JP2001230675A - Method for hierarchically encoding and decoding acoustic signal - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce the effects of a transmission error or packet loss by deforming an error spread over hierarchies, in bit rate scalable encoding and decoding, based on hierarchical quantization such as controlling of an amplitude. SOLUTION: A signal, deformed into input signal by a first deformation function, is inputted to the quantizer of the first hierarchy, a quantized code and a signal reconstituted by inverse quantization are outputted, the error signal calculator of the first stage outputs an error d1 (f) between the input signal and the signal reconstituted on the first stage, the quantizer of the next hierarchy inputs a signal deformed into error signal d1 (f) by a second deformation function and outputs the quantized code of that hierarchy and the signal reconstituted by inverse quantization, the error signal calculator of the second stage outputs an rear d2 (f) between the sum of the signal reconstituted on the first hierarchy, and the signal reconstituted on the second hierarchy and the input signal and the errors deformed on the high-order hierarchies, are similarly successively quantized on multiple stages.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a high-efficiency signal encoding method and an encoding method for digitally encoding an audio signal such as a voice or a musical sound signal with a minimum amount of information. TECHNICAL FIELD The present invention relates to a hierarchical encoding method and an encoding method capable of reconstructing voices and musical sounds when there is an error.

[0002]

2. Description of the Related Art Hierarchical scalable encoding and decoding of a signal in the related art has a configuration as shown in FIG. The encoder includes a plurality of layers of quantizers and an error signal calculator between each layer to quantize an input signal vector, and the lowest layer quantizer receives the input signal x (f) as an input, A quantized code (compressed bit string) and a signal x ＾ (f) reconstructed by inverse quantization are output, and the first-stage error signal calculator calculates an error d ₁ between the input signal x (f) and x ＾ (f). (f) (= x (f) −x ＾ (f)), and the quantizer of the next layer receives d ₁ (f) as input and re-quantizes it by the quantization code of the layer and inverse quantization. Composed signal d ₁
＾ (f) is output, and the second-stage error signal calculator calculates the input signal x
(f) and the error of the signal obtained by adding d ₂ ＾ (f) d ₂ (d ₂ = x (f) − (d ₁ ＾ (f)
+ X ＾ (f))), and similarly, the upper-layer-deformed error is sequentially quantized in multiple stages.

In a decoder, a signal is reconstructed by inversely quantizing the quantization code of each layer, and a final signal is reproduced by addition. Assuming that the output y ₁ (f) of the first stage, the output y ₂ (f) of the second stage, and the quantization error of each layer are q ₁ (f) and q ₂ (f), y ₁ (f) = x ＾ (f) = x (f) + q ₁ (f) (1) y ₂ (f) = y ₁ (f) + d ₁ ＾ (f) = x ＾ (f) + x (f) −x ＾ (f ) + Q ₂ (f) = x (f) + q ₂ (f) (2) In this case, errors in lower layers are all fed back to higher layers, and higher layers can reproduce higher quality signals. Further, a signal corresponding to the bit rate can be reproduced even without a signal of a higher layer, which is a feature of the scalable coding. However, when there is an error or a frame loss in the bit sequence of the lower layer, the output y _2- (f) is d ₁ ＾ (f)
Only the result ^{_{y 2 - (f) = d}} 1 ^ (f) = x (f) -x ^ (f) + q 2 (f) = -q 1 (f) + q 2 (f) (3) x (f) A large error occurred, and it was not possible to rescue the upper hierarchy.

[0004]

In the hierarchical scalable coding and decoding methods described in the prior art, if there is an error or a frame loss in a bit string, it is difficult to reconstruct speech or musical sound. . INDUSTRIAL APPLICABILITY The present invention can transmit musical sounds and voices with the highest possible quality in accordance with the environment of a network or a decoder, and is particularly suitable for hierarchical transmission of audio signals suitable for wireless transmission in which transmission bits may have errors. It is an object to provide a method and an encoding method.

[0005]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, the present invention provides a bit rate scalable coding by hierarchical quantization, and performs a transmission error by performing a modification such as an amplitude control on an error across layers. And the effect of packet loss.

[0006]

FIG. 2 is a block diagram for explaining processing in an encoder and a decoder according to an embodiment of the present invention.
In the encoder, in order to quantize the input signal vector, a quantizer Q ₁ of the lowest hierarchy receives a signal obtained by multiplying the input signal x (f) by α _1, and performs a quantization code (compressed bit sequence) and inverse quantization. Q ₁ ^-1
And outputs a signal [α ₁ x (f)] ＾ reconstructed by Here, α ₁ is a constant of 1 or less. That is, in the first stage of inverse quantization, the input signal is not used as it is as an input, but a modified signal is used. Here, an example of multiplication by a constant is used, but processing of a coefficient or filter depending on frequency may be used.

Next, the first-stage error signal calculator calculates [α ₁ x
(f)] ＾ and the error d ₁ (f) d ₁ (f) = x (f) − [α ₁ x (f)] ＾ (4) between the input signal x (f) and (4) is output. The quantizer of the next layer receives a signal obtained by multiplying d ₁ (f) by α _2, and receives a quantization code of the layer and a signal reconstructed by inverse quantization (α ₂ d ₁ (f)) ＾. Output. The second-stage error signal calculator calculates an error d ₂ (d ₂ = x (f) between the signal obtained by adding the modified g ₂ () to x ＾ (f) and d ₁ ＾ (f) and the input signal x (f). ) −g ₂ (x ＾
(f), d ₁ ＾ (f))), and similarly, a signal transformed into an upper layer is input, and quantization is sequentially performed in multiple stages. This modification is not necessary in the last layer.

