JPS6342532A - Audio encoding device - Google Patents

Abstract

PURPOSE:To considerably improve the information compression ratio by encoding a low band component and using a representative value in every prescribed audio section of respective time series data of plural filter coefficients obtained from an adaptive equalizing filter as auxiliary information for high band reproducing. CONSTITUTION:When an audio signal X(n) is inputted, a thinning LPF 111 eliminates every other sample point of the audio signal X(n) to generate a low band audio signal XL. An encoding part 112 encodes the low band audio signal XL to generate an encoded low band audio signal EL. When an approximated audio signal X1'(n) is inputted to an ADF 121, the ADF 121 outputs time series data of each filter coefficient, namely, auxiliary information Hau for high band reproducing, and this information is stored in a coefficient memory 131 in frame units. A coefficient converting part 133 converts and quantizes inputted representation auxiliary information Au to a known LSP parameter tolerant of quantization. Thus, information whose quantity is considerably larger than that in a conventional system is compressed to improve the transmission rate.

Description

[Detailed Description of the Invention] [Summary] In a speech coding system that requires compensation of high-frequency components,
The low-frequency component is encoded, and the representative value for each predetermined speech section of each time-series data of a plurality of filter coefficients obtained from the adaptive equalization filter is used as auxiliary information for high-frequency reproduction. As a result, the information pressure compression ratio can be increased (improved), and high frequency components can also be reproduced satisfactorily.

[Industrial application field]

The present invention provides an audio encoding device, in particular, divides an audio signal into a low frequency component and a high frequency component, performs encoding processing on the low frequency component, and generates a high frequency component obtained from an adaptive equalization filter for the high frequency component. In a speech encoding device that uses auxiliary information necessary for reproduction, an improved speech code that maintains good quality of the reproduced high-frequency components and efficiently compresses the auxiliary information necessary for high-frequency component reproduction. related to conversion equipment.

[Conventional technology]

Various methods have been proposed for audio encoding, but when encoding audio signals, it is required to encode in a way that does not impair the quality of the audio (i.e., increases the compression ratio). On the other hand, it is well known that the dominant components that determine the quality of voice are mainly in the low frequency band.

Therefore, as a typical audio encoding method that efficiently compresses audio while maintaining audio quality, the audio signal is divided into low frequency components and high frequency components, the low frequency component is encoded, and the high frequency component is encoded. A frequency division type audio encoding system is known in which auxiliary information necessary for reproduction is sent.

As a conventional frequency division type speech encoding method mainly containing low-frequency components, a speech coding method as shown in FIG. 5 has been adopted.

In FIG. 5, 30 is an encoding device and 40 is a decoding device.

In the encoding device 30, 31 is a low frequency encoding means that encodes the low frequency component (low frequency audio signal XL) of the input audio signal X to generate an encoded low frequency audio signal EL.

32 is a high frequency reproduction auxiliary information generating means, which includes an adaptive equalization filter (ADF).
r) 321 and reproduced approximate audio signal X+'
(not shown) and the original audio signal Output as hau. The equalized residual signal of the ADF 321 is formed from time-series data of the sampling period during encoding processing.

The encoding device 30 sends the encoded low frequency audio signal EL and the high frequency reproduction auxiliary information hau generated in this manner to the decoding device 40 as audio encoding information.

In the decoding device 40, reference numeral 41 denotes a decoding/reproducing means, in which the received encoded low frequency signal EL' (the dash indicates the encoder 3
This is to distinguish it from the EL on the 0 side, and the same applies to the next hau'), and the 30 examples of encoding devices approximate it.

42 is an adaptive equalization filter (ADF');
It has the same configuration as the ADF 321 on the 0 side. ADF'
42 corrects each filter coefficient in the ADF' 42 using the received high frequency reproduction auxiliary information hau', and generates an audio signal corresponding to the original audio signal X of the 30 encoding devices from the input approximate audio signal X+. Play k.

As described above, encoding is performed mainly on the low-frequency components of the audio signal, and for the high-frequency components of the audio signal, the equalized residual signal of the ADF is sent as auxiliary information for reproducing the high-frequency components. This makes it possible to thickly compress audio information and to reproduce even high-frequency components favorably.

[Problem that the invention seeks to solve]

The conventional audio encoding method shown in FIG. 5 provides a high information compression ratio and can reproduce even high-frequency components favorably.

