JPS6235680B2 - - Google Patents

Description

[Detailed description of the invention]

The present invention relates to a high-efficiency coding transmission system for audio signals, and particularly to a low-speed audio signal transmission system with a transmission rate of around 10K bits/second. As one of the highly efficient coding transmission methods for audio signals, a method has been proposed in which the audio signal is converted from the time domain to the frequency domain using orthogonal transformation, and the spectrum of the signal is quantized and encoded in the frequency domain before being transmitted. ing. A typical example of this method is IE Transactions on Acoustics Speech and Signal Processing (IEEE
Transactions on Acoustics, Speech, and
Signal Processing) magazine, Volume 25, August issue, 1977, pp.299-309 (PP.299-309, VOL.ASSP-25,
AUGUST, 1977), “Adaptive Transform Coding of Speech.
Signals” (“Adaptive Transform Codig of
An adaptive transform coding (hereinafter referred to as ATC) method is known, which is published in a paper entitled "Speech Signals" (Reference 1). Vocoder driven adaptive transform coding was proposed by R.E. CROCHIERE and others as a method that enables high-quality audio signal transmission with a large amount of information. Transform
Codig” (hereinafter referred to as VD-ATC) method is known, and its details can be found in the IEEE Transactions on Acoustics Speech and Signal Processing (IEEE Transactions on Acoustics Speech and Signal Processing). on Acoustics，Speech，
and Signal Processing), Volume 27, 1979, 10.
Monthly issue, pp.512-530 (PP.512-530, VOL.ASSP-
27, OCTOBER, 1979) “Frequency Domain Coding of Speech”
(“Frequency Domain Coding of Speech.”) (Reference 2). FIG. 1 is a block diagram explaining the VD-ATC system. In the figure, on the transmitting side, input terminal 10
A sampled audio signal system example x(n) is input to . The buffer memory 15 stores the input signal in units of M samples in order to block process the input signal in units of M samples. The window function circuit 20 multiplies the M sample signals stored in the buffer memory 15 by a predetermined window function. Second
The figure shows an example of the window function described in Document 2. In FIG. 2, n indicates the sample time, and m indicates the number of overlapping samples. Here M
If a sample value series consisting of sample values is defined as a block, the solid line part in FIG. 2 shows the window function of the current block, and the broken line part shows the window functions of the blocks before and after it. The output value of the window function circuit 20 is input to a discrete cosine transform circuit (hereinafter referred to as a DCT circuit) 30. Here, M-sample discrete cosine transform (Discrete Cosine Transform;
According to the above-mentioned document 2, the DCT (hereinafter referred to as DCT) is calculated according to equations (1) and (2). here

[Formula] In equation (1), v(n) (n=0,1,...,M
−1) is the output value of the window function circuit 20, and V _c (k)
(k=0, 1,...,M-1) indicates the DCT coefficient of M samples of V(n). (Similar expressions will be used hereinafter.) Inverse cosine transform (hereinafter referred to as IDCT) is calculated according to the following formula. The method for calculating the DCT coefficient for M samples is 2M
Discrete Fourier transform of the sample
A calculation method using Fourier Transform (hereinafter referred to as DFT) is known. This calculation method will be briefly explained below. Now, let u(n) be a sample value series of M samples, and define it as follows. If the DFT of 2M samples is U(k), U(k) is Therefore, the DCT series of M samples, V _c (k), can be found as follows. V _c (k)=Re{c(k)exp(-jπk/2M)U(k)} k=0,1,2...,M-1 (6) In the above formula, the symbol Re{ } is the real part represents. In addition, as a calculation method that significantly reduces the amount of calculation,
IE Transactions by J.MAKHOUL
IEEE Transactions on Acoustics, Speech and Signal Processing
Acoustics, Speech, and Signal Processing)
"A Fast Cosine Transform in One and Two Dimensions" published in 1980, Volume 28, February issue, pages 27-34 of One
It is explained in detail in the paper titled "And Two Dimensions", so I will not explain it here.
Returning to FIG. 2 again, the output value of the DCT circuit 30,
That is, the M sample DCT coefficients are generated by the autocorrelation calculation circuit 40 and the DCT coefficient quantization encoding circuit 1.
10 is input. The autocorrelation calculation circuit 40 calculates the square value of each DCT coefficient of M samples which are input values, and applies the inverse of 2M samples to these calculation results.
DFT (IDFT) is applied to calculate M pseudo-autocorrelation numbers. The M pseudo-autocorrelation numbers are provided to the pitch extraction circuit 50 and the prediction parameter calculation circuit 70, respectively. The pitch extraction circuit 50 has M
The pitch period P and the pitch gain P _G are calculated using the autocorrelation coefficients P and P G and are supplied to the quantizer 60 .
