Digital data decoding device

The present invention relates to a digital data decoding device that decodes digital data such as compressed digital audio data and digital video data recorded on a recording medium such as a minidisc (MD) or a flash memory.

However, since some of the information in the compressed digital audio data is deleted when it is encoded by the compression coding technology, it is supplied to a full-scale audio reproduction device capable of faithfully reproducing music sounds, voices, and the like. In some cases, people often feel unsatisfied or uncomfortable. This is due to the following reasons. That is, in the compression coding technology, when coding, the information of the spectrum which is considered to be difficult to hear or distinguish by the human ear due to the auditory psychological characteristics is deleted. However, in the above-mentioned full-scale audio reproduction device, even the information of such a spectrum can be faithfully reproduced. Therefore, when the above-mentioned compressed digital audio data is reproduced, the information that should be originally obtained is obtained. This is because it will not be played.

Therefore, recently, in order to solve the above problem, a technique has been proposed in which information deleted at the time of compression coding is pseudo-interpolated when the compressed digital audio data is decompressed and decoded. For example, some conventional digital audio data decoding devices include a nuclear decoding means and an extended decoding means. The nuclear decoding means decodes the input coded sequence to generate the first frequency spectrum information. Based on the first frequency spectrum information, the extended decoding means is equivalent to the tuning structure indicated by the first frequency spectrum information extended on the frequency axis in the frequency band not represented by the coded sequence. Generate second frequency spectrum information showing the structure (see, for example, Patent Document 1).<patcit num="1"><text>JP-A-2003-108197 (Claim 1, [0009] to [0019], FIGS. 1 to 5)</text></patcit>

<p> In the conventional digital audio data decoding device described above, the wave tuning structure, which is a relatively common property of audio signals, is extracted within the band represented by the coded sequence, and extended spectrum information is added to the high frequency range. (Refer to paragraph [0017] of Patent Document 1 above), or assuming that the energy distribution of the frequency spectrum information can be expressed by a chord function in the tuning period T, the low frequency range between one tuning period T The frequency spectrum information is repeatedly copied or amplified in the high frequency range as it is (see paragraph [0019] of Patent Document 1 above).</p><p> However, the wave-tuning structure included in the decoded low-frequency frequency spectrum information of a certain coded sequence is not always the same as the actual high-frequency tuning structure deleted during coding. Also, the assumption that the energy distribution of frequency spectrum information can be represented by a cosine function in the tuning period T does not always apply. On the other hand, even if this assumption is valid, it cannot be said that the actual high frequencies deleted during coding could be restored simply by repeatedly copying and amplifying the low frequency spectrum information to the high frequencies. Therefore, the interpolation is incomplete, and the above-mentioned unsatisfactory or uncomfortable feeling may not be resolved. Further, in the above-mentioned conventional digital audio data decoding device, the number of quantization bits is non-zero even if a component close to the original sound can be added to the high frequency band completely deleted during compression coding. There is a problem that the quantization noise of the band cannot be removed. The inconvenience described above is that the analog video signal is compressed and encoded into digital video data based on visual and psychological characteristics, recorded on a recording medium, and compressed and encoded digital video data read from the recording medium. The same applies to the case of decompressing and decoding.</p><p> The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a digital data decoding apparatus capable of solving the above-mentioned problems.</p>

<p> In order to solve the above problem, in the invention according to claim 1, frames having a plurality of signals existing within a predetermined time are sequentially input in a time series, further divided into a plurality of frequency bands, and set for each frequency band. It is related to the digital data decoding device that decodes the spectral data that is quantized and encoded for each frequency band based on the index to be calculated, and among the inversely quantized spectral data belonging to a certain frame, the quantization bit is assigned. The spectral data that exists in the frequency band with a small value or the spectral data that should exist in the frequency band with zero quantization bit allocation is the same as either one or both of the frames before and after the certain frame. It is characterized by including an interpolation processing unit that corrects or interpolates using the spectrum data existing in the frequency band.</p><p> The invention according to claim 2 relates to the digital data decoding apparatus according to claim 1, wherein the interpolation processing unit has the same frequency of either one or both of frames before and after the certain frame. A synthetic masking threshold created by synthesizing a masking threshold calculated using the spectral data existing in the band and a minimum audible limit characteristic, and a quantization noise power or energy of the spectral data obtained by the interpolation. The coefficient to be multiplied by the spectral data obtained by the interpolation is determined so that the ratio with and is larger than the ratio in the band in which the quantization bit allocation is large, and the spectral data obtained by the interpolation is multiplied by each. It is characterized by doing.</p><p> The invention according to claim 3 relates to the digital data decoding apparatus according to claim 1, and the interpolation processing unit is out of the range in which the spectrum data obtained by the correction is determined by the index and the number of quantization bits. When present in, the value of the spectrum data obtained by the correction is re-corrected to a value within the range based on the index and the number of quantization bits.</p><p> The invention according to claim 4 relates to the digital data decoding apparatus according to any one of claims 1 to 3, wherein the interpolation processing unit is a quantum of inversely quantized spectral data belonging to a certain frame. In the frequency band where the conversion bit allocation is small, the spectrum data rounded to 0 is the spectrum data obtained by the interpolation, the spectrum data obtained by the interpolation and multiplication of the coefficients, or the correction and re-correction of the value. It is characterized in that any of the spectrum data obtained in the above and the spectrum data existing in the same frequency band of either one or both of the frames before and after the certain frame are used for interpolation.</p>

<p> According to the present invention, when decoding compressed digital data, it is possible to increase the resolution of spectra with low quantization resolutions such as binary and quadrature, and the spectra deleted during coding are unsatisfactory. Interpolation can be performed without feeling uncomfortable. As a result, the analog signal composed of the reproduced musical sound, voice, etc. is of high quality.</p>

