JP2008026372A - Encoding rule conversion method and device for encoded data - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an encoding rule conversion method and device for converting the encoding rule of data which are encoded with a first encoding rule, to a second encoding rule in a short period of time, without degradation in the quality. <P>SOLUTION: In the encoding rule conversion method of the encoded data, a quantization scale calculating section 300 calculates an AAC(advanced audio coding) scale value Q' from an MP3 quantization scale value Q, on the basis of a primary function indicating correlation of the MP3 quantization scale value and the AAC quantization scale value. A quantization section 311 quantizes MDCT (modified discrete cosine transform) coefficient, on the basis of the AAC quantization scale value Q', which is calculated in the quantization scale calculating section 300. Consequently, repeated processings for determining the quantization scale value are not performed in the AAC encoding process. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to encoding of encoded data for converting audio data encoded with the first encoding rule to audio data encoded with a second encoding rule different from the first encoding rule. The present invention relates to a chemical rule conversion method and apparatus.

  As an audio compression method defined by an international standard, the MP3 (MPEG (Moving Picture Experts Group) -1 Audio Layer 3) standard standardized by ISO is widely used. MP3 is one of the most popular audio compression methods, and many portable playback devices have an MP3 decoder.

  On the other hand, although it is not compatible with MP3, AAC (Advanced Audio Coding) is defined as an encoding method that achieves a higher compression rate while maintaining sound quality, and it is used in several online music distribution and various broadcasting services, etc. It has been adopted. AAC is said to be capable of realizing CD-quality audio at about 48 to 64 kbps per channel and reducing the code amount by about 30% compared to MP3. Therefore, AAC products and services related to AAC will appear in the market in the future, and the demand for converting MP3 files to AAC files is expected to increase.

  The simplest way to convert MP3 to AAC is transcoding in the uncompressed domain, where MP3 is completely decoded to PCM data and then this PCM data is re-encoded to AAC.

  FIG. 5 is a block diagram showing the configuration of the main part of a conventional MP3 decoder 10, which extracts a Huffman code decoding unit 101 for decoding MP3 Huffman code into nonlinear quantized data and side information in the frame. The side information decoding unit 102 for decoding the data, the inverse quantization unit 103 for dequantizing the data based on the side information, the alias reduction unit 104 for reducing aliasing caused by the hybrid filter bank, and the inverse deformation An IMDCT (Inverse Modified Discrete Cosine Transform) unit 105 that performs discrete cosine transform and a Synthesis Subband Filter Bank unit 106 that synthesizes 32 subbands to restore PCM data are the main components. Such an MP3 decoder is disclosed in Patent Document 1.

  FIG. 6 is a block diagram showing a configuration of a main part of a conventional AAC encoder 20, and an input audio signal is blocked (called a frame) every predetermined number of samples and divided into two paths. It is processed.

  The psychoacoustic analysis unit 201 obtains a frequency spectrum and various parameters by performing a fast Fourier transform (FFT) on the input frame. The MDCT (modified DCT) unit 202 converts the input audio signal into a frequency spectrum (hereinafter also referred to as an MDCT coefficient) with the block length determined by the psychoacoustic analysis unit. A TNS (Temporal Noise Shaping) unit 203 concentrates the quantization noise on a portion with a large signal level by changing the noise level accompanying the compression processing according to the volume of the sound, and the sound is In a small part, the noise is reduced and the hearing is improved.

  The backward prediction processing unit 204 performs prediction filtering on the MDCT coefficient. The nonlinear quantization unit 205 quantizes the MDCT coefficient with the goal of being below the allowable quantization noise power for each scale factor band obtained by the psychoacoustic analysis unit. The quantized MDCT coefficient is further subjected to Huffman coding by the Huffman coding unit 206 to reduce redundancy. This quantization / Huffman encoding process is performed in an iterative loop, and is repeated until the amount of code actually generated falls below the number of bits allocated to the frame. Such an AAC encoder is disclosed in Patent Document 2.