However, since the equalized residual signal related to ADF control used as auxiliary information for high-frequency component reproduction is generated in the form of time-series data of the sampling period during encoding processing, the amount of data cannot be reduced. unable to decrease significantly,
Therefore, there is a problem that there is a limit to the amount of information that can be compressed.

In the conventional audio encoding method shown in FIG. An object of the present invention is to provide a speech encoding method that makes it possible to reproduce satisfactorily even the range components.

[Means for solving problems]

The means taken by the present invention to solve the above-mentioned problems in the conventional audio encoding system will be explained with reference to FIG.

FIG. 1 is a block diagram showing the basic configuration of the present invention.

In FIG. 1, numeral 10 is an encoding device that performs the audio encoding process of the present invention.

Reference numeral 11 denotes a low frequency encoding means, which encodes the low frequency component (low frequency audio signal XL) of the input audio signal X to generate an encoded low frequency audio signal EL.

Reference numeral 12 denotes a high frequency reproduction auxiliary information generating means, which is equipped with an ADF, that is, an adaptive equalization filter 121, and is configured to adjust A so that the error between the reproduced approximate audio signal Xr' and the original audio signal X becomes zero.
When equalization processing is performed by DF121, ADF12
Each time series data of a plurality of filter coefficients obtained from 1 is output as high frequency reproduction auxiliary information Hau for high frequency component reproduction.

Reference numeral 13 denotes representative auxiliary information generating means, which obtains a representative value of the high frequency reproduction auxiliary information 1iau for each predetermined audio section and outputs it as a representative auxiliary information sieve for high frequency component reproduction.

[For production]

When the audio signal X is input, the low frequency encoding means 11 encodes the low frequency audio signal XL of the audio signal X to generate an encoded low frequency audio signal EL. This encoding is performed using, for example, the well-known adaptive differential P CM (AdapLive dif
ferential PCM: ADPC
M).

The high frequency reproduction auxiliary information generating means 12 generates an ADF 121 when equalization processing is performed by the eight DFs 12 so that the error between the reproduced approximate audio signal Xl' and the original audio signal X becomes zero.
Each time series data of a plurality of filter coefficients obtained from the filter coefficients is output as high frequency reproduction auxiliary information 91 Hau. ADPCM
When the encoded low-frequency audio signal EL is generated, a reproduced signal of the low-frequency audio signal is generated in the encoding process, so the reproduced low-frequency signal is used to add 1' to the approximate audio signal. It can be played.

As mentioned above, the high frequency reproduction auxiliary information Hau is
The representative auxiliary information generating means 13 obtains the representative value of the high frequency reproduction auxiliary information Hau for each predetermined audio section, and calculates the representative value of the high frequency reproduction auxiliary information Hau for each time series data, and It is output as representative auxiliary information 91Au for component reproduction.

By doing the above, it is possible to reproduce high-frequency components well, and the amount of auxiliary information for high-frequency component reproduction can be significantly increased compared to the conventional method that directly uses the equalized residual signal of the ADF. Therefore, the overall information compression ratio can be greatly improved.

〔Example〕

Embodiments of the present invention will be described with reference to FIGS. 2 to 4.

FIG. 2 is an explanatory block diagram of the configuration of an embodiment of the present invention, FIG. 3 is an explanatory diagram of the ADF, and FIG. 4 is an explanatory diagram of the signal spectrum and signal waveform of the embodiment.

(A) Configuration of the embodiment In FIG. 2, an encoding device 10, a low-frequency encoding means 11,
The high frequency reproduction auxiliary information generating means 12, ADF 121, and representative auxiliary information generating means 13 are as described in FIG.

In the low frequency encoding means 11 of the encoding device 10, 111 is a decimation low pass filter (decimation LPF) which generates a low frequency audio signal XL by performing sampling conversion.

The encoder 112 encodes the low frequency audio signal XL to generate an encoded low frequency audio signal EL.

In the high frequency reproduction auxiliary information generating means 12, 122 is an approximate audio reproducing section, which generates an approximate audio signal Xl that approximates the audio signal X.
'Play.'

An adder 123 generates an error signal e between the original audio signal X and the audio signal X reproduced by the ADF 121.

In the representative auxiliary information generating means 13, 131 is a coefficient memory in which high frequency reproduction auxiliary information 1 inputted from the ADF 121 is stored.
au is stored in units of predetermined audio sections.