The prediction parameter calculation circuit 70 calculates a predetermined number of prediction parameter value sequences (as an example, a K parameter value sequence) using M autocorrelation numbers, and converts this K parameter value sequence into a quantizer 60. supply to. Note that the K parameter is also called a Percall parameter. The quantizer 60 quantizes the pitch period P, the pitch gain _PG , and the K parameter value series using a predetermined number of quantization bits, and provides the obtained quantization parameter values to the inverse quantizer 65 and the side information encoder 130. . The inverse quantizer 65 inversely quantizes the input quantization parameter value and supplies the obtained pitch period P' and pitch gain P _G ' to the spectrum regeneration circuit 90. Further, the K parameter value series obtained by inverse quantization is provided to the parameter conversion circuit 80. The parameter conversion circuit 80 converts the input K parameter value series into a parameter value series (for example, an α parameter value series) suitable for spectrum reproduction, and provides this α parameter value series to the spectrum reproduction circuit 90 . Note that the α parameter is also called a linear prediction coefficient. The spectrum reproducing circuit 90 calculates the pitch spectrum σ _p (k) using the pitch period P' and the pitch gain P _G '. Furthermore, the envelope spectrum σ _f (k) is calculated using the α parameter value series. Here, the envelope spectrum is also called formant spectrum. Furthermore, the spectrum is reproduced using σ _p (k) and σ _f (k). Below, the reproduced spectrum is σ(k)
It will be written as (k=0, 1..., M-1). Next, σ _f (k) and σ(k) are input to the bit allocation step size calculation circuit 100. The calculation algorithm of the bit allocation step size calculation circuit 100 will now be described. First, we will show the step size calculation algorithm. Δ(k)=Q・A(b(k))・σ(k) (k=0, 1,...,M−1) (7) Here, Δ(k) is the k-th DCT coefficient σ(k) is the reproduced spectrum. b(k) is the number of allocated bits, and A(b(k)) is a constant determined by b(k). Q is a load factor and determines the noise characteristics of the quantizer with respect to load. Next, the bit allocation calculation algorithm is shown. kth DCT
The number of bits allocated to the coefficient is calculated by the following equation. b(k)=δ+1/2log ₂ w(k)σ ² (k)/D (k=0, 1,...,M-1) (8) Here, δ is w(k) is the frequency weighting function, and D is the quantization noise power expressed by the following equation. Here, σ _e ² (k) indicates the quantization noise power of the k-th DCT coefficient. The frequency weighting function w(k) is
It is expressed as follows. w(k)=σ _f ² 〓(k) (k=0, 1,..., M-1) (10) Here, σ _f (k) is the aforementioned envelope spectrum. γ is a factor that determines the frequency characteristics of quantization noise power, and takes a value of −1≦γ≦0. FIG. 3 shows changes in the quantization noise power σ _e ² (k) depending on the value of γ.
In FIG. 3, the horizontal axis shows the value of k, and the vertical axis shows the logarithmic power. Moreover, the solid line part shows the signal power spectrum, and the broken line part shows the quantization noise power spectrum. When γ=0, the quantization noise power spectrum exhibits flat characteristics, and when γ=−1, it exhibits similar characteristics to the signal power spectrum.
As the value of γ, a value of −1<γ<0 is normally used. The method of changing the characteristics of the quantized noise power spectrum depending on the value of γ is called noise shaping. Next, the DCT coefficient adaptive quantization encoding circuit 110 will be explained. The DCT coefficient adaptive quantization encoding circuit 110 uses the quantization step size Δ(k) and the number of allocated bits b(k), which are the output values of the bit allocation step size calculation circuit 100, to calculate the output value of the DCT circuit 30. is
Adaptively quantize and encode DCT coefficients. The encoded DCT coefficients are provided to multiplexer 120. Side information encoder 130 encodes the quantization parameter value from quantizer 60 as side information and supplies it to multiplexer 120. The multiplexer 120 transmits the output code of the side information encoder 130 and the output code of the DCT coefficient adaptive quantization coding circuit 110 via the transmission side output terminal 140 every M sample times. Next, the operation on the receiving side will be explained. On the receiving side, the transmitted code received at the receiving side input terminal 150 is given to a demultiplexer 160, which converts the received code into
The code is separated into a code representing the DCT coefficient and a code representing side information. The codes representing the separated DCT coefficients are provided to a DCT coefficient adaptive decoder 200 and the codes representing side information are provided to a side information decoder 170. Side information decoder 170
decodes the input code, separates the pitch period P' and pitch gain P _G ' from the K parameter value series, and sends the K parameter value series to the parameter conversion circuit 1.