Embodiment 1. FIG. 1 is a block diagram showing a configuration of an interpolation processing unit constituting the digital data decoding apparatus according to the first embodiment of the present invention, and FIG. 2 is an interpolation processing unit shown in FIG. 1. It is a block diagram which shows the structure of the digital audio recording / reproduction system. The digital audio recording / playback system of this example uses the ATRAC (Adaptive TRanceform Acoustic Coding) method, which is one of the audio compression coding technologies used in minidiscs (MD) and the like, as shown in Fig. 2. It is composed of a digital audio data coding device 1 and a digital audio data decoding device 2. The number of quantization bits is, for example, 0 to 16 bits, and since the quantization data has a code, the number of quantization bits 1 does not exist, and the minimum number of quantization bits other than 0 is 2.

The digital audio data coding device 1 compresses and encodes the digital audio data DAD into the digital audio coded data CDAD, and then records the digital audio data DAD on a recording medium 3 such as a mini disc. Here, the digital audio data DAD is an analog audio signal composed of musical sounds, voices, etc., encoded by a coding method called linear PCM (Pulse Code Modulation) without compression. When the recording medium 3 is set in the digital audio data decoding device 2, the digital audio data decoding device 2 reads the digital audio encoded data CDAD from the recording medium 3 and then decodes it into the digital audio data DAD'. ..

The digital audio data coding device 1 includes a frequency band dividing unit 11, a time-frequency conversion unit 12, a power calculation unit 13 for each band, a masking calculation unit 14, a minimum audible limit synthesis unit 15, and the number of quantization bits. It has a calculation unit 16, a scale factor calculation unit 17, a quantization unit 18, and a packing unit 19. At the input end of the digital audio data encoding device 1, for example, uncompressed digital audio data DAD (multi-bit data) sampled at a sampling frequency of 44.1 kHz is placed at predetermined time intervals, for example, in frame units of about 11.6 ms. Entered.

The frequency band division unit 11 has a quadrature mirror filter (QMF) which is a kind of band division filter, and divides the input digital audio data DAD into a plurality of frequency bands (subband frames). The subband frame consists of, for example, a low band subband frame SB11 of about 0 to 5.5 kHz, a medium band subband frame SB12 of about 5.5 to 11 kHz, and a high band subband frame SB13 of about 11 to 22 kHz. , Low band subband frame SB21 of about 0 to 5.5kHz, medium band subband frame SB22 of about 5.5 to 11kHz, medium and high band subband frame SB23 of about 11 to 16.5kHz, and high band subband frame of about 16.5 to 22kHz. Some consist of 4 bands of SB24.

The time-frequency transform unit 12 performs a modified discrete cosine transform (MDCT) process on the digital audio data divided into a plurality of subband frame units, so that the MDCT coefficient of the frequency component of the corresponding frequency band is applied. Orthogonal transform to (spectral data). The power calculation unit 13 for each band is provided for each of the above frequency bands, and the MDCT coefficient of the frequency component of the corresponding frequency band is summed to the square, etc., and the spectral power Si (i =) of each of i frequency bands. Calculate 1,2, ..., I, for example, I = 25). However, the calculation method of the spectral power Si is not particularly limited. Here, a critical band (unit: Bark) or the like is used as the frequency band. The critical band refers to a characteristic part of a wideband audio spectrum in which specific psychoacoustic regularities such as frequency selectivity and masking threshold are effective. In addition, power means energy per unit time. The masking threshold will be described later.

The masking calculation unit 14 calculates a masking threshold value due to the simultaneous masking effect that each spectral power Si gives to other frequency bands for each of the frequency bands, and sets the maximum value as the masking threshold value for the spectral power Si. To do. Here, the simultaneous masking effect means that when sounds of a plurality of frequency components are generated at the same time, a sound having a large sound pressure level masks another sound having a low sound pressure level existing in a frequency in the vicinity thereof. , Refers to the audible effect that a person cannot hear or is hard to hear.

The minimum audible limit synthesizer 15 synthesizes the minimum audible limit characteristic lt (f) (dB) and the like represented by Eq. (1) and the above masking threshold value, and the final masking threshold shown in FIG. Determine the value Mi (i = 1,2, ..., I, eg, I = 25) for each frequency band. In this synthesis process, the larger value of the minimum audible limit characteristic lt (f) (dB) and the above masking threshold value is adopted. The minimum audible limit characteristic lt (f) may be stored in the table ROM in advance. In FIG. 3, the masking curve in which the broken lines are combined is shown. lt (f) =-0.6 × 3.64 × (f / 1000)^-0.8+6.5 × exp {-0.6 × (f / 1000-3.3)²}-10^-3× (f / 1000)⁴ ... (1) Note that f above is the frequency (Hz).

The quantization bit number calculation unit 16 first, for each i (frequency band index), the spectral power Si calculated by the power calculation unit 13 for each band and each frequency band calculated by the minimum audible limit synthesis unit 15. Calculate the ratio SMRi with the masking threshold Mi (see equation (2)) for all frequency bands. SMRi = Si / Mi ... (2)

Next, the quantization bit number calculation unit 16 performs the spectral power Si and the quantization noise power Ni (n) when the spectral power Si of each frequency band is quantized by n bits (n = 0 to 16). Calculate the ratio SNRi (n) to and (see equation (3)). SNRi (n) = Si / Ni (n) ... (3) The above ratio SNRi (n) is statistically a constant (20 × log) according to the signal characteristics.₁₀2ⁿ), So it may be calculated in advance by statistical processing. Further, the quantization bit number calculation unit 16 uses the ratio of the above ratio SNRi (n) to the above ratio SMRi to determine the ratio MNRi (n) between the masking threshold Mi and the quantization noise power Ni (n) (Equation (Equation (Equation)). 4) Calculate). MNRi (n) = SNRi (n) / SMRi ... (4)