Reference numeral 132 is an averaging unit that calculates the representative value of the high frequency reproduction auxiliary information fl Hau, that is, the representative value of the high frequency reproduction auxiliary information for each predetermined audio section from the average value for each predetermined audio section of the time series data of each filter coefficient. To 〇.

133 is a coefficient conversion unit which converts a known line spectrum pair (L S
P: Line spectrum pair
) The representative auxiliary information Au is converted and quantized by conversion processing.

On the right side of FIG. 2 is a decoding device 20, which performs a process of reproducing an audio signal from received encoded audio information.

In the decoding device 20, reference numeral 21 denotes a decoding unit, in which the received encoded low-frequency audio signal EL' (the dash indicates E on the encoding device 10 side
This is added to distinguish it from L, and Au′ which will be explained next.
) is decoded to reproduce the low-frequency audio XL corresponding to the low-frequency audio signal XL of the 10 encoding devices.

Reference numeral 22 denotes an approximate speech reproduction unit which generates an approximate sound of the 10 encoding devices.

23 is a coefficient inverse transformation unit which receives the LSP-converted representative auxiliary information and converts it to the original representative auxiliary information! Generate IaAu'.

24 is an adaptive equalization filter (ADF');
It has the same configuration as the ADF 121 on the 0 side, and reproduces the audio signal X corresponding to the original audio signal X from the input representative auxiliary information Au' and the approximate audio signal 'X+.

FIG. 3 shows an example of the ADF'121.

In ADF 121 in FIG. 3, 121al to 121
a* is a delay element represented by a Z-transform symbol, which causes a time delay equal to the sampling period.

12 l b+ to 12 l b+ are filter coefficient setting units that set filter coefficients a14a. 121C is an adder which adds the outputs of the filter coefficient setting section 121b+"121b4 to generate the audio signal X (nl).

121d is an adder which generates a sum signal S(nl) of this audio signal X(n) and the approximate audio signal X+'(nl. Note that (nl is the value at sample point n).

121e is a coefficient control unit that controls the filter coefficient setting unit 12 so that the error signal e (nl) output from the adder 123 becomes zero.
1bi to 121b, and the values of + filter coefficients a1 to a are sequentially controlled.

Each time series data of each variably controlled filter coefficient al-84, that is, 1lau, which is high frequency reproduction auxiliary information used in the present invention, is outputted as time series data of each sampling period to the representative auxiliary information generation means 13. Added. (Note that the equalized residual signal output from the adder 123 is high frequency reproduction auxiliary information used in the conventional method shown in FIG. 5.) (B) Operation of the embodiment This will be explained with reference to the subekram and signal waveforms in FIG.

In the embodiment described below, it is assumed that the predetermined voice section is a short frame of several mφC (5m5ec in the example) during which the voice parameters do not seem to change substantially, and the sampling frequency fs is 6. .4
Suppose that it is kHz. Therefore, there are 32
There is a sample (=6.4!(zx5msec)).

The input audio signal X is 0 as shown in Figure 3 (al).
-f s/2 (0 to 3.2kHz), and its signal waveform is 1 as shown in the figure (dl).
It is in the form of a sampling signal sampled 32 times per frame. Below, the amplitude value at sample point n is X(n): (n=1 t 2.3 t-+32)
Let us express it as This also applies to other signal waveforms.

(B-1) Operation audio signal X of encoder 5 (When nl is input, thinning LPF 1
11 performs a process of thinning out every other sample point of the audio signal X (n) to generate a low frequency audio signal XL. The low-frequency sound signal XL generated in this way has a spectrum characteristic of O=f s/4 (0 to 1.6 kHz), as shown in Figure 3), and its signal waveform is the same. As shown in the figure (el), sampling signal XL (2n) sampled 16 times per frame: (2n=2,4
,6,...,32).

The encoding unit 112 encodes the low frequency audio signal XL to generate an encoded low frequency audio signal E. This encoding process is
This is performed using the well-known ADPCM encoding method. The generated encoded low frequency audio signal E is a prediction residual signal of the low frequency audio signal XL.

As a result, an encoded low-frequency audio signal EL having a small number of bits, that is, a high information compression ratio is obtained, and the original low-frequency audio signal can be reproduced satisfactorily.

In the ADPCM encoding method, in the encoding process, the encoded low frequency audio signal EL is decoded, and the low frequency audio signal corresponding to the low frequency audio signal XL reproduced on the decoding device 20 side is generated. XL' is played. This reproduced low frequency audio signal XL' was input to the approximate audio reproduction section 123.