75, and the pitch period P' and pitch gain P _G ' are supplied to the spectrum regeneration circuit 180. Here, the parameter conversion circuit 175 performs the same operation as the parameter conversion circuit 80 on the transmitting side, converting the K parameter value series to the α parameter value series,
Spectrum regeneration circuit 18 for α parameter value series
Give to 0. The spectrum regeneration circuit 180 performs the same operation as the spectrum regeneration circuit 90 on the transmitting side, and supplies the regenerated spectrum σ(k) and the envelope spectrum σ _f (k) to the bit allocation step size calculation circuit 190. The bit allocation step size calculation circuit 190 is a bit allocation step size calculation circuit 100 on the transmitting side.
Perform the same operation as quantization step size Δ
(k) and the allocated bit number b(k) are supplied to the DCT coefficient adaptive decoder 200. The DCT coefficient adaptive decoder 200 decodes the value input from the demultiplexer 160 using the Δ(k) and b(k).
Get the DCT coefficients. The obtained DCT coefficients are the inverse DCT
(IDCT) circuit 210, M samples of
Perform inverse cosine transformation on the DCT coefficients and reproduce M samples of reproduced signal (n) (n=0, 1,...,M-1)
get. The buffer memory circuit 220 once stores M samples (n) and then outputs them through the receiving side output terminal 230. The VD-ATC method explained above is 16k bit/
Very high quality audio signals can be obtained at transmission speeds of around seconds. However, this VD
- Since the ATC method has to transmit both DCT coefficients and side information, it is difficult to significantly compress the amount of transmitted information, and the quality of the reproduced audio deteriorates at low-speed transmission (for example, 9.6 kbit/s or less). It has the disadvantage of deterioration. An example of audio quality deterioration is that pitch information is not reproduced sufficiently, resulting in reproduced audio sounding hoarse. This is due to the fact that the envelope spectrum, pitch spectrum, etc. cannot be reproduced satisfactorily in low-speed transmission. An object of the present invention is to provide an adaptive audio signal transmission method that can reduce the amount of transmitted information and reproduce high-quality audio signals even during low-speed transmission. The transmission method according to the present invention is an adaptive audio signal transmission method in which an orthogonal transform sequence obtained by orthogonally transforming a sample value sequence obtained by sampling an audio signal is adaptively quantized and transmitted. generating a first signal sequence representing a characteristic of the first signal sequence, based on the first signal sequence, generating a second signal sequence representing a characteristic of the first signal sequence; generating a third signal sequence representing a spectrum, calculating a difference sequence between the first signal sequence and the third signal sequence, and generating a fourth signal sequence by adaptively quantizing the difference sequence; It is characterized in that the second signal series and the fourth signal series are transmitted in combination. Next, the present invention will be explained in detail with reference to the drawings. FIG. 4 is a block diagram showing a first embodiment of the present invention. In this embodiment, DCT is used as the orthogonal transformation. On the transmitting side in FIG. 4, a buffer memory 15 and a window function circuit 20
performs the same operations as the components with the same numbers in FIG. The DCT circuit 300 performs discrete cosine transformation (DCT) on the M output values of the window function circuit 20.
, and the DCT coefficient V _c (k) (k=0, 1,...,M-
1) Output. The DCT calculation method is as described above.
As mentioned in the explanation of the VD-ATC method, it can be calculated using DFT or the method disclosed in Document 3. The absolute value circuit 310 receives input V _c (k)
The code information of (k=0, 1, . . . , M-1) is given to the adaptive quantization encoder 390. Further, the absolute value |V _c (k)| of V _c (k) is calculated, and the calculation result is provided to the logarithm calculation circuit 320 and the subtracter 380 . The logarithm calculation circuit 320 calculates the absolute value of the DCT coefficient |V _c (k)|(k
=0,1,...,M-1) logarithm LOG |V _c (k)|
Calculate the value of (k=0, 1,...,M-1). 2M
Irverse Discrete Fourier transform of a point
Fourier Transform (hereinafter referred to as IDFT) circuit 3
20 is the output value of the logarithm calculation circuit 320.
Calculate the IDFT and pseudo-cepstrum C(h) (n=
0,1,...,2M-1). The reason why it is expressed as pseudo here is because the cepstrum is obtained from the DCT coefficients. Details of the kebstrum can be found in the 1977 issue of PROCEEDINGS OF THE IEEE by DGCHILDERS et al.
“The Cepstrum: A Guide to Processing” published in October 2016, Volume 65, pp. 1428-1443.
The pseudo-cepstrum C(n) is described in the paper titled "Processing"), so its explanation will be omitted here.