After that, the quantization bit number calculation unit 16 increases the number of bits n in order from 0, and each time, the ratio MNRi of the masking threshold value Mi of each frequency band and the quantization noise power Ni (n) (n) is calculated, bits are assigned in order from the frequency band that minimizes the ratio MNRi (n), and each time the number of bits n is updated, the frequency band that minimizes the ratio MNRi (n) is similarly allocated. Bits are allotted to, and bits are allotted until a predetermined number of allottable bits according to a desired bit rate is reached. That is, in the spectral power Si, bits are sequentially allocated from the frequency band having the longest length of the portion exceeding the masking threshold value Mi. The quantization bit number WL (i) calculated in this way is equally distributed to the spectra in the same frequency band. However, the method of bit allocation that utilizes the auditory psychological characteristics based on the spectral power Si is not limited to this.

On the other hand, the scale factor calculation unit 17 is about every 2 dB from the absolute maximum value of the MDCT coefficient (for example, 16 bits, full amplitude is 0 dB) of the frequency component of each frequency band orthogonally converted by the time frequency conversion unit 12. Calculate the scale factor (index) (for example, 60 for 0 dB). This scale factor represents the scale (magnitude) factor of the spectrum (MDCT coefficient) and is generally calculated by encoding the absolute value of the maximum spectrum in the quantized frequency unit. Will be done. That is, if the absolute maximum value of the MDCT coefficient of each frequency band is Kmax (i) and the scale factor at that time is SF (i), the scale factor SF (i) satisfying Eq. (5) is calculated. SF (i) x 2^-1/3 Kmax (i) &lt;SF (i) ... (5)

The quantization unit 18 includes the number of quantization bits WL (i) for each frequency band calculated by the quantization bit number calculation unit 16, the scale factor SF (i) calculated by the scale factor calculation unit 17, and the time frequency. The quantized quantization coefficient MK (m) shown in Eq. (6) is calculated based on the MDCT coefficient K (m) of the frequency component of each frequency band that has been orthogonally converted by the conversion unit 12. MK (m) = Round {K (m) × (2^{WL (i) -1}-1) / SF (i)} ... (6) In Eq. (6), m is the index of the MDCT coefficient, i is the index of the quantization frequency band, and Round is a function that rounds off to the nearest whole number. is there.

The packing unit 19 includes a quantization coefficient MK (m) quantized by the quantization unit 18, a quantization bit number WL (i) of each frequency band calculated by the quantization bit number calculation unit 16, and a scale factor calculation unit. After packing the scale factor SF (i) calculated in 17 together with the frame information, it is encoded in the digital audio coded data CDAD and recorded in the recording medium 3.

The digital audio data decoding device 2 includes an unpacking unit 21, an inverse quantization unit 22, a pure tone determination unit 23, a switching unit 24, an interpolation processing unit 25, a frequency time conversion unit 26, and frequency band synthesis. It has a part 27. The unpacking unit 21 has a quantization coefficient MK (m), a quantization bit number WL (i), and a scale factor SF (based on the frame information constituting the digital audio coded data CDAD read from the recording medium 3). i) Unpack.

The inverse quantization unit 22 inverse-quantizes the quantization coefficient MK (m), the number of quantization bits WL (i), and the scale factor SF (i), and based on these, the inverse quantum shown in Eq. (7). The quantized inverse modified discrete cosine transform (IMDCT) coefficient I (m) (spectral data) of the frequency component of each frequency band is calculated. I (m) = SF (i) × MK (m) / (2^{WL (i) -1}-1) ... (7) In Eq. (7), m represents the index of the IMDCT coefficient and i represents the index of the inverse quantization frequency band.

The pure tone determination unit 23 is the maximum value SF of the scale factor SF (i) dequantized by the dequantization unit 22._maxAnd mean SF_avDifference value from (= ΣSFj / J) (SF_max-SCIENCE FICTION_av) And the difference value (SF)_max-SCIENCE FICTION_avThe level of pure tone of the digital audio coded data CDAD is determined based on the magnitude of), and the switching control of the switching unit 24 is performed based on the determination result. That is, the pure tone determination unit 23 uses the difference value (SF)._max-SCIENCE FICTION_av) Is very large (for example, larger than 70 dB), it is judged that the pure tone is high, and the quantization coefficient MK (m), which is inversely quantized by the inverse quantization unit 22, is quantized with respect to the switching unit 24. The number of bits WL (i) and the scale factor SF (i) are controlled to be supplied to the frequency-time converter 26. On the other hand, the pure tone determination unit 23 uses a difference value (SF)._max-SCIENCE FICTION_av) Is 70 dB or less, for example, it is determined that the pure tone is low, and the quantization coefficient MK (m) and the number of quantization bits WL (i) that are inversely quantized by the inverse quantization unit 22 with respect to the switching unit 24. ) And the scale factor SF (i) are controlled to be supplied to the interpolation processing unit 25.

Here, the reason why the pure tone determination unit 23 is provided and the interpolation process described later is not performed on the digital audio coded data CDAD having high pure tone will be described. In the case of digital audio-coded data CDAD with high pure tone quality such as a sine wave, the data bits are concentrated in an extremely narrow band. Therefore, if the interpolation processing described later is performed, the above-mentioned unsatisfactoryness and discomfort will not be eliminated. The sound quality deteriorates. For details of the pure tone determination unit 23, refer to, for example, Japanese Patent Application Laid-Open No. 2005-195983.