The approximate audio reproduction section 122 uses this reproduced low-frequency audio signal XL'
When received, the amplitude information at the sample points thinned out by the thinning LPFIII is supplemented by interpolation processing, and an approximate audio signal X+' corresponding to the approximate audio signal X+ reproduced on the decoding device 20 side is reproduced. Approximate audio signal X1
' is 0 to f s/2 (as shown in Figure 3 (C1)
It has a spectrum characteristic of 0 to 3.2kHz), and
As shown in l, the sampling signal X1' (n) sampled 32 times per frame: (n-1, 2,
3, ++, 32).

When the approximate audio signal X+'(n) is input, the ADF 121 outputs the audio signal X (n) and adds it to the adder 123. The adder 123 outputs an error signal e (nl) between this audio signal X (n) and the original audio signal
Return to 121. The coefficient control section 121e of the ADF 121 sets the filter coefficients a1(nl: (i=1, 2, 3) of the filter coefficient setting sections 12 lb1 to 12 lb4 so that the input error signal e , 4
) is sequentially updated and controlled by, for example, a coefficient update equation shown in the following equation +11.

ai (n + 1) = ai (n) *γ+εHsg
n (e (nil honor sgn (S (n -1)
)...(1) Here, sgn (X): is called a sign function and is passed as follows, and s (n) is the sum signal output from the adder 121d, T
is the leakage coefficient for transmission errors, ε is the update rate coefficient,
Each square indicates multiplication (same as X).

Filter coefficient a i variably controlled by this sequential control
(n) (i = 1 to 4) is 32 per frame
Sample existing time series data a 1 (J): (i
=1.

2.3,4; j=1,2,3. ..., 32), and becomes time series data of each filter coefficient, that is, high frequency reproduction auxiliary information 11au.

High frequency reproduction auxiliary information 11a output from ADF 121
u is stored in the coefficient memory 131 in units of frames.

The averaging unit 132 calculates the high frequency reproduction auxiliary information! according to the following formula.
Each filter coefficient ai (i=1, 2°3.
4) Find the average value per frame of the time series data and set it as the respective representative value Ai (i-1, 2, 3°4). That is, 32j=1 i=1, 2, 3.4 The averaging unit 132 calculates each representative value Ai obtained in this way.
(i=1 to 4) as representative auxiliary information A for high frequency component reproduction.
It is added to the coefficient conversion unit 133 as u.

The coefficient conversion unit 133 converts and quantizes the input representative auxiliary information Au into known LSP parameters that are resistant to quantization.

The encoded low-frequency audio signal E generated by the encoder 112 and the representative auxiliary information subjected to LSP conversion! iAu is sent to the decoding device 20 as audio encoded information.

In that case, the encoded audio information may be multiplexed by a multiplexing device and then transmitted to the decoding device 20 via a transmission path (none of which is shown).

Next, the amount of information necessary for transmitting the encoded audio information generated as described above will be specifically explained in comparison with the conventional method shown in FIG.

In the audio encoding system according to the present invention, the encoded low-frequency audio signal EL, which is obtained by quantizing the prediction residual of the low-frequency audio signal XL (2n), is 2 bits x 16 samples = 32 bits. Also, the frame synchronization is 1 bit, and the auxiliary information %iAu for reproducing high frequency components converted to LSP is 3 bits x 4th order = 1
It is 2 bits. Therefore, the amount of information required for the audio encoding method of the present invention is 45 bits in total. In that case, the frame length is 5m5ec, so the transmission rate of audio encoded information is 45 bits/5m5ec=9.0 kb
ps.

On the other hand, in the case of the conventional method shown in Fig. 5, under the same conditions, the encoded low frequency audio signal E is also 32 bits, but the high frequency reproduction information (Au) for high frequency component reproduction is Even if a certain equalized residual signal (a(1): i = 1 to 32) is converted to a minimum of 1 bit, it requires 1 bit x 32 = 32 bits, so adding 1 bit of frame synchronization makes the total minimum It will be 65 bits. Therefore, the transmission rate of the encoded information is at least 65 bits 15m5s=13 kb
It is pS.

As described above, according to the audio encoding method of the present invention, the amount of information can be compressed to a greater extent than the conventional method, and the transmission rate can be increased.