It is input to 0. The lifter circuit 340 multiplies the pseudo cepstrum C(n) by a predetermined window function, separates and extracts a low time part and a high time part, supplies the low time part to a quantizer 350, and pitches the high time part. It is applied to the detection circuit 345. In general, it is known that the low time part of the cepstrum contains the envelope spectrum information of the input audio signal, and the high time part contains the pitch (pitch period and pitch gain), so the same applies to the pseudo cepstrum. It is believed that this holds true. Pitch detection circuit 3
45 detects pitch period P and pitch gain P _G and supplies them to quantizer 350 . Quantizer 350
are the low time portion of the pseudo cepstrum output from the lifter circuit 340 and the pitch period P and pitch gain P output from the pitch detection circuit 345.
_G is quantized with a predetermined number of quantization bits. Hereinafter, the low time portion of the pseudo cepstrum and the quantized values of the pitch period P and the pitch gain P _G will be referred to as side information. The side information output from the quantizer 350 is input to a decoder 355 and a side information encoder 410. Decoder 355
decodes the side information and sends it to the DFT calculation circuit 36
Give to 0. Discrete Fourier transform
Fourier Transform (hereinafter referred to as DFT) circuit 3
60 performs 2M-point DFT on the decoded side information to obtain a reproduced spectrum in the logarithmic domain. The reproduced spectrum of the first M points among these is σ
If _L ′(k) (k=0, 1,...,M-1), σ
_L '(k) is composed of a pitch spectrum σ _P '(k) and an envelope spectrum σ _f '(k) (k=0, 1, . . . , M-1). Here, the envelope spectrum σ
Examples of _f ′(k), pitch spectrum σ _P ′(k), and reproduction spectrum σ _L ′(k) are shown in FIGS. 5a, b, and c.
In Figure 5, the horizontal axis is the value of k (0 to M
-1), and the vertical axis represents logarithmic amplitude. still,
In the figure, each spectrum value is shown consecutively for visual clarity. The reproduced spectrum σ _L ′(k) in the logarithmic domain is
Provided to exponent calculation circuit 370 and bit allocation step size calculation circuit 400. The exponent calculation circuit 370 calculates an exponent for the input value.
That is, the processing in the logarithm calculation circuit 320 and the inverse processing are performed. Output value σ′(k) of index calculation circuit 370
(k=0, 1,...,M-1) is given to the subtracter 380 and is the output value of the absolute value circuit 310 |V
The difference e(k) (k=0, 1,...,M-1) between _c (k)| and the output value σ'(k) of the exponent calculation circuit 370 is calculated, and the calculation result is converted into an adaptive quantization code. 390. The adaptive quantization encoder 390 adaptively quantizes the e(k) using the quantization step size and the number of allocated bits output from the bit allocation quantization step size calculation circuit 400.
Encoding is performed using the code information of V _c (k) supplied from the absolute value calculation circuit 310. The code sequence from the adaptive quantization encoder is provided to multiplexer 420. The bit allocation quantization step size calculation circuit 400 calculates the difference e(k) using the reproduced spectrum σ _L '(k) supplied from the DFT calculation circuit 360.
Calculate the number of quantization bits and quantization step size for As a method of calculating the quantization step size, the above equation (7) can be applied. The simplest method for allocating the number of quantization bits is to allocate the same number of bits (for example, 2 bits). In such a quantization bit allocation method, the quantization noise power spectrum has similar characteristics to the signal power spectrum. That is,
In FIG. 3, the characteristics are similar to those obtained when γ=-1. In the conventional VD-ATC method, when quantizing with the same number of bits, it is necessary to allocate a considerable number of bits (4 bits or 5 bits) to the DCT coefficient V _c (k). Although there was a drawback that the amount of transmitted information increased significantly, in the present invention, the difference e(k) is quantized instead of the CT coefficients, so there is almost no deterioration in sound quality even with a small number of bits. , the amount of transmitted information does not increase significantly. Of course, it is also possible to apply adaptive bit allocation or noise shaping, and in that case, the above-mentioned
Equations (8) to (10) may be applied. However, in order to effectively apply noise shaping in the case of this embodiment, the value of γ must be set to the conventional VD-ATC
A value different from the value mentioned in the method description must be adopted. That is, in the case of this embodiment, the characteristics of the quantization noise power spectrum are flat when γ=1, and are similar to the signal power spectrum when γ=0. Therefore, a value of 0<γ<1 must be adopted as the value of γ. The multiplexer 420 combines the output code sequence of the adaptive quantization encoder 390 and the size information encoder 410.