Under the control of the pure tone determination unit 23, the switching unit 24 determines the quantization coefficient MK (m), the number of quantization bits WL (i), and the scale factor SF (i) that are inversely quantized by the inverse quantization unit 22. It is supplied to either the interpolation processing unit 25 or the frequency time conversion unit 26. The interpolation processing unit 25 interpolates the IMDCT coefficient I (m) (spectral data) in which the number of quantization bits WL (i) is 0 bits, 2 bits, or 3 bits. The frequency time conversion unit 26 performs IMDCT processing on the IMDCT coefficient I (m) (spectral data) supplied from the interpolation processing unit 25 or the inverse quantization unit 22 for each subband frame, thereby corresponding to the time axis. Orthogonal conversion to the data of. The frequency band synthesizing unit 27 has an inverse quadrature mirror filter (Inverse QMF; Quadrature Mirror Filter), which is a kind of band synthesizing filter, and band-synthesizes a plurality of input frequency bands (sub-band frames) for digital audio. Decrypt to data DAD'.

Next, the function and configuration of the interpolation processing unit 25 will be described. In the first embodiment, the interpolation processing unit 25 performs the following processing. (1) Interpolation processing for the band where the number of quantization bits WL (i) is "2" bits or "3" bits (a) Multiple quantization bits are rounded to "2" bits or "3" bits The spectrum data of is corrected by using a plurality of spectrum data existing in other bands having a quantization bit number WL (i) of "4" bits or more and spectrum data existing in its own band. (b) The above band should have a value because the quantization bit number WL (i) is assigned to "2" bits or "3" bits, but the spectrum is rounded to 0. The data is corrected using the plurality of spectrum data corrected in (a) and the plurality of spectrum data existing in other bands having the quantization bit number WL (i) of "4" bits or more.

(2) At the stage of compression coding, the ratio SMRi is smaller than the offset value determined by the adopted bit rate, so the number of quantization bits WL (i) is set to "0" bits. Correction processing for multiple spectral data that should originally exist in the band (a) For all spectral data, use multiple spectral data that exist in other bands with a quantization bit rate of WL (i) of "0" bits or more. Interpolate. (b) Masking threshold value and minimum audible limit characteristic lt (f) (dB) calculated using multiple spectral data in which the number of quantization bits WL (i) exists in other bands of "2" bits or more. Based on the synthetic masking curve prepared by synthesizing and, the coefficients of all the spectral data obtained by the interpolation in (a) are corrected. (3) In the processing of (1) and (2), when a plurality of spectral data exist in the same band of either one or both of the previous frame and the subsequent frame, these plurality of spectral data are used. Interpolate.

Next, the configuration of the interpolation processing unit 25 will be described with reference to FIG. The interpolation processing unit 25 includes a quantization bit number determination unit 31, an IMDCT coefficient primary correction unit 32, a primary gain control unit 33, a power calculation unit 34 for each band, a masking calculation unit 35, and a minimum audible limit. It is composed of a synthesis unit 36, an MNR calculation unit 37, a coefficient storage unit 38, an IMDCT coefficient second-order interpolation unit 39, and a second-order gain control unit 40.

The quantization bit number determination unit 31 determines how many bits the quantization bit number WL (i) of the input IMDCT coefficient I (m) is, and supplies the determination result to the IMDCT coefficient primary correction unit 32. .. The IMDCT coefficient primary correction unit 32 uses the number of quantization bits WL (i), the scale factor SF (i), and the IMDCT coefficient I (m) (spectral data) inversely quantized by the inverse quantization unit 22. Based on this, in the frequency region, Lagrange interpolation or spline interpolation is performed on the spectral data having the quantization bit number WL (i) of "2" bits or "3" bits, and the existing quantized spectral data is subjected to the above Lagrange interpolation or Replace with the spectral data on the interpolation curve obtained by spline interpolation. Lagrange interpolation is an interpolation that calculates a single interpolation polynomial that passes through all the points in a certain interval, and has a feature that a target value can be calculated even if the intervals of known data are not equal. Spline interpolation is a representative of interpolation that calculates a piecewise polynomial that constitutes an equation that passes through all the given points by dividing it into certain intervals and calculating the equation for each interval.

Further, the IMDCT coefficient primary correction unit 32 has the number of quantization bits WL (i), the scale factor SF (i), and the IMDCT coefficient I (IMDCT coefficient I) after the interpolation processing of the previous frame stored in the coefficient storage unit 38. Based on m) (spectral data) and the IMDCT coefficient I (m) (spectral data) inversely quantized by the inverse quantization unit 22, the number of quantization bits WL (i) for the same frequency in the previous and next frames. Performs Lagrange interpolation or spline interpolation on the "2" bit or "3" bit spectrum data, and replaces the existing quantized spectrum data with the spectrum data on the interpolation curve obtained by the above Lagrange interpolation or spline interpolation. In this case, the IMDCT coefficient primary correction unit 32 is adjacent to the spectrum data to be corrected when the spectrum data adjacent to the spectrum data to be corrected exists at the same frequency of the preceding and following frames and the adjacent frequency of the same frame. Of the spectral data to be used, the spectral data corrected based on the spectral data having a large amplitude value is used. Here, the above-mentioned "same frequency of the previous and next frames" means that if t1, t2, and t3 are the respective frame numbers and the interpolated frame is t2 in FIG. 3, the interpolated band is i. , T1, t3, multiple IMDCT coefficients (spectrums) of the same frequency I_i(t1), I_iI using (t3)_iIt shows that the interpolation of (t2) is performed. Further, the above-mentioned "adjacent frequencies of the same frame" means that if the interpolated frame is t2 and the interpolated band is i, I_i-1(t2), I_{i + 1}I using (t2)_iIt shows that the interpolation of (t2) is performed. In Figure 3, the solid line is I_iRepresents the spectral power of.