(B-2) Operation of the decoding device When the decoding device 20 receives encoded audio information in a multiplexed manner, the encoded low-frequency audio signal EL' and LSP are converted into the encoded low-frequency audio signal EL' by a multiplexing decoding device (not shown). The representative auxiliary information Au' for high-frequency component reproduction is separated and extracted, the encoded low-frequency audio signal EL' is inputted to the decoder 21, and the LSP-converted representative auxiliary information Au' is sent to the coefficient inverse transformer 23. input.

The decoding unit 21 decodes the input encoded low frequency audio signal EL' to generate a low frequency signal corresponding to the low frequency audio signal XL (see FIGS. 4(b) and (e)) of the 10 encoding devices. audio signal
Play L.

When the approximate audio reproduction unit 22 receives the reproduced low-frequency audio signal XL, it supplements the amplitude information at the thinned out sample points by interpolation processing, and reproduces the approximate audio signal X1' of the 10 encoding devices (Fig. 4). An approximate audio signal X+ corresponding to (see (C) and (f)) is generated and input to the ADF' 24.

On the other hand, the coefficient inverse transform unit 23 performs LSP=coefficient inverse transform on the added LSP-transformed representative auxiliary information Au' to generate representative auxiliary information Au' for original high frequency component reproduction, ADF' 24 is manually operated.

The ADF'24 updates each frame coefficient value of its filter coefficient setting section for each frame based on the input representative auxiliary information Au', and updates the frame coefficient value of the filter coefficient setting section based on the inputted approximate audio signal X+.
An audio signal X corresponding to the original audio signal X (see FIGS. 4(al and d)) is reproduced.

In this way, one of the high frequency components of the reproduced audio signal
The average amplitude per frame is the same as the average amplitude per frame of the high-frequency components of the original audio signal It can be reproduced well with little spectral distortion.

Although one embodiment of the present invention has been described above, the structure of the present invention is not limited to each structure of this embodiment.

Thinning LPF III can perform thinning processing using a thinning rate of an integer greater than 2.

In that case, the interpolation rate of the approximate audio reproduction units 122 and 24 is also
Correspondingly, it is set to an integer greater than 2.

Further, the coefficient update process of the ADF 121 can be performed using various known coefficient update formulas in addition to the above-mentioned formula (11).

〔Effect of the invention〕

As explained above, according to the present invention, the following effects can be obtained.

(b) It becomes possible to significantly compress the amount of auxiliary information necessary for reproducing high-frequency components of audio signals, and the overall information compression ratio can be increased (improved).

(b) Since the average amplitude per frame of the high-frequency components of the reproduced audio signal is the same as the average amplitude per frame of the high-frequency components of the original audio signal, the high-frequency components also have a small spectrum. It can be reproduced well with distortion.

[Brief explanation of the drawing]

Fig. 1...Explanatory diagram of the basic configuration of the present invention, Fig. 2...
An explanatory diagram of the configuration of an embodiment of the present invention, Fig. 3: An explanatory diagram of the adaptive equalization filter (ADF) of the embodiment, Fig. 4: Explanation of the signal spectrum and signal waveform of the embodiment Fig. 5: An explanatory diagram of a conventional audio encoding method. In FIG. 1 and FIG. 2, 10... encoding device, 11... low frequency encoding means, 12
. . . High frequency reproduction auxiliary information generation means, 121 . . . Adaptive equalization filter (ADF), 13 . . . Representative auxiliary information generation means, 20 .
... Approximate audio reproduction section, 23... Coefficient inverse transformation section, 24.
...Adaptive equalization filter (ADF').

Claims (2)

[Claims] (1) Encode the low-frequency components of the input audio signal (X) to generate an encoded low-frequency audio signal (EL), and perform equalization processing using an adaptive equalization filter to reproduce the high-frequency components. A speech encoding device that outputs frequency reproduction auxiliary information includes (a) an adaptive equalization filter (121), and converts each time series data of a plurality of filter coefficients obtained from the adaptive equalization filter (121) into high frequency components. High frequency reproduction auxiliary information generating means (Hau) for outputting as high frequency reproduction auxiliary information (Hau)
(b) Find the representative value of the high frequency reproduction auxiliary information (Hau) for each predetermined audio section and obtain the representative auxiliary information (Au) for high frequency component reproduction.
A speech encoding device comprising representative auxiliary information generating means (13) for outputting as follows. (2) The auxiliary information generating means (13) generates a line spectrum pair (
2. The audio encoding device according to claim 1, wherein the audio encoding device outputs auxiliary information (Au) that has been subjected to LSP) conversion processing.