In Patent Document 3, the audio data (MP3) of the first coding rule that is difficult to edit is converted into the audio data (ATRAC) of the second coding rule that is easy to edit. A technique is disclosed in which editing processing is performed on audio data with the second encoding rule, and the data is restored to audio data with the first encoding rule after editing.
JP 2003-99095 A JP 2006-145882 A JP 2001-184840 A

  In the coding rule conversion of audio data, in the transcoding scheme in which the data to be converted is re-encoded after complete decoding, all decoding processes and encoding processes are executed, especially when all the encoding processes are executed, Since it takes a long time in the iterative loop of quantization and Huffman coding, there is a technical problem that the conversion time becomes long. In addition, since two types of encoding processing are performed on the same data, there is a technical problem that quality deteriorates.

  The object of the present invention is to solve the above-mentioned problems of the prior art and to convert the data encoded by the first encoding rule into the second encoding rule in a short time while minimizing quality degradation. It is an object of the present invention to provide a coding rule conversion method and apparatus that can convert the data into a code.

In order to achieve the above-described object, the present invention encodes the first coding rule data encoded with the first coding rule with the second coding rule in which the quantization at the time of encoding is repeated in an iterative loop. An encoded data conversion method for converting encoded data into converted second encoded rule data includes the following procedure.
(1) A decoding process for decoding the first coding rule data into the PCM data, a decoding process for coding the PCM data into the second coding rule data, and a decoding process for the first coding rule data In the inverse quantization procedure, in the inverse quantization procedure, a first inverse quantization scale value is obtained for each sample, and a predetermined function calculation is performed on each first quantization scale value. And calculating a second quantization scale value and, in the encoding process of the second encoding rule, a procedure of quantizing data using each of the second quantization scale values. .
(2) The first coding rule data is encoded by the number of frames corresponding to the least common multiple of the number of samples of one frame of the first coding rule data and the number of samples of one frame of the second coding rule data. It is characterized in that the chemical rule is converted.
(3) The encoding process of the second encoding rule data includes a DCT procedure for converting the time domain of the data into the frequency domain, and each frame in the inverse quantization procedure in the decoding process of the first encoding rule data And a procedure for determining a window size based on the stored frame structure in the DCT procedure in the encoding process of the second code rule data.

According to the present invention, the following effects are achieved.
(1) The first encoding rule data encoded by the first encoding rule is converted into the second encoding rule data encoded by the second encoding rule in which the quantization at the time of encoding is repeated in an iterative loop. At the time of conversion, the parameter relating to the quantization scale obtained in the decoding process of the first coding rule data can be inherited by the coding process in the second coding rule, so that the most in the coding process of the second coding rule. It is possible to simplify a time-consuming repetitive process.
(2) Since the coding rule conversion is performed by the number of frames corresponding to the least common multiple of the number of samples of one frame of the first coding rule data and the number of samples of one frame of the second coding rule data. Thus, the difference between the frame size of the first coding rule data and the frame size of the second coding rule data can be resolved.
(3) The first encoding rule data encoded by the first encoding rule is converted into the second encoding rule data encoded by the second encoding rule in which the quantization at the time of encoding is repeated in an iterative loop. At the time of conversion, parameters related to the frame structure obtained in the decoding process of the first coding rule data can be inherited by the coding process in the second coding rule, which is optimal in the coding process of the second coding rule. The frame structure can be selected.

  DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the best embodiment of the present invention will be described in detail with reference to the drawings. Here, taking the conversion from MP3 to AAC as an example, the outline of the present invention will be described first, and then the details will be described.

  In the present invention, the frame structure obtained by the decoding process of MP3 and the above-described technical problem of the transcoding scheme are simplified by simplifying the iterative process of quantization that takes the most time in the encoding process of AAC. The parameters related to the quantization scale are inherited by the AAC encoding process.

  In MP3 decoder, the frequency component is changed to 32 subband signals by applying 6-point (short) or 18-point (long) MDCT (modified DCT) while overlapping the frames by half in each subband. Divided. The 6-point MDCT is applied to a portion that changes abruptly in time, particularly where pre-echo occurs.