, and multiplexes these code sequences and transmits them to the receiving side via the transmitting side output terminal 140. Next, the operation on the receiving side will be explained. On the receiving side, the transmitted code sequence is input to the receiving side input terminal 15.
0 and applied to demultiplexer 430. The demultiplexer 430 calculates the difference e(k)(k
=0, 1, . Side information decoder 440
decodes the input code sequence to obtain decoding side information, and outputs these to the DFT calculation circuit 450. Here, the DFT calculation circuit 450 performs the same operation as the DFT calculation circuit 360 on the transmitting side, and uses the decoding side information to calculate the spectrum σ _L ′(k) (k=0, 1, ..., M -1) is played. σ _L '(k) is output to bit allocation step size calculation circuit 460 and exponent calculation circuit 490. Bit allocation step size calculation circuit 4
60 performs the same operation as the bit allocation step size calculation circuit 400 on the transmitting side. In addition, the exponent calculation circuit 490 is connected to the exponent calculation circuit 37 on the transmitting side.
Performs the same operation as 0. The output value σ'(k) (k=0, 1, . . . , M-1) of the exponent calculation circuit 370 is output to the adder 480. adaptive decoder 470
performs the opposite operation to the adaptive quantization encoder 390 on the transmitting side. In other words, the code information of the DCT coefficients is separated from the input code sequence and
It is applied to the IDCT circuit 500. Further, the adder 48 decodes the code representing the difference e(k) using the allocated bit number and the quantization step size.
Give to 0. Adder 480 calculates the difference e(k) and σ′(k)
and represents the absolute value of the DCT coefficient | _c (k)
|(k=0, 1,...,M-1) is obtained. ｜ _c (k)
| is given to an IDCT (inverse cosine transform) circuit 500. The IDCT circuit 500 calculates _{the DCT coefficient c (} k) (k=0,
1,...,M-1), and the inverse DCT of M point is DCT
It is applied to the coefficient _c (k) and reproduced signal (n) ((n
=0,1,...,M-1). IDCT circuit 50
The output value (n) of 0 is the buffer memory circuit 22
0, and after the signal for M samples has been accumulated, it is outputted as a reproduced audio signal via the receiving side output terminal 230. This configuration provides the following effects compared to the conventional VD-ATC method. (1) According to this embodiment, the DCT coefficients are not directly quantized, but the difference e(k) is quantized and transmitted, so the amount of information required to quantize and transmit the difference e(k) is It is possible to reduce this to nearly 1/2 compared to the case of transmitting DCT coefficients in the conventional method, and it is possible to obtain the same sound quality as the conventional VD-ATC method with a small amount of transmitted information. (2) In the conventional VD-ATC system, the side information contributes to the calculation of the quantization step size and the number of allocated bits, and when the amount of transmitted information is set to 9.6 kbps or less, the loss of the spectrum increases, causing problems on the receiving side. Deterioration in sound quality occurs due to the inability to faithfully reproduce the spectral structure required for audio signal reproduction, that is, the spectral envelope structure and pitch structure. However, according to the configuration of this embodiment, the transmission The spectral structure required for audio signal reproduction can be reproduced using the side information obtained.
Even if the amount of transmitted information is reduced to 9.6k bits/second or less, high-quality playback audio can be obtained. (3) Even if a simple method is used to calculate the number of allocated bits (as an example, allocating 2 bits to every difference) to reduce the amount of calculation,
There is almost no deterioration in sound quality. In addition, in calculating the DCT coefficient, 5. a, 5. b. As shown in equation (6), when calculating using DFT, the cepstrum may be calculated using DFT coefficients. The cepstrum obtained in this way is capable of approximating the spectral structure of the input signal better than the pseudo cepstrum calculated from the DCT coefficients. In this case, it is also possible to use an M-point DCT calculation circuit instead of the 2M-point DFT calculation circuits 360 and 450. By using the M-point DCT calculation circuit, the amount of calculation required for DCT calculation can be reduced to approximately 1/2, and the amount of differential transmission information can be reduced. In Fig. 4, the cepstrum is directly quantized, but since the cepstrum is relatively easily affected by quantization, the low time domain cepstrum, which represents the spectral envelope, is not so sensitive to the effects of quantization. Parameter values (as an example,
Quantization may be performed after converting to K parameter). In this case, the number of quantization bits of the K parameter can be reduced, and the amount of side information to be transmitted can be reduced. Note that it is also possible to use a means other than the cepstrum to express the spectrum structure with a small number of bits (for example, it is possible to use the K parameter obtained from the pseudo-autocorrelation coefficients as in the conventional VD-ATC method). It is. As an example, FIGS. 9a to 9i show signals at various parts when an arbitrary block of an input audio signal sequence is processed by the method of the present invention. Note that in the figure, the discrete signal sequence is expressed as a continuous signal. Figure 9a
4 shows the output signal waveform of the window function circuit 20 in the embodiment of FIG. This is a signal waveform of M samples within an arbitrary block of the input audio signal sequence. The horizontal axis in FIG. 9a is time, and the vertical axis is amplitude.