In the first-order gain control unit 33, the quantization coefficient MK (m) is the number of quantization bits WL (i) and the scale factor SF (i) based on the above equation (6) regarding the quantization coefficient MK (m). And the size of the spectrum data corrected by the IMDCT coefficient primary correction unit 32 is adjusted so as to fall within the theoretical range of the MDCT coefficient K (m) determined from the quantization coefficient MK (m). For example, as shown in FIG. 4, the number of quantization bits WL (i) is "2" bits, and the scale factor SF (i) is 2.⁵, When the number of spectrum data in the quantization frequency band i is 6, the spectrum data before the correction processing is rounded to 0, and the spectrum data after the correction processing is 0.5 × 2.⁵If it exceeds, the correction range is 0 to 0.5 x 2⁵The gain factor is determined and multiplied by all spectral data in the corrected quantization frequency band i so that it fits within. If the entire corrected range does not fall within the theoretical range of the MDCT coefficient K (m), then 0.5 × 2⁵Or it is clipped at 0. In FIG. 4, x represents the value of the spectrum data before the quantization rounding, and represents the value of the spectrum data after the quantization rounding.

The power calculation unit 34 for each band is provided for each frequency band, and the determination result supplied from the quantization bit number determination unit 31 is the input quantization bit number WL (IMDCT coefficient I (m). When i) indicates that it is a non-zero bit, the IMDCT coefficient I (m) of the input frequency component of the corresponding frequency band is summed to the square, etc., and the spectral power Si of each i frequency band Calculate (i = 1,2, ..., I, for example, I = 25). However, the calculation method of the spectral power Si is not particularly limited. Here, a critical band (unit: Bark) or the like is used as the frequency band.

The masking calculation unit 35 calculates a masking threshold value due to the simultaneous masking effect that each spectral power Si calculated by the power calculation unit 34 for each band gives to other frequency bands, and sets the maximum value for the spectral power Si. Set as the masking threshold. This masking threshold value has the effect of suppressing the fluctuation of the data value in the interpolation by the IMDCT coefficient quadratic interpolation unit 39, which will be described later. The minimum audible limit synthesis unit 36 synthesizes the minimum audible limit characteristic lt (f) (dB) and the like represented by Eq. (1) and the like and the masking threshold value calculated by the masking calculation unit 35. The final masking threshold Mi (i = 1,2, ..., I, eg, I = 25) shown in FIG. 3 is determined for each frequency band. The minimum audible limit characteristic lt (f) may be stored in the table ROM in advance.

The MNR calculation unit 37 assumes that the input quantization bit number WL (i) is linearly converted based on the above equation (4), that is, at the bit rate adopted at the time of compression coding. Assuming that the bits are assigned mechanically based on the determined masking threshold, the range of MNRi (n) for each frequency band is calculated based on the number of quantization bits WL (i). In equation (4), the ratio SNRi (n) is statistically a constant (20 × log) according to the signal characteristics, as described above.₁₀2ⁿ), So it may be calculated in advance by statistical processing. As a result, for example, when the number of quantization bits WL (i) is "3" bits, MNRi (n) is larger than 6 dB and 12 dB or less.

First, the MNR calculation unit 37 masks the spectral power Si calculated by the power calculation unit 34 for each band and each frequency band calculated by the minimum audible limit synthesis unit 36 for each i (frequency band index). Calculate the ratio SMRi to the threshold value Mi (see equation (2) above) for all frequency bands. Next, the MNR calculation unit 37 calculates the ratio of the spectral power Si to the quantization noise power Ni (n) when the spectral power Si of each frequency band is quantized by n bits (n = 0 to 16). Calculate SNRi (n) (see equation (3) above). As described above, the above ratio SNRi (n) is statistically a constant (20 × log) according to the characteristics of the signal.₁₀2ⁿ), So it may be calculated in advance by statistical processing. Further, the MNR calculation unit 37 uses the ratio of the ratio SNRi (n) to the ratio SMRi to determine the ratio MNRi (n) between the masking threshold Mi and the quantization noise power Ni (n) (the above equation (4)). (See) is calculated so that the number of quantization bits WL (i) is included in the range of MNRi (n) for all frequency bands in the calculated range of MNRi (n) for each frequency band. Adjust the overall MNRi (n) offset. This offset value represents the intercept during the linear transformation.

The coefficient storage unit 38 is, for example, a semiconductor memory such as RAM or flash memory, an FD drive in which an FD (flexible disk) is mounted, an HD drive in which an HD is mounted, or an MO disk drive in which an MO (magneto-optical) disk is mounted. , CD-R (Recordable), CD-RW (ReWritable), DVD-R, DVD-RW, etc. are installed in the CD / DVD drive. The coefficient storage unit 38 stores the IMDCT coefficient I (m) (spectral data) after interpolation processing of either one or both of the previous and subsequent frames.

The IMDCT coefficient second-order interpolation unit 39 has a sound spectrum of the IMDCT coefficient I (m) inversely quantized by the inverse quantization unit 22 for a band in which the number of quantization bits WL (i) is 0, that is, a silent band. Lagrange interpolation or spline interpolation is performed using the data, sound spectrum data with IMDCT coefficient first-order correction and first-order gain control, and silence spectrum data in any of the "0" bit bands. Quantize.

The gain control unit 40 is interpolated so that the spectrum data of the "0" bit band interpolated by the IMDCT coefficient second-order interpolation unit 39 is within the value of MNRi (n) or less calculated by the MNR calculation unit 37. The coefficients to be multiplied by the spectral data are determined and multiplied by all the spectral data in the corrected quantization frequency band i.

