter bank unit 3, a switch 4, a flatness measure calculation unit 5, a grouping unit 6, a quantization distortion / rate control unit 7, a host processor 8, and scaling. A section 9, a quantization section 10, an encoding section 11, and a formatter 12. Each of these units can be configured by hardware, and the function of each unit can be realized by software processing using a DSP (Digital Signal Processor) or the like.

  The block cutout unit 1 is given a PCM signal to be encoded. This PCM signal is a time domain signal. The block cutout unit 1 cuts out data from the PCM signal for each sample number of a specified block size. The block cutout unit 1 outputs the cutout signal.

  The psychoacoustic model unit 2 performs orthogonal transform such as DFT (Discrete Fourier Transform), DCT (Discrete Cosine Transform), or MDCT (Modified DCT) on the signal output from the block cutout unit 1, thereby Convert signal to frequency domain signal. The psychoacoustic model unit 2 calculates a parameter called perceptual entropy from the transform coefficient (frequency component) obtained by the orthogonal transform. This parameter is information required to encode one block by analyzing the auditory frequency resolution, diffusion of frequency components, unpredictability, and signal tonality from the conversion coefficient. This is for estimating the quantity, and details of the calculation method are defined in the international standard 13818-7 established by ISO / IEC.

  The psychoacoustic model unit 2 includes a block switching unit 2a and an SMR calculation unit 2b. Based on the perceptual entropy, the block switching unit 2a determines a block length (long block, short block) used in orthogonal transform processing (MDCT processing) when actually encoding. The block switching unit 2a outputs block length information indicating the result of this determination to the filter bank unit 3 and the switch 4. The SMR calculation unit 2b is a permissible noise that is not perceived even if it exists as a sound masked with respect to a signal, that is, noise, for each equally-spaced band on a scale (bark, mel, etc.) in consideration of auditory frequency resolution. The SMR (Signal to Mask Ratio) indicating the ratio of the amount of is calculated. The SMR calculation unit 2 b outputs the calculated SMR to the quantization distortion / rate control unit 7.

  The filter bank unit 3 orthogonally transforms the output signal of the block cutout unit 1 according to the block length information output from the psychoacoustic model unit 2. The filter bank unit 3 outputs the spectral data obtained by the orthogonal transformation. In addition, when AAC is employ | adopted as an encoding system, the orthogonal transformation in the filter bank part 3 becomes MDCT.

  The switch 4 selects any one of the flatness measure calculation unit 5, the grouping unit 6, and the scaling unit 9 based on the block length information. The switch 4 gives the output signal of the filter bank unit 3 to the selected side.

  The flatness measure calculation unit 5 calculates the spectral flatness measure in the output signal of the filter bank unit 3 for each sub-block. The grouping unit 6 groups the output signals of the filter bank unit 3 based on the spectrum flatness measure calculated by the flatness measure calculation unit 5.

  The quantization distortion / rate control unit 7 calculates a code amount that can be allocated for each frame based on the coding rate instructed by the host processor 8 and the SMR output from the psychoacoustic model unit 2. The quantization distortion / rate control unit 7 controls the scaling unit 9, the quantization unit 10, and the encoding unit 11 using the calculated code amount as the target code amount of the encoded frame. For example, the quantization distortion / rate control unit 7 calculates a quantization distortion amount from the quantization coefficient given from the quantization unit 10 and gives an output instruction to the quantization unit 10 according to the result. Further, the quantization distortion / rate control unit 7 checks whether or not the code amount notified from the encoding unit 11 is within the target code amount, and when the code amount is within the target code amount, the quantization distortion / rate control unit 7 proceeds to the encoding unit 11. Give output instructions.

  The scaling unit 9 determines a scale factor based on the spectral data output from the filter bank unit 3 or the grouping unit 6. The scaling unit 9 scales the spectral data output from the filter bank unit 3 using the determined scale factor. Then, the scaling unit 9 outputs the spectral data after scaling to the quantization unit 10. The scaling unit 9 outputs a scaling coefficient to the formatter 12 in response to an instruction from the quantization distortion / rate control unit 7.

  The quantizing unit 10 corrects the spectral data output from the scaling unit 9 according to a prescribed formula, and then quantizes all the spectral data. The quantization unit 10 outputs the quantized data to the quantization distortion / rate control unit 7 as information for determining whether the quantization distortion error is an allowable error based on the SMR value. The quantization unit 10 outputs the quantized data to the encoding unit 11 in response to an output instruction from the quantization distortion / rate control unit 7.