FIG. 9b is an output signal of the DCT circuit 300 of FIG. 4, and shows a discrete cosine transform waveform of the signal waveform of FIG. 9a. The horizontal axis in FIG. 9b is frequency, and the vertical axis is amplitude. FIG. 9c shows the logarithm calculation circuit 320 of FIG.
The horizontal axis shows the frequency and the vertical axis shows the amplitude. FIG. 9d shows the IDFT calculation circuit 330 of FIG.
is the output signal of This signal is called the cepstrum. The horizontal axis shows time and the vertical axis shows amplitude. FIG. 9e shows the output signal of the DFT calculation circuit 360 of FIG. The horizontal axis shows frequency and the vertical axis shows amplitude.
FIG. 9f shows the output waveform of the index calculation circuit 370 of FIG. The horizontal axis is frequency and the vertical axis is amplitude. FIG. 9g shows the number of allocated bits which is the output of the bit allocation quantization step calculation circuit 400 of FIG. The horizontal axis is the frequency, and the vertical axis is the number of allocated bits. Figure 9h shows the IDCT circuit 50 of Figure 4.
Indicates an input signal of 0. The horizontal axis is frequency and the vertical axis is amplitude. FIG. 9i is an output signal waveform of the IDCT circuit 500 of FIG. 4, and shows a reproduced audio signal. The horizontal axis shows time and the vertical axis shows amplitude. FIG. 6 is a block diagram showing a second embodiment of the present invention. In this embodiment, the subtraction operation on the transmitting side and the addition operation on the receiving side are performed not in the linear domain but in the logarithmic domain. In FIG. 6, components labeled with the same numbers as in FIG. 4 indicate that they operate in the same way as in FIG. 4. In such a configuration, on the sending side,
Since an exponent calculation circuit is not required and the difference spectrum can be quantized with a smaller number of quantization bits, it is possible to further reduce the amount of transmitted information. FIG. 7 is a block diagram showing a third embodiment of the present invention. In FIG. 7, components labeled with the same numbers as in FIG. 4 perform the same operations as in FIG. 4. First, the sending side will be explained. The DFT calculation circuit 600 performs M-point DFT on the M-point output sample values of the window function circuit 20. DFT coefficient is X(k)
(k=0,1,...,M-1), the _real part and _imaginary part of
..., M-1), X _R (k) and X _I (k) are given to the absolute value phase calculation circuit 610. calculation circuit 610
An example of the calculation method is shown below. |X(k)|=√( _R (k)) ² + ( _I (k)) ² (11.a) arg(X(k))=tan ^-1 (X _I (k)/X _R (k )) (11.b) (k=0, 1,...,M-1) In the above equation, |X| represents the absolute value of X, and arg(x) represents the phase of X. Here, the input signal of DFT is an audio signal and is a real number. Due to the properties of DFT, it is known that when a real number signal is subjected to DFT, the real part of the DFT coefficient becomes an even function and the imaginary part becomes an odd function. Therefore, the absolute value of the DFT coefficient is an even function, and the phase is an odd function. Therefore, it is clear that all the information of the input signal is contained in the absolute value and phase sequence of M/2 points of the DFT coefficient. From now on, M/2
The absolute value and phase series of the points are respectively |X(k)|
〓, arg(X(k))〓(k=0, 1,..., M/2-1). |X(k)| at point M is given to the subtracter 380 and logarithm calculation circuit 320, and arg(X
(k)) is provided to a phase quantization encoder 620. Logarithm calculation circuit 320 performs the same operation as logarithm calculation circuit 320 in FIG. The IDFT calculation circuit 335 performs an M-point inverse DFT calculation on the output value of the logarithm calculation circuit 320. According to the configuration of this embodiment, the cepstrum C _P (n) (n=0, 1, . . . , M-1) is obtained as the output of the IDFT calculation circuit 335. As is well known, the cepstrum C _P (n) is an even function. lifter circuit 340,
The pitch detection circuit 345, quantizer 350, decoder 355, and side information encoder 410 operate in the same way as the components denoted by the same numbers in FIG. 4, so their explanation will be omitted. Also, DFT calculation circuit 3
65 performs DFT calculation of M points. According to the configuration of this embodiment, the spectral structure is obtained by performing DFT calculation on the cepstrum.