Next, among the operations of the digital audio recording / playback system having the above configuration, first, the outline of the operation of the digital audio data decoding device 2 will be described. When the recording medium 3 is set on a turntable (not shown) that is rotationally driven by a spindle motor (not shown), the spindle motor rotationally drives the recording medium 3. As a result, the unpacking unit 21 has a quantization coefficient MK (m), a quantization bit number WL (i), and a scale based on the frame information constituting the digital audio coded data CDAD read from the recording medium 3. Unpack factor SF (i).

Next, the inverse quantization unit 22 dequantizes the quantization coefficient MK (m), the number of quantization bits WL (i), and the scale factor SF (i), and based on these, the above equation (7). ), The IMDCT coefficient I (m) of the frequency component of each dequantized frequency band is calculated. In addition, the pure tone determination unit 23 is the maximum value SF of the scale factor SF (i) dequantized by the dequantization unit 22._maxAnd mean SF_avDifference value with (SF_max-SCIENCE FICTION_av) And the difference value (SF)_max-SCIENCE FICTION_av) Is very large (for example, larger than 70 dB), it is judged that the pure tone is high, and the quantization coefficient MK (m), which is inversely quantized by the inverse quantization unit 22, is quantized with respect to the switching unit 24. The number of bits WL (i) and the scale factor SF (i) are controlled to be supplied to the interpolation processing unit 25. On the other hand, the pure tone determination unit 23 uses a difference value (SF)._max-SCIENCE FICTION_av) Is 70 dB or less, for example, it is determined that the pure tone is low, and the quantization coefficient MK (m) and the number of quantization bits WL (i) that are inversely quantized by the inverse quantization unit 22 with respect to the switching unit 24. ) And the scale factor SF (i) are controlled to be supplied to the frequency-time converter 26.

As a result, when the pure tone determination unit 23 determines that the pure tone is low, the switching unit 24 has the quantization coefficient MK (m) and the number of quantization bits WL, which are inversely quantized by the inverse quantization unit 22. (i) and the scale factor SF (i) are supplied to the interpolation processing unit 25. Therefore, when the pure tone determination unit 23 determines that the pure tone is low, the interpolation processing unit 25 sets the number of quantization bits WL (i) to "0" bits, "2" bits, or "3" bits. Interpolate for a certain IMDCT coefficient I (m) (spectral data).

Then, when the pure tone quality determination unit 23 determines that the pure tone quality is low, the frequency time conversion unit 26 receives the IMDCT coefficient I (m) (spectral data) supplied from the interpolation processing unit 25 for each subband frame. ) Is orthogonally converted to the data on the corresponding time axis by performing IMDCT processing. On the other hand, when the pure tone determination unit 23 determines that the pure tone is high, the frequency time conversion unit 26 is supplied from the inverse quantization unit 22 via the switching unit 24 for each subband frame. The IMDCT coefficient I (m) (spectral data) is orthogonally converted to the data on the corresponding time axis by performing the IMDCT process. Next, the frequency band synthesizing unit 27 band-synthesizes a plurality of input frequency bands (sub-band frames) and decodes them into digital audio data DAD'.

Hereinafter, the operation of the interpolation processing unit 25 will be described in more detail. First, the quantization bit number determination unit 31 determines how many bits the quantization bit number WL (i) of the inversely quantized IMDCT coefficient I (m) is, and determines the determination result as the IMDCT coefficient primary correction unit. Supply to 32. The IMDCT coefficient primary correction unit 32 uses the number of quantization bits WL (i), the scale factor SF (i), and the IMDCT coefficient I (m) (spectral data) inversely quantized by the inverse quantization unit 22. Based on this, in the frequency region, Lagrange interpolation or spline interpolation is performed on the spectral data having the quantization bit number WL (i) of "2" bits or "3" bits, and the existing quantized spectral data is subjected to the above Lagrange interpolation or Replace with the spectral data on the interpolation curve obtained by spline interpolation.

Further, the IMDCT coefficient primary correction unit 32 has the number of quantization bits WL (i), the scale factor SF (i), and the IMDCT coefficient I (IMDCT coefficient I) after the interpolation processing of the previous frame stored in the coefficient storage unit 38. Based on m) (spectral data) and the IMDCT coefficient I (m) (spectral data) inversely quantized by the inverse quantization unit 22, the number of quantization bits WL (i) for the same frequency in the previous and next frames. Performs Lagrange interpolation or spline interpolation on the "2" bit or "3" bit spectrum data, and replaces the existing quantized spectrum data with the spectrum data on the interpolation curve obtained by the above Lagrange interpolation or spline interpolation. In this case, the IMDCT coefficient primary correction unit 32 is adjacent to the spectrum data to be corrected when the spectrum data adjacent to the spectrum data to be corrected exists at the same frequency of the preceding and following frames and the adjacent frequency of the same frame. Of the spectral data to be used, the spectral data corrected based on the spectral data having a large amplitude value is used.

In the first-order gain control unit 33, the quantization coefficient MK (m) is the number of quantization bits WL (i) and the scale factor SF (i) based on the above equation (6) regarding the quantization coefficient MK (m). And the size of the spectrum data corrected by the IMDCT coefficient primary correction unit 32 is adjusted so as to fall within the theoretical range of the MDCT coefficient K (m) determined from the quantization coefficient MK (m). For example, as shown in FIG. 4, the number of quantization bits WL (i) is "2" bits, and the scale factor SF (i) is 2.⁵, When the number of spectrum data in the quantization frequency band i is 6, the spectrum data before the correction processing is rounded to 0, and the spectrum data after the correction processing is 0.5 × 2.⁵If it exceeds, the correction range is 0 to 0.5 x 2⁵The gain factor is determined and multiplied by all spectral data in the corrected quantization frequency band i so that it fits within. If the entire corrected range does not fall within the theoretical range of the MDCT coefficient K (m), then 0.5 × 2⁵Or it is clipped at 0.

