JP3478267B2 - Digital audio signal compression method and compression apparatus - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a digital audio signal compression method and a compression device used in audio signal compression according to the MPEG / Audio standard, ATRAC standard, Dolby digital standard and the like.

[0002]

2. Description of the Related Art FIG. 3 shows MPEG (Moving Picture Coding E).
FIG. 3 is a circuit diagram showing a configuration of a digital audio signal compression circuit according to the xperts Group) / Audio standard. In this figure, the input digital audio signal Da is divided into blocks (called frames) for each predetermined number of samples, and divided into two paths for processing. First, the filter bank 1 in one path has an input signal of equal bandwidth 32
The band is divided into subband signals. In this case, each subband signal is downsampled to a sampling frequency of 1/32. The scale factor extraction / normalization circuit 2 detects the sample having the maximum absolute value for each subband signal in one frame. A value obtained by converting the value into a logarithm and quantizing the value is called a scale factor.
Each subband sample is then divided by this scale factor and their values are normalized to within ± 1.

On the other hand, the psychoacoustic analysis unit 3 calculates the frequency spectrum by FFT (Fast Fourier Transform),
A masking threshold value for each sub-band is calculated and output based on each frequency data thus obtained. The bit allocation unit 4 determines the number of quantization bits for each subband by iterative loop processing under the limitation of the number of bits that can be used in one frame determined by the output of the psychoacoustic analysis unit 3 and the bit rate. The quantizer 5 uses the scale factor extraction / normalization circuit 2 with the number of quantization bits set for each subband.
Quantize the subband signal output from. The bitstream generation unit 6 multiplexes the quantized subband samples, the bit allocation information for each subband, and the scale factor, attaches a header thereto, and creates and outputs a bitstream.

Next, an example of a processing procedure in the conventional psychoacoustic analysis unit 3 will be described. The procedure described below is based on the MOD in the psychoacoustic model according to ISO / IEC 11172-3.
It is the procedure of EL1. (1) Obtain frequency characteristics by FFT and obtain 512 frequency data. (2) Obtain each sound pressure level of 32 sub-bands. (3) Determine the absolute threshold. (4) Select the frequency (masker) that can be heard as a sound. (5) Reduce the masker. (6) Calculate individual mask thresholds. (7) Calculate the global mask threshold. (8) Determine the minimum mask threshold for each subband. (9) Calculate the signal-to-mask ratio (SMR) for each subband. Then, the SMR is output to the bit allocation unit 4 as bit allocation information.

[0005]

By the way, the above-described processing in the psychoacoustic analysis unit 3 has a drawback that the calculation takes time. In particular, calculation of the above (6) and (7) took a long time. For example, the calculation in (7) above

[Equation 1] Even if the calculation of log is excluded, i is 1 to about 130, m and n are the numbers of sounds and noises of about 10 to 20, and therefore LTtm and LT
Each nm must be calculated 1000 times or more.
LTtm and LTnm are expressed by the sum of three terms, but two of them are linear functions and it takes time to calculate.

The present invention has been made in consideration of such circumstances, and an object thereof is to provide a digital audio signal compression method and a compression apparatus capable of significantly reducing the calculation time as compared with the conventional method. .

[0007]

The present invention has been made to solve the above problems, and the invention according to claim 1 is
The input digital audio signal is frequency-divided into multiple sub-bands, and psychoacoustic analysis processing is performed.
In the digital audio signal compression method of allocating bits of each subband according to the result of the psychoacoustic analysis process and quantizing and outputting the signal of each subband according to the bit allocation, the psychoacoustic analysis process processes the input digital audio signal. After frequency analysis and conversion into frequency components, the maximum value of frequency components for each subband is detected.
The detected maximum value is weighted according to the frequency band of that subband, and the variance of the frequency component of each subband is calculated using each weighted maximum value. It is a digital audio signal compression method.