High quality logarithmic domain reproduction spectrum｜X _L ′(k)｜
(k=0, 1,...,M-1) can be obtained. Of course, |X _L ′(k)| is an even function, so
Among the output series of M points of the DFT calculation circuit 360,
The M/2 point sequence (hereinafter referred to as |X _L '(k)|) is supplied to an exponent calculation circuit 370 and a bit allocation/quantization step size calculation circuit 400. Components 370 and 400 perform the same operations as the similarly numbered components in FIG. however,
When the configuration of this embodiment is adopted, component 3
70 and 400 target M/2 point series. The subtracter 380 outputs the absolute value of the M point of the absolute value phase calculation circuit 610 |X(k)|(k=0, 1,...,M-1)
The M/2 point series, that is, |X(k)|〓, and the M/2 point output series of the index calculation circuit 370 |′(k)|
〓 is performed, and the difference is output to the adaptive quantization encoder 39. Adaptive quantization encoder 39
0 performs the same operation as the component with the same number in FIG. Phase quantization encoder 620
quantizes arg(X(k)). As an example of the quantization method of the phase quantization encoder 620, a simple method may be one in which equal bits (for example, about 2 bits) are assigned to each phase component to perform quantization. The multiplexer 630 outputs the output value of the phase quantizer 620 to the adaptive quantization encoder 3.
90 output code sequences and side information encoder 410
It receives the output code sequences of , multiplexes them, and transmits them to the receiving side via the transmitting side output terminal 140. On the receiving side, the code received at the input terminal 150 is applied to the demultiplexer 640, which separates the code sequence representing the phase, the code sequence representing the difference, and the code sequence representing side information from the received code sequence, and demultiplexes the phase. The phase decoder 65
0, a code sequence representing the difference is applied to the adaptive post-coder 470, and a code sequence representing the side information is applied to the side information decoder 440. Phase decoder 65
0 decodes the phase sequence of M/2 points from the input code sequence and outputs it to the conversion circuit 660. The side information decoder 440 performs the same operations as the components with the same numbers in FIG. The cepstrum is given to a DFT calculation circuit 455. The DFT calculation circuit 450, the exponent calculation circuit 490, and the bit allocation quantization step size calculation circuit are respectively provided on the transmitting side.
DFT calculation circuit 365, exponent calculation circuit 370, and bit allocation quantization step size calculation circuit 4
Performs the same operation as 00. adaptive decoder 470
decodes the difference information and outputs it to adder 480. The adder 480 receives the output value of the adaptive decoder 470 and the M/2 point output series of the exponent calculation circuit 490.
X′(k)｜〓(k=0,1,...,M/2-1)
..., M/2-1) is output to the conversion circuit 660. The conversion circuit 660 uses the fact that the phase series is an odd function and the absolute value series is an even function to obtain the phase series and absolute value series of M points from the phase series and absolute value series of M/2 points, and converts these into The real part series _R (k) of M points is
It is output to the IDFT calculation circuit 670. The IDFT calculation circuit 670 performs inverse DFT calculation on _R (k) and _I (k) at M points, and reproduces the reproduced signal (n) (n=0, 1,
..., M-1) is obtained. Buffer memory circuit 220
and the receiving side output terminal 230 perform the same operations as the components with the same numbers in FIG. With the configuration of this embodiment, effects similar to (1) to (3) described in the description of the first embodiment can be obtained. Further, as in the first embodiment, the cepstrum may be converted into another parameter (for example, K parameter), quantized, and transmitted, or means other than the cepstrum may be used. Furthermore, as shown in the second embodiment, the subtraction and addition regions may be made into logarithmic regions. The effect of doing so is the same as in the first and second embodiments. Generally, it is known that the human ear is not very sensitive to the phase component of a signal, so in this embodiment, there is almost no deterioration in sound quality even if the phase component is subjected to fairly coarse quantization. By coarsening the quantization of the phase component, there is an effect that the amount of transmitted information can be reduced. In addition, as a method of phase quantization, the phase component is approximated by a linear function in another block of the method explained in this example, and the difference between the slope of the linear function and the corresponding value of the linear function of each phase component is may be quantized and transmitted. Furthermore, it is also possible to apply a minimum phase condition to the input speech sequence. That is, processing is performed on the assumption that the input speech sequence satisfies the minimum phase condition. Due to the aforementioned reason that the human ear is not very sensitive to phase, applying the minimum phase condition causes less audio degradation. In this case, the phase component can be reproduced on the receiving side, so there is no need to transmit the phase component. Therefore, the amount of information to be transmitted regarding the difference is halved, making it possible to significantly reduce the amount of information to be transmitted overall. FIG. 8 is a block diagram showing a fourth embodiment of the present invention to which the minimum phase condition is applied. In FIG. 8, the components labeled with the same numbers as those in FIG. 7 perform the same operations as in FIG. 7, so their explanations will be omitted here. The operation of lifter circuit 700 in FIG. 8 in the high time region is the same as lifter circuit 340 in FIG. However, in low time domain operation, the cepstral window applied to the cepstrum is as follows.