In the power calculation unit 34 for each band provided for each frequency band, the determination result supplied from the quantization bit number determination unit 31 is input, and the quantization bit number WL (i) of the IMDCT coefficient I (m) is input. When indicates that is a non-zero bit, the IMDCT coefficient I (m) of the input frequency component of the corresponding frequency band is summed to the square, etc., and the spectral power Si of each of the i frequency bands is obtained. calculate.

The masking calculation unit 35 calculates the masking threshold value based on each spectral power Si calculated by the power calculation unit 34 for each band. The minimum audible limit synthesis unit 36 synthesizes the minimum audible limit characteristic lt (f) and the like represented by the above equation (1) and the above masking threshold value calculated by the above masking calculation unit 35, and shows the figure. Determine the final masking threshold Mi shown in 3 for each frequency band.

The MNR calculation unit 37 assumes that the input quantization bit number WL (i) is linearly converted based on the above equation (4), and the frequency is based on the quantization bit number WL (i). Calculate the range of MNRi (n) for each band. For example, when the number of quantization bits WL (i) is "3" bits, MNRi (n) is larger than 6 dB and 12 dB or less.

First, the MNR calculation unit 37 compares the spectral power Si calculated by the power calculation unit 34 for each band with the masking threshold value Mi of each frequency band calculated by the minimum audible limit synthesis unit 36 for each i. Calculate SMRi (see equation (2) above) for all frequency bands. Next, the MNR calculation unit 37 calculates the ratio of the spectral power Si to the quantization noise power Ni (n) when the spectral power Si of each frequency band is quantized by n bits (n = 0 to 16). Calculate SNRi (n) (see equation (3) above). Further, the MNR calculation unit 37 uses the ratio of the ratio SNRi (n) to the ratio SMRi to determine the ratio MNRi (n) between the masking threshold Mi and the quantization noise power Ni (n) (the above equation (4)). (See) is calculated so that the number of quantization bits WL (i) is included in the range of MNRi (n) for all frequency bands in the calculated range of MNRi (n) for each frequency band. Adjust the overall MNRi (n) offset. This offset value represents the intercept during the linear transformation.

The coefficient storage unit 38 stores the IMDCT coefficient I (m) (spectral data) after interpolation processing of either one or both of the previous and subsequent frames.

The IMDCT coefficient second-order interpolation unit 39 has a sound spectrum of the IMDCT coefficient I (m) inversely quantized by the inverse quantization unit 22 for a band in which the number of quantization bits WL (i) is 0, that is, a silent band. Lagrange interpolation or spline interpolation is performed using the data, sound spectrum data with IMDCT coefficient first-order correction and first-order gain control, and silence spectrum data in any of the "0" bit bands. Quantize.

The gain control unit 40 is interpolated so that the spectrum data of the "0" bit band interpolated by the IMDCT coefficient second-order interpolation unit 39 is within the value of MNRi (n) or less calculated by the MNR calculation unit 37. The coefficients to be multiplied by the spectral data are determined and multiplied by all the spectral data in the corrected quantization frequency band i.

As described above, the interpolation processing unit 25 performs the following processing. (1) Interpolation processing for the band where the number of quantization bits WL (i) is "2" bits or "3" bits (a) The number of quantization bits WL (i) exists in other bands of "4" bits or more The number of quantization bits is rounded to "2" bits or "3" bits based on the plurality of spectral data having the maximum value and the value next to the maximum value and the spectral data existing in its own band among the plurality of spectral data to be processed. Correct multiple spectral data. Further, when the corrected spectrum data exists within the range determined by the scale factor SF (i) and the number of quantization bits WL (i), the spectrum data obtained by the above interpolation and coefficient multiplication is adopted as it is. If it exists outside the above range, it is corrected to a value within the above range. (b) Since the above band has the number of quantization bits WL (i) assigned to "2" bits or "3" bits, it should have a value, but the spectrum is rounded to 0. Regarding the data, using the plurality of spectral data interpolated and corrected in (a) and the plurality of spectral data existing in other bands having the quantization bit number WL (i) of "4" bits or more, (a). ), Correct.

(2) At the stage of compression coding, the ratio SMRi is smaller than the offset value determined by the adopted bit rate, so the number of quantization bits WL (i) is set to "0" bits. Correction processing for multiple spectral data that should originally exist in the band (a) For all spectral data, among multiple spectral data existing in other bands with a quantization bit rate of WL (i) of "2" bits or more. , The silent frequency band i is interpolated based on a plurality of spectral data having a maximum value and a value next to the maximum value. (b) The masking threshold value calculated using the spectral data existing in the other band where the number of quantization bits WL (i) is "2" bits or more and the minimum audible limit characteristic lt (f) (dB) are combined. To create a synthetic masking curve. The offset SMRi obtained by dividing the power value of the spectrum data of the band i interpolated in (a) above by the composite masking curve is assumed to have set WL (i) to "0" during compression coding (the offset ( If it is smaller than the threshold value, the spectrum data existing in the band i interpolated in (a) above is used as it is, and if it is larger than the offset (threshold) value, the band obtained in (a) above is used. Correct the coefficient of the spectral data existing in i. (3) In the processing of (1) and (2), when a plurality of spectral data exist in the same band of either one or both of the previous frame and the subsequent frame, these plurality of spectral data are used. Interpolate.