Further, the invention according to claim 2 is the digital audio signal compression method according to claim 1, wherein the psychoacoustic analysis processing calculates the variance,
The value obtained by the said calculation, characterized by calculating the data corresponding to the auditory sensitivity obtained in advance. The invention according to claim 3 provides an input digital audio signal.
A frequency division into a plurality of subbands , a psychoacoustic analysis means for performing a psychoacoustic analysis processing, a bit allocation means for allocating bits of each subband according to the result of the psychoacoustic analysis processing, and a bit allocation according to the bit allocation. In a digital audio signal compression apparatus comprising a quantizing means for quantizing and outputting a signal of each sub-band, the psychoacoustic analyzing means frequency-analyzes the input digital audio signal and converts it into frequency components. And, detect the maximum value of the frequency component for each sub-band,
Second means for weighting the detected maximum value according to the frequency band of the subband, and third means for calculating the variance of the frequency component for each subband using each weighted maximum value. And a digital audio signal compression device.

[0009] The invention of claim 4, claim
In the digital audio signal compressing device according to the third aspect, the psycho-acoustic analysis means calculates a value corresponding to a variance obtained by the calculation by the third means, the data corresponding to a previously-obtained auditory sensitivity. Is further provided.

[0010]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of the present invention will be described below with reference to the drawings. The compression circuit to which the compression method according to this embodiment is applied is the same as that in FIG. 3, and the processing procedure described below is used in the psychoacoustic analysis unit 3 in FIG. FIG. 1 is a flow chart for explaining the compression method according to the embodiment. Below, each step S
1 to S6 will be sequentially described. In the following description, the number of subbands is 32, and the number of frequency data in each subband is 16.

Step S1 (frequency analysis) FFT processing (calculation of the formula described in step S1) on the input digital audio data (real number)
In the case of MPEG / Audio Layer2 (MP2), 1024 data is obtained as frequency data F (k). This frequency data F (k) is complex number data, and FFT of real number input
Depending on the symmetry of, 512 data is meaningful.
It should be noted that j in the formula in the figure is an imaginary unit.

Step S2 (Sound Pressure Level Measurement) The sum of the squares of the real part and the imaginary part of each frequency data F (k), that is, the square of the absolute value P (k) of F (k) is obtained. This value P (k) corresponds to the sound pressure level. Step S3 (measurement of average value) The sound pressure data of 512 described above is divided into 32 sub-bands (bands) every 16 data. Then, the average value E (sb) of the sound pressure level P (k) is obtained for each subband.
In addition, sb is a sub-band number, and is 0 to 3 from the low tone side
1 is assigned. For example, the subband of sb = 3 includes sound pressure levels P (48) to P (63).

Step S4 (weighting of maximum value) The maximum value P ′ of the sound pressure level P (k) for each subband
(K) is detected, and sqrt is added to the detected sound pressure level P ′ (k).
The weighted sound pressure level P ′ (k) is obtained by multiplying the value by (33 / (sb + 1)) {sqrt: square root}.
To get Step S5 (calculation of variance) For each subband, the variance V (sb) of the sound pressure level P (k) is calculated using the calculation results of steps S2 to S4 described above. Here, the weighted sound pressure level P ′ (k) is used as the maximum sound pressure level P (k) in each band.

Step S6 (calculation of SMR) The SMR of each subband is calculated. That is, the logarithm of the variance V (sb) is taken for each subband, multiplied by 2.5, and 0.5 times the auditory sensitivity data Q ′ (sb) is subtracted from the value. Here, the hearing sensitivity data Q ′ (sb) is data corresponding to the hearing sensitivity curve of the human ear,
It is stored in the memory in advance. FIG. 2 shows an example of the sensitivity auditory data. In this figure, "Fs" is a sampling frequency when converting an analog audio signal into a digital audio signal.