[Table] In the above equation, (n) represents a cepstral window, and n ₀ is any positive value shorter than the pitch period t. In addition, in the four embodiments described above,
Regarding DFT calculation, Fast Fourier transform (Fast Fourier transform) is used as an algorithm for high-speed calculation.
Transform (hereinafter referred to as FFT) is known, so if FFT is used, the calculation speed will be significantly reduced. Furthermore, when reproducing the structural spectrum of the input signal from the cepstrum, both the pitch spectrum and the envelope spectrum were used in the above embodiment, but only the envelope spectrum was used, excluding the pitch spectrum. You may also perform playback. In this way, since the information regarding the pitch is transmitted while being included in the difference series, the pitch detection circuit is not required, and it is possible to prevent the pitch detection circuit from erroneously detecting the pitch information. As described above, according to the present invention, since the difference between the orthogonal transform coefficients and the spectrum reproduced from the side information is transmitted, the transmitted information is halved compared to the conventional method that transmits the orthogonal transform coefficients. It is possible to obtain a high-quality audio signal even at low transmission speeds of 9.6K bits/second or less, where the sound quality deteriorated with conventional methods.

[Brief explanation of the drawing]

Figure 1 shows an example of the conventional adaptive transform coding method, VD-ATC (Vocoder Diwen
Figure 2 is a diagram showing an example of a window function, Figure 3 is a diagram showing an example of a signal spectrum when noise shaping is applied, and Figure 4 is a block diagram of the Adaptive Transform Coding method.
Figure 5 is a block diagram showing a first embodiment of the present invention, Figure 5 is a diagram showing an example of a signal spectrum representing the spectral structure of an audio signal, and Figure 6 is a block diagram showing a second embodiment of the present invention. , FIG. 7 is a block diagram showing a third embodiment of the invention, FIG. 8 is a block diagram showing a fourth embodiment of the invention, and FIG. 9 is a block diagram showing a fourth embodiment of the invention.
(a) to (i) are waveform diagrams illustrating the operation of the device in FIG. 4; In the figure, 10... transmitting side input terminal, 15...
... Buffer memory circuit, 20 ... Window function circuit, 3
0,300... Inverse dispersive cosine transform (DCT) circuit, 40... Autocorrelation calculation circuit, 50... Pitch detection circuit, 60... Quantizer, 65... Inverse quantizer, 70... Prediction parameter calculation circuit, 80
. . . Parameter conversion circuit, 90 . . . Spectrum regeneration circuit, 100 . Side information encoder, 140... Transmission side output terminal, 150... Receiving side input terminal, 160, 430, 640... Demultiplexer, 170, 440... Side information decoder, 175... Parameter conversion circuit, 180... …
Spectrum regeneration circuit, 190... Bit allocation step size calculation circuit, 200... DCT coefficient adaptive decoder, 210,500... Inverse discrete cosine transform (IDCT) circuit, 220... Buffer memory circuit, 230... Receiving side Output terminal, 310
... Absolute value circuit, 320 ... Logarithm calculation circuit, 33
0,335,670...Inverse discrete Fourier transform (IDFT) calculation circuit, 340,700...Lifter circuit, 345...Pitch detection circuit, 350...Quantizer, 355...Decoder, 360,365, 4
50,455,600...Discrete Fourier Transform (DFT) calculation circuit, 37,490...Exponent calculation circuit, 380...Subtractor, 390...Adaptive quantization encoder, 400,460...Bit allocation quantum quantization step size calculation circuit, 470...adaptive decoder, 480...adder, 610...absolute value phase calculation circuit, 620...phase quantization encoder, 650
A phase decoder, 660 . . . , a conversion circuit is shown, respectively.

Claims (1)

[Claims] 1. In an adaptive audio signal transmission method that adaptively quantizes and transmits an orthogonal transform sequence obtained by orthogonally transforming a sample value sequence obtained by sampling an audio signal, a first generating a signal sequence, generating a second signal sequence representing a characteristic of the first signal sequence based on the first signal sequence, and generating a third signal sequence representing a spectrum based on the second signal sequence; generating a signal sequence, calculating a difference sequence between the first signal sequence and the third signal sequence, generating a fourth signal sequence by adaptively quantizing the difference sequence; and the fourth signal sequence are transmitted in combination.