  The encoding unit 11 compresses and encodes the output of the quantization unit 10 according to a predetermined encoding method. For example, in the case of AAC, the Huffman encoding method is applied as the above-described encoding method. The encoding unit 11 outputs the encoded code amount to the quantization distortion / rate control unit 7. The encoding unit 11 outputs the encoded data to the formatter 12 in response to an output instruction from the quantization distortion / rate control unit 7.

  The formatter 12 multiplexes the output of the encoding unit 11 and the scaling coefficient output from the scaling unit 9 according to a predetermined format. The formatter 12 outputs the result of the above multiplexing as an encoded audio signal.

  When signals grouped by the grouping unit 6 are processed, side information such as a scale factor is shared in the processing in the scaling unit 9, the quantization unit 10, the encoding unit 11, and the formatter 12.

  Next, the operation of the encoding apparatus configured as described above will be described. The point of the present invention lies in the grouping process when it is determined that the block is a short block. Therefore, here, this process will be mainly described. For other processing, processing similar to that of an existing encoding device that conforms to AAC can be applied.

  Based on the perceptual entropy, the block switching unit 2a determines a frame including a transient signal such as an attack signal as a short block and determines other frames as a long block.

  When it is determined that the block is a short block, the switch 4 selects the flatness measure calculation unit 5 and the grouping unit 6.

  On the other hand, when it is determined that the block is a short block, the filter bank unit 3 divides the time domain audio signal by each of eight windows W0 to W7 as shown in FIG. , Spectral data of 8 sub-blocks from No. 0 to No. 7 are sequentially obtained. The spectral data of the eight sub blocks are sequentially input to the flatness measure calculation unit 5 and the grouping unit 6.

The flatness measure calculation unit 5 calculates the flatness of the input spectral data for each sub-block. As the flatness of the spectral data, a spectral flatness measure calculated by the following equation (1) can be used.

Ma and Mg in the above equation (1) are the arithmetic mean and geometric mean within the sub-block for the signal strength or power value for each sample, and the following equations (2) and (3) Is calculated by

Here, k is a sub-block number, n is the number of sub-block samples, and spec (i) is the signal strength or power value for each sample.
That is, the spectral flatness measure is an arithmetic geometric average within the sub-block for the signal strength or power value for each sample.

  FIG. 2 is a diagram showing an operation flow in the grouping unit 6 in FIG. The grouping unit 6 performs the processing shown in FIG. 2 for each frame.

In step Sa1, the grouping unit 6 collects the spectral flatness measures sfm (0) to sfm (7) calculated by the flatness measure calculation unit 5 for each sub-block as described above. In step Sa2, the grouping unit 6 determines the maximum value of these spectral flatness measures sfm (0) to sfm (7). In step Sa3, the grouping unit 6 substitutes the number of the spectral flatness measure sfm, which is the maximum value, into the variable _kmin . That is, for example, if the spectral flatness measure sfm (3) is the maximum value, “3” is substituted into the variable _kmin .

In step Sa4, the grouping unit 6 checks whether the value of the variable _kmin is “0”, “1” to “6”, or “7”.

If the value of the variable _kmin is “0”, the grouping unit 6 proceeds from step Sa4 to step Sa5. In step Sa5, the grouping unit 6 groups the 0th sub-block as the first group and the 1-7th sub-blocks as the second group.

If the value of the variable _kmin is “1” to “6”, the grouping unit 6 proceeds from step Sa4 to step Sa6. Grouping unit 6 at step Sa6 is a k _min -1 th subblock 0 th first group, the sub-blocks of the second group of k _min th, the sub-blocks of k _min +1 th to 7 th as the third group Group.

If the value of the variable _kmin is “7”, the grouping unit 6 proceeds from step Sa4 to step Sa7. In step Sa7, the grouping unit 6 groups the 0th to 6th sub-blocks as the first group and the 7th sub-block as the second group.

  That is, the grouping unit 6 separates sub-blocks having the flatest spectrum shape as one block, and collects other consecutive sub-blocks as one group. For example, if the spectral shape of the spectral data in the third sub-block is the flattest, the grouping unit 6 assigns each sub-block to {0,1,2}, {3}, {4,5,6,7}. Group them as follows.

  A plurality of sub-blocks grouped in the same group in this way share side information such as a scale factor.