The above is the processing procedure according to the embodiment of the present invention. As is clear from the above, the SMR calculation method of the psycho-acoustic analysis unit 3 according to this embodiment basically calculates the variance of the sound pressure level P (k) for each subband, and calculates the calculated variance value. Is used as the SMR. Here, the reason why the dispersion can be used as the SMR is as follows. That is, even if the average value of 16 sound pressure levels in each sub-band is the same, the variance has a large value if the variation is large, and conversely, the variance has a small value if the variation is small. If all sound pressure levels are equal, the variance will be zero. On the other hand, when there is little variation in the frequency components in one subband, the human ears hear noise if the phases of the frequency components are different. On the other hand, if there is a peak in the subband,
The sound is perceived. That is, when the variance of frequency components (amplitude or sound pressure level) is calculated, a waveform with a smaller variance is more important than a waveform with a smaller variance, and therefore, it is also necessary to increase the number of bits. A small waveform can reduce the number of bits.

By the way, SMR and dispersion have the following properties. (1) Data amount allocation 1 (bit) is about 6 (d) of SMR
Corresponds to B). (2) The variance has the dimension of the square of the sound pressure level. (3) The sound pressure level has the dimension of the square of the amplitude. (4) It is appropriate to increase the data amount by 1 (bit) when the amplitude doubles. That is, the fourth root is taken to convert the variance into the amplitude, and the common logarithm is taken to obtain 20 times (d
It is appropriate to convert it to B) to obtain SMR. In practice, due to the property of logarithm, the common logarithm of variance may be taken and multiplied by 5. That is, the SMR is basically obtained by the formula: SMR (sb) = 5 × log ₁₀ V (sb).

However, since the human ear has various auditory characteristics, it is possible to obtain better results by utilizing those characteristics. First, the human ear has a property that the frequency resolution is better at lower frequencies, and the resolution becomes worse at higher frequencies. Considering this property, the following correction method can be considered. When calculating the variance, only the maximum sound pressure level P (k) in each subband is increased in inverse proportion to the frequency. That is, assuming that the maximum sound pressure level in each subband is P ′ (k), a correction of P ′ (k) = P ′ (k) × (32 / sb) is added to this P ′ (k), The variance is calculated using the corrected sound pressure level P ′ (k).

Experimentally, since high-frequency characteristics are significantly deteriorated when directly inversely proportional, it is possible to obtain a better result when inversely proportional to the square root. Since sb starts from 0, "1" is added for convenience of calculation. After all, it is preferable to correct the sound pressure level by the following equation. P ′ (k) = P ′ (k) × sqrt (33 / (sb + 1)) The weighting in step S4 described above is this correction. In the calculation of the average value (step S3), the sound pressure level P (k) before correction is used.

Next, the human ear has a frequency characteristic represented by a so-called auditory curve. Letting Q (sb) be the auditory sensitivity in the center of each subband, this value is expressed in units of sound pressure level (dB). The smaller the value (the more negative the value), the better the ear sensitivity. Shows. Therefore,
The SMR is calculated by the following equation that takes into account the hearing sensitivity Q (sb).
Is preferred to be calculated. SMR2 (sb) = 5 × log 10 V (sb) -Q (s
b)

However, actually, for example, considering MP2 and a sampling frequency of 48 KHz, sb = 0.
The frequency corresponding to is 0 to 750 Hz, and there are many fundamental waves of sound in this range, so if the sensitivity is lowered, the sound quality deteriorates, and noise that feels like noise increases.
Therefore, better results are obtained without correction up to about 2 KHz at which the ear sensitivity is improved to some extent. The value thus corrected is designated as Q '(sb). Figure 2 shows this Q '
The value of (sb) is shown. In addition, experimentally, a better result can be obtained by taking the sum at a ratio of 0.5: 0.5 than by simply taking the sum. That is, it is more preferable to calculate the SMR by the following equation. SMR3 (sb) = 2.5 × log 10 V (sb) -0.
5Q ′ (sb) Step S6 described above shows this calculation processing.