  On the other hand, AAC has been studied to achieve higher quality compression coding than MP3, and many techniques used in AAC have already been introduced in MP3. One of the reasons for improved sound quality over MP3 in AAC is the introduction of the MDCT filter bank. In AAC, the filter bank adaptively fluctuates between 128 and 1024 points (half windows overlap with 256 and 2048 points).

  A filter bank with high frequency resolution is often required for audio signals, but 1024 filter banks can be applied with AAC, but up to 576 with MP3. However, since a high frequency resolution is not required in the case of a transient signal, it is possible to adaptively set the window length shorter. In this case, the frequency resolution can be set at 192 points in MP3, but 128 points in AAC. Thus, one of the differences between MP3 and AAC is the difference in the method of conversion to the frequency domain. Furthermore, each frame size for conversion is also different.

  In general, in the encoding process of MP3 and AAC, code bits are assigned using the same psychoacoustic model. Furthermore, since code assignment is typically an iterative operation, the time required for this takes up most of the encoding process. Therefore, reducing the time for this process greatly reduces the time for the entire audio encoding.

  In the present invention, the time required for coding rule conversion can be shortened by calculating the AAC parameters determined by the iterative calculation from the MP3 parameters.

  The inverse quantization coefficient of MP3 is a global gain (quantization step) and a scale factor (for each subband signal in one frame, the value of the sample having the maximum absolute value is converted into a logarithm and quantized). It is expressed as the sum of

  Here, xr [i] is a quantized MDCT coefficient, is [i] is 576 data (MDCT coefficients) obtained by decoding a Huffman code, Scalefactor is a scale factor applied for each scale factor band, Global_Gain , Subblock_Gain, Scalefactor_Scale are values obtained from the granule information, and i, gr, w, and cb indicate the MDCT coefficient index, the granule index, the window index, and the codebook index, respectively. On the other hand, the requantization coefficient of AAC is also calculated as follows using the scale factor.

  Here, g and sfb mean window group and scale factor band, respectively. In the above formulas (1) and (2), the distribution of the quantization scale value indicated by the exponent part of 2 is expressed as shown in FIG. 3, and the quantization scale values Q and Q ′ of MP3 and AAC are absolute. Although the values are different, it can be seen that they are highly correlated.

  In the present invention, paying attention to the correlation between the quantization scale values Q and Q ′, using the linear function representing the correlation, the quantization scale value Q of the MP3 can be used for the AAC quantum without repetition processing. In particular, the time of the AAC encoding process is shortened by obtaining the conversion scale value Q ′.

  FIG. 1 is a block diagram showing the configuration of the main part of a coding rule conversion apparatus 1 according to the present invention.

  The Huffman code decoding unit 301 decodes the MP3 Huffman code into nonlinear quantized data. The side information decoding unit 302 takes out the side information in the frame and decodes it. The inverse quantization unit 303 performs inverse quantization on the data based on the side information. The alias reduction unit 304 reduces aliasing (folding distortion) caused by the hybrid filter bank.

  The IMDCT unit 305 performs inverse deformation discrete cosine transform. A subband filter bank (FB) unit 306 combines the 32 subbands and reproduces PCM data. The quantization scale calculation unit 300 calculates the AAC quantization scale value Q from the MP3 quantization scale value Q based on the linear function indicating the correlation between the MP3 quantization scale value and the AAC quantization scale value described with reference to FIG. 'Is calculated.

  The psychoacoustic analysis unit 307 obtains a frequency spectrum by performing FFT on the input frame in the psychoacoustic analysis unit. The MDCT unit 308 converts the input audio signal into a frequency spectrum (MDCT coefficient) with the block length determined by the psychoacoustic analysis unit 307. The TNS unit 309 concentrates the quantization noise on the part where the signal level is high by changing the level of the noise accompanying the compression process according to the volume of the sound, and reduces the noise at the part where the sound is low, thereby improving the audibility. Improve.