  The shape of the spectrum becomes flatter as the signal is transient. For this reason, the sub-block having the flatst spectral shape, that is, the sub-block having the largest spectral flatness measure is the sub-block in which the influence of the transient signal appears most. Thus, by the grouping of the present embodiment as described above, it is possible to properly perform the grouping by correctly reflecting the influence of the transient signal. As a result, since the side information is properly shared, the compression rate can be improved.

  For example, in the example shown in FIG. 3, the influence of the transient signal has a greater effect on the MDCT processing in the window W3 than in the MDCT processing in the window W2, and the spectrum shape of the third sub-block corresponding to the window W3 becomes flat. As a result, appropriate grouping as in the above specific example is performed.

  Note that the spectral flatness measure may be calculated for processing different from the grouping. In this case, the grouping can be realized by a simpler process by using the spectral flatness measure calculated for such another process for the grouping.

This embodiment can be variously modified as follows.
When the second or third largest spectral flatness measure exceeds a threshold, the sub-blocks for which the spectral flatness measure is calculated may be grouped so as to be independent.

  Even if the frame can be determined to be a long block based on the perceptual entropy, a method of converting to a short block may be adopted if both the preceding and succeeding frames are determined to be short blocks. In this case, since a frame converted from a long block to a short block does not include a transient signal, it is preferable to group all eight sub-blocks as one group without performing grouping. In this way, the side information can be shared by all the side blocks, and the compression rate is improved. In order to cope with the occurrence of a short block frame that does not include a transient signal as described above, for example, a method of monitoring the result of conversion processing from a long block to a short block, or spectral flatness A method of monitoring whether the maximum value of the measure exceeds a threshold value can be considered.

  If the number of sub-blocks included in one group should be limited, grouping may be performed so as to create more groups. For example, if the number of subblocks to be included in one group should be limited to three, in the specific example shown in the above embodiment, each subblock is {0,1,2}, {3}, {4,5, Groupings such as 6,7} are grouped as {0,1,2}, {3}, {4,5,6}, {7}.

  Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment.

The block diagram of the audio coding apparatus which concerns on this embodiment of this invention. The figure which shows the operation | movement flow in the grouping part 6 in FIG. The figure explaining the determination method of the position of the transient signal in the past.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Block cutout part, 2 ... Psychological auditory model part, 3 ... Filter bank part, 4 ... Switch, 5 ... Flatness measure calculation part, 6 ... Grouping part, 7 ... Quantization distortion / rate control part, 8 ... Host processor , 9: scaling unit, 10: quantization unit, 11: encoding unit, 12: formatter.

Claims (5)

The audio signal is subjected to time-frequency conversion in units of sub-blocks constituting one frame to obtain spectral data, and the sub-blocks included in one frame are grouped to form a plurality of the sub-blocks included in the same group. In an audio encoding device that encodes the audio signal using an audio encoding method in which side information is shared in a block,
Based on the spectral data, calculation means for calculating the flatness of the spectrum for each sub-block,
Selecting means for selecting, as a single block, the one with the highest flatness calculated from the sub-blocks included in the one frame;
Grouping for grouping the sub-blocks included in the one frame so as to form a group of the single block and a group including at least two consecutive sub-blocks of the sub-blocks other than the single block. And an audio encoding device. The audio signal is subjected to time-frequency conversion in units of sub-blocks constituting one frame to obtain spectral data, and the sub-blocks included in one frame are grouped to form a plurality of the sub-blocks included in the same group. In an audio encoding device that encodes the audio signal using an audio encoding method in which side information is shared in a block,
Based on the spectral data, calculation means for calculating the flatness of the spectrum for each sub-block,
Selecting means for selecting, as a single block, a sub-block whose flatness calculated by the calculating means exceeds a predetermined threshold;
Grouping means for making each of the single blocks an independent group and creating a group including at least two consecutive subblocks of the subblocks other than the single block included in the one frame. An audio encoding device.   3. The audio code according to claim 1, wherein the grouping unit performs the grouping so as to create a group including all of the continuous sub-blocks among the sub-blocks other than the single block. 4. Device.   The grouping means is a frame in which the block length used in the orthogonal transformation process is determined to be a long block based on perceptual entropy, and the frame converted to the short block according to the preceding and following frames is the one frame The audio encoding apparatus according to claim 1 or 2, wherein a group including all sub-blocks included therein is created.   When the maximum flatness of the flatness of the spectrum for each sub-block calculated by the calculating means is less than a predetermined threshold, the grouping means is a group consisting of all sub-blocks included in the one frame. The audio encoding device according to claim 1 or 2, wherein:

Images