As described above in detail, according to the above-described embodiment, the variance is calculated for each subband and the SMR is obtained from the calculation result, so that it is far more than the conventional calculation (see [Equation 1]). The SMR can be obtained by a simple calculation. In the experiment, the SMR took about 1/3 the time of the conventional method.
Could be asked. Note that the method according to the above-described embodiment has a tendency that the characteristics in the high frequency range are quickly deteriorated when the bit rate is lower than in the conventional psychoacoustic model. As a result, in the conventional psychoacoustic model, the quantization noise of the high frequency "Piro Piro" which is conspicuous in the conventional psychoacoustic model is not conspicuous in the method according to this embodiment, and instead the quantization noise of the low frequency "Gosogoso" becomes conspicuous. .

[0022]

As described above, according to the present invention, the frequency component is obtained for each subband, the variance of the obtained frequency component is calculated, and bit allocation is performed based on the calculation result. The advantage is that the calculation time for bit allocation can be significantly shortened compared to the conventional one.

[Brief description of drawings]

FIG. 1 is a flowchart showing a processing procedure of an embodiment of a method according to the present invention.

FIG. 2 is a diagram showing hearing sensitivity data used in the same embodiment.

FIG. 3 is a block diagram showing a configuration example of a digital audio signal compression circuit.

[Explanation of symbols]

3 ... psychoacoustic analysis unit, 4 ... bit allocation unit, 5 ... quantization circuit.

Continuation of the front page (56) References JP-A-6-242798 (JP, A) JP-A-4-104618 (JP, A) JP-A-9-134200 (JP, A) Yamao Yoshio, High-efficiency coding Trends, The Acoustical Society of Japan, 1991, Vol. 12, No. 12, p. 955-961 Akihiko Sugiyama, High Efficiency Coding of Acoustic Signals, Journal of the Television Society, 1994, Vol. 48, No. 4, p. 447-454 Takehiro Moriya, Takao Kaneko, Basic Technology for Information Compression Coding of Speech / Music, Interface, August 1998, p. 92-99 (58) Fields surveyed (Int.Cl. ⁷ , DB name) G10L 19/00 G10L 19/02 H03M 7/30 JISST file (JOIS)

Claims (4)

(57) [Claims] 1. An input digital audio signal is frequency-divided into a plurality of subbands, psychoacoustic analysis processing is performed, bits are assigned to each subband according to the result of the psychoacoustic analysis processing, and each subband is assigned according to the bit allocation. In the digital audio signal compression method for quantizing and outputting a subband signal, the psychoacoustic analysis processing frequency-analyzes the input digital audio signal to convert it into frequency components, and
Detected maximum value of frequency component for each band and detected maximum value
Its performs weighting corresponding to the frequency band of the sub subband, digital audio signal compression method, characterized by using each maximum value by weighting a process of calculating the variance of said frequency components for each subband . 2. The psychoacoustic analysis process, after calculating the variance, calculates data corresponding to a previously obtained auditory sensitivity to a value obtained by the calculation. A method for compressing a digital audio signal according to. 3. An input digital audio signal is frequency-divided into a plurality of subbands, and psychoacoustic analysis means for performing psychoacoustic analysis processing, and bit allocation of each subband is performed according to the result of the psychoacoustic analysis processing. In a digital audio signal compression apparatus comprising bit allocation means and quantizing means for quantizing and outputting signals of respective subbands according to the bit allocation, the psychoacoustic analysis means frequency-analyzes the input digital audio signal. The first means for converting to frequency components and the maximum value of frequency components for each sub-band are detected and detected.
Third means for calculating the variance of said frequency components of each sub-band by using the second means for performing a maximum value weighted according to the frequency band of the sub-bands out, each maximum value were weighted A digital audio signal compression apparatus comprising: 4. The psychoacoustic analysis means includes the third hand.
4. The digital audio signal compression apparatus according to claim 3, further comprising fourth means for calculating data corresponding to a previously-obtained auditory sensitivity based on the variance value obtained by the calculation by the step. .