  The backward prediction processing unit 310 performs linear prediction on the assumption that the MDCT coefficient is a signal on the time axis, and performs prediction filtering on the MDCT coefficient. The quantization unit 311 quantizes the MDCT coefficient based on the AAC quantization scale value Q ′ calculated by the quantization scale calculation unit 300. The quantized MDCT coefficient is Huffman encoded by the Huffman encoder 312 to reduce the redundancy.

  Next, a procedure for converting MP3 encoded data into AAC encoded data will be described in detail with reference to the flowchart of FIG.

  In step S1, MP3 frames corresponding to the least common multiple of the number of samples of one frame of MP3 and the number of samples of one frame of AAC are accumulated in the buffer. Since one frame of MP3 is 1152 samples and one frame of AAC is 1024 samples, here, 9216 samples which are the least common multiple of them are used as conversion units, and samples for 8 (= 9216/1152) frames are stored in the buffer. Accumulated. As a result, the problem of the difference in frame size between MP3 and AAC can be absorbed.

  In step S2, the MP3 data in the buffer is decoded by the Huffman code decoding unit 301 and the side information decoding unit 302, and 576 data of granule information, scale factor, and quantized samples are obtained. The granule information includes parameters such as global gain, sub-block gain, and scale factor scale.

  In step S3, the inverse quantization unit 303 receives the above-mentioned granule information, scale factor, and 576 data of the quantized samples as input, and performs inverse quantization based on the above equation (1). MDCT coefficient xr [i] is calculated. In step S4, the quantization scale value obtained in the process of calculating the quantized MDCT coefficient xr [i] is stored as a variable Q [i] (i is a sample number). At this time, the frame structure (long or short) of each sample is also stored.

  In step S5, alias processing is performed on the quantized MDCT coefficients to reduce aliasing distortion. In step S6, IMDCT synthesis processing is performed, and the frequency domain data is converted into time domain data. At this time, for a frame having a long block structure, 576 data is converted to a 32 × 18 data structure, and for a short block frame, 192 data is converted to a 32 × 6 data structure. In step S7, the subbands divided into 32 are combined to restore PCM data.

  In step S8, the psychoacoustic analysis unit 307 performs FFT on the PCM data by a predetermined number of frames to obtain the frequency spectrum. Further, auditory masking is calculated based on the frequency spectrum, and allowable quantization noise power for each preset frequency band and psychoacoustic entropy PE for the frame are obtained.

  In step S9, the MDCT unit 308 converts the input audio signal into a frequency spectrum (MDCT coefficient) with the block length determined by the psychoacoustic analysis unit 307. At this time, the frame structure stored in step S4 is referred to, and as shown in an example in FIG. 4, a short window (SW) is formed in a frame that largely overlaps a frame that was a short block (SB) in MP3. The long window (LW) is selected in other frames.

  In step S10, the TNS unit 309 performs linear prediction on the assumption that the MDCT coefficient is a signal on the time axis, and performs prediction filtering on the MDCT coefficient. In step S11, the backward prediction processing unit 310 performs backward prediction processing for predicting the current MDCT coefficient value from the quantized MDCT coefficients in the past two frames for each MDCT coefficient, and compares it with the immediately preceding data. The amount of data is reduced by taking the difference. In step S12, based on the quantization scale value Q [i] inherited from MP3, the quantization scale calculation unit 300 calculates the AAC quantization scale value Q '[i].

  In step S13, the code amount is determined based on the quantization scale value Q '[i]. In step S14, the quantization unit 205 performs quantization on the prediction residual. That is, in this embodiment, iterative processing for determining the quantization scale value is not performed.

  In step S15, the global gain is finely corrected according to the psychoacoustic model. In step S16, the Huffman code is applied to the quantized MDCT coefficient to reduce the redundancy. In step S17, it is determined whether or not unprocessed MP3 data remains, and the process returns to step S1 and the above-described processes are repeated until the conversion process for all the MP3 data is completed.

  In the above-described embodiment, the present invention has been described by taking the conversion from MP3 to AAC as an example. However, the present invention is not limited to this, and data encoded by the first encoding rule. Is converted into data encoded by the second encoding rule, if the parameters obtained by the decoding process of the first encoding rule can be used in the encoding process by the second encoding rule, The same applies to conversion between coding rules.