As described above, according to the first embodiment of the present invention, the resolution of a spectrum having a low quantization resolution of binary or quadrature can be increased, and the number of quantization bits of the spectral data having a high masking effect can be reduced. By diverting the compression coding method (algorithm) as much as possible, when decoding compressed digital data, the information of the spectrum deleted at the time of coding is interpolated without feeling unsatisfactory or uncomfortable. can do. As a result, the analog signal composed of the reproduced musical sound, voice, etc. is of high quality.

Further, according to the first embodiment of the present invention, the ratio MNRi (n) is calculated, and the coefficient to be multiplied by the plurality of interpolated spectral data is determined based on the ratio MNRi (n). , The magnitude of silent spectral data can be appropriately interpolated by compression coding at any bit rate.

Embodiment 2. In the first embodiment described above, an example in which the digital audio data decoding device 2 is configured by hardware is shown, but the present invention is not limited to this. That is, among the above digital audio data decoding devices 2, the interpolation processing unit 25 is a CPU (central processing device), an internal storage device such as ROM or RAM, an FD drive, an HD drive, an MO disk drive, or a CD / DVD. It may be composed of an external storage device such as a drive, an output means, and a computer having an input means. Then, it may be configured that the function of the CPU is stored in a semiconductor memory such as ROM or a storage medium such as FD, HD or CD-ROM as an interpolation processing program. In this case, the internal storage device or the external storage device serves as the coefficient storage unit 38, and the interpolation processing program is read from the storage medium into the CPU and controls the operation of the CPU. When the interpolation processing program is started, the CPU calculates the power for each band by the quantization bit number determination unit 31 constituting the interpolation processing unit 25, the IMDCT coefficient primary correction unit 32, the primary gain control unit 33, and the band. It functions as a unit 34, a masking calculation unit 35, a minimum audible limit synthesis unit 36, an MNR calculation unit 37, an IMDCT coefficient secondary interpolation unit 39, and a secondary gain control unit 40, and is controlled by an interpolation processing program. The above processing is executed.

Although the embodiments of the present invention have been described in detail with reference to the drawings, the specific configuration is not limited to these embodiments, and design changes and the like within a range not deviating from the gist of the present invention, etc. Even if there is, it is included in the present invention. For example, in each of the above-described embodiments, the present invention has shown an example of application to a digital audio recording / playback system to which the ATRAC method is applied, but the present invention is not limited thereto. Since the present invention only needs to have an IMDCT coefficient I (m) (spectral data), a quantization bit rate WL (i) and a scale factor SF (i), for example, MP3 (MPEG Audio Layer-3) method, AAC ( It can also be applied to digital audio recording / playback systems to which audio compression coding technology such as Advanced Audio Coding) method or WMA (Windows Media Audio) (Windows is a registered trademark) method is applied.

Further, in each of the above-described embodiments, the number of quantization bits is 0 to 16 bits, but the present invention is not limited to this, and the number of quantization bits may be any number of bits. It can also be applied to a method in which the quantization coefficient is encoded by a Huffman code. In connection with this, in each of the above-described embodiments, an example of performing interpolation processing on a band in which the number of quantization bits WL (i) is 0 bits, 2 bits, or 3 bits has been shown. , Not limited to this, the interpolation processing may be performed for the band in which the number of quantization bits WL (i) is 4 bits or more.

Further, in each of the above-described embodiments, an example in which the present invention is applied to decode uncompressed digital audio data from compressed digital audio data recorded on a recording medium such as a minidisc (MD). Shown, but not limited to this. In the present invention, for example, an analog video signal that changes at a speed equivalent to that of a normal consumer is compressed and encoded into digital video data based on visual and psychological characteristics, recorded on a recording medium, and read from the recording medium. It can also be applied when decoding the output compressed encoded digital video data into uncompressed digital video data.

Further, in each of the above-described embodiments, an example in which a mini disc (MD) is used as a recording medium is shown, but the recording medium is not limited to this, and the recording medium is, for example, a compact disc (CD) or a DVD (Digital Versatile Disk). , Hard disk (HD), or semiconductor memory such as flash memory. Further, in each of the above-described embodiments, MDCT is mentioned as a conversion method for generating coded digital data, but the present invention can be applied to any conversion method for orthogonal conversion such as DCT. Further, in each of the above-described embodiments, an example in which simultaneous masking is used as masking is shown, but the present invention is not limited to this, and temporal masking may be used, or both simultaneous masking and temporal masking may be used.

<figref num="1">It is a block diagram which shows the structure of the interpolation processing part which comprises the digital audio data decoding apparatus which concerns on Embodiment 1 of this invention.</figref><figref num="2">FIG. 5 is a block diagram showing a configuration of a digital audio recording / playback system to which the interpolation processing unit shown in FIG. 1 is applied.</figref><figref num="3">It is a figure which shows the spectral power of each frequency band calculated by the power calculation part in the minimum audible limit synthesis part shown in FIG. 1 and FIG.</figref><figref num="4">It is a figure which shows an example of the MDCT coefficient.</figref>

Code description

1 Digital audio data coding device 2 Digital audio data decoding device 3 Recording medium 11 Frequency band division section 12-hour frequency conversion section 13,34 Power calculation section for each band 14,35 Masking calculation section 15,36 Minimum audible limit synthesis section 16 Quantization bit number calculation unit 17 Scale factor calculation unit 18 Quantization unit 19 Packing unit 21 Unpacking unit 22 Inverse quantization unit 23 Pure sound judgment unit 24 Switching unit 25 Interpolation processing unit 26 Frequency time conversion unit 27 Frequency band synthesis unit 31 Quantization bit number determination unit 32 IMDCT coefficient primary correction unit 33 Primary gain control unit 37 MNR calculation unit 38 Coefficient storage unit 39 IMDCT coefficient secondary interpolation unit 40 Secondary gain control unit