It is the block diagram which showed the structure of the principal part of the encoding rule converter based on this invention. It is the flowchart which showed the procedure which converts MP3 data into AAC data. It is the figure which showed the correlation of each quantization scale value of MP3 and AAC. FIG. 3 is a diagram illustrating a relationship between an MP3 frame structure and an AAC frame structure. It is the block diagram which showed the structure of the conventional MP3 decoder. FIG. 10 is a block diagram showing a configuration of a conventional AAC encoder.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Coding rule converter, 10 ... MP3 decoder, 20 ... AAC encoder, 300 ... Quantization scale calculation part, 301 ... Huffman code decoding part, 302 ... Side information decoding part, 303 ... Dequantization part, 304 ... Alias Reduction unit, 305... IMDCT unit, 306... Synthesis Subband Filter Bank unit, 307... Psychoacoustic analysis unit, 308... MDCT unit, 309. Chemical department

Claims (8)

A code for converting the first coding rule data encoded by the first coding rule into the second coding rule data encoded by the second coding rule in which quantization at the time of encoding is repeated in an iterative loop. In the coding rule conversion method for coded data,
A decoding process for decoding the first coding rule data into PCM data and a coding process for coding the PCM data into second coding rule data;
A procedure of dequantizing the quantized data in the decoding process of the first code rule data;
In the inverse quantization procedure, a procedure for obtaining a first inverse quantization scale value for each sample;
Performing a predetermined function calculation on each first quantization scale value to calculate a second quantization scale value;
A process of quantizing data using the second quantization scale values in an encoding process of a second encoding rule,
The predetermined function is configured to convert the first quantization scale value to the second quantization scale based on an additive relationship between the quantization scale value of the first coding rule and the quantization scale value of the second coding rule for the same sample. A coding rule conversion method for encoded data, characterized by converting into a value.   The first coding rule data has its coding rule for each frame number corresponding to the least common multiple of the number of samples of one frame of the first coding rule data and the number of samples of one frame of the second coding rule data. The encoding rule conversion method for encoded data according to claim 1, wherein conversion is performed.   3. The coding rule conversion method for coded data according to claim 1, wherein the first coding rule is MP3 and the second coding rule is AAC. 4. The encoding process of the second encoding rule data includes a DCT procedure for converting the time domain of the data to the frequency domain;
In the inverse quantization procedure in the decoding process of the first code rule data, a procedure for storing a frame structure of each frame;
4. The encoding of encoded data according to claim 3, wherein a DCT procedure in the encoding process of the second encoding rule data includes a procedure for determining a window size based on the stored frame structure. 5. Law conversion method. A code for converting the first coding rule data encoded by the first coding rule into the second coding rule data encoded by the second coding rule in which quantization at the time of encoding is repeated in an iterative loop. In a coding data conversion device for coded data,
Means for dequantizing the quantized data of the first code rule;
Means for performing a predetermined function calculation on the first inverse quantization scale value obtained at the time of the inverse quantization to calculate a second quantization scale value;
Means for quantizing data to be encoded according to a second encoding rule using each of the second quantization scale values,
The predetermined function is configured to convert the first quantization scale value to the second quantization scale based on an additive relationship between the quantization scale value of the first coding rule and the quantization scale value of the second coding rule for the same sample. An encoding rule conversion apparatus for encoded data, characterized by converting into a value. Buffer means for storing the first coding rule data by the number of frames corresponding to the least common multiple of the number of samples of one frame of the first coding rule data and the number of samples of one frame of the second coding rule data. ,
The coding rule conversion device for coded data, wherein the coding rule conversion is executed in units of frames stored in the buffer.   7. The encoded rule conversion apparatus for encoded data according to claim 5, wherein the first encoding rule is MP3 and the second encoding rule is AAC. DCT means for converting the time domain of the data to be encoded by the second encoding rule into the frequency domain,
8. The coding rule conversion apparatus for coded data according to claim 7, wherein the DCT means determines a window size of each frame based on a frame structure of each frame of the first coding rule data.

Images