[0008] When there is no code error or frame erasure, the decoder dequantizes the quantization code of each layer to reconstruct a signal. y ₁ (f) = [α ₁ x (f)] ＾ = α ₁ x (f) + α ₁ q ₁ (f) (5) y ₂ (f) = y ₁ (f) + [α ₂ d ₁ ( f)] ＾ = [α ₁ x (f)] ＾ + α ₂ (x (f) −x ＾ (f) + q ₂ (f)) = (1−α ₂ ) [α ₁ x (f)] ＾ + α _{2 (x (f) + q} 2 (f)) = (α 1 + α 2 -α 1 α 2) x (f) + α 1 (1-α 2) q 1 (f) + α 2 q 2 (f) (6 ) component of the input signal component alpha ₁ next to the input signal included in the output y ₁ (f) of the first stage, and the output up to the second stage, which is contained in the y ₂ (f) is (alpha ₁ + alpha ₂ - α ₁ α ₂ ) times. As a result, q
The quantization error of ₁ (f) and q ₂ (f) becomes relatively large, and the SNR becomes lower than in a normal case not including α.

If there is an error or loss of the quantization code of a certain layer, the signal of that layer is set to 0 and the signal is reconstructed. If an error detection code can be obtained as transmission line information, it may be used. If not, an error detection code may be attached to each frame or each layer. In the absence example information of the first layer, the output signal y ₂ only the second layer ^- (f) is the ^ [alpha ₁ d ₁ (f)]. ^{_{y 2 - (f) = α}} 2 ( [α ₁ d ₁ (f)] _{^ + q 2 (f))} = α 2 α 1 x (f) -x ^ (f) + α 2 q 2 (f) = α ₂ (1−α ₁ ) x (f) −α ₁ α ₂ q ₁ (f) + α ₂ q ₂ (f) (7) As can be seen from the above, even if there is no output from the first layer, α ₂ ( Since the component of the input signal multiplied by 1−α ₁ ) is included in the output of the second layer, the SNR is higher than that of the conventional method when compared under the condition that the first layer is not present. Similarly, the components of the input signal are also distributed to the higher layers, and the first layer
The lack of hierarchy can be compensated to some extent.

In the embodiments described so far, the transformation of the quantization input is made a constant multiple, but it can be extended to a processing depending on the frequency. FIG. 3 shows a second embodiment of the present invention. Similar to FIG. 2, but the signal transformation (here a constant multiple) is performed after the output signal by inverse quantization. The same effect as in the first embodiment is obtained, but the reproduced signal on the decoder side is slightly different.

Y ₁ (f) = x ＾ (f) = x (f) + q ₁ (f) (8) The output of the second layer is y ₂ (f) = y ₁ (f) + d ₁ ＾ (f) _{= x (f) + q 1} (f) + x (f) -α 1 (x (f) + q 1 (f)) + q 2 (f) = (2-α 1) x (f) + (1-α 1 ) q ₁ (f) + q ₂ (f) (9)

[0012] When there is no information in the first layer, the output signal y ₂ only the second layer ^- (f) becomes d ₁ ^ (f). ^{_{y 2 - (f) = d}} 1 ^ (f) = x (f) -α 1 (x (f) + q 1 (f)) + q 2 (f) = (1-α 1) x (f) -α ₁ q ₁ (f) + q ₂ (f) (10) The effects and extensions are the same as in the first embodiment, and the two embodiments can be combined. In this embodiment, since the coefficient of x (f) in equation (9) is (2-α ₁ ), the normal volume of the second layer is higher than it should be. Also, when the output of the first layer cannot be used as in the equation (10), the volume decreases because the coefficient is (1−α ₁ ). In the case of a normal decoder, since it is known in advance which layer of information is to be used, it is sufficient to modify the output by multiplying the output by a correction constant so that the volume becomes the original x (f). However, if the decoder is fixed and chipped according to an international standard or the like, this correction cannot be made, so it is necessary to allow a difference in volume or to separately correct the volume.

[0013]

According to the scalable coding of the present invention, since the entire system becomes redundant, when compared with the same number of bits, the quantization distortion in the case where there is no error is larger than that of the conventional method. , So that even if information in any layer is lost, the damage is reduced.

[Brief description of the drawings]

FIG. 1 is a diagram showing a basic configuration of a conventional hierarchical encoder and decoder.

FIG. 2 is a diagram showing a configuration of a hierarchical encoder and a decoder in the first embodiment of the present invention.

FIG. 3 is a diagram showing a configuration of a hierarchical encoder and a decoder according to a second embodiment of the present invention.

 ──────────────────────────────────────────────────続 き Continuation of the front page (72) Inventor Takeshi Mori 2-3-1 Otemachi, Chiyoda-ku, Tokyo Within Nippon Telegraph and Telephone Corporation (72) Inventor Akio Kami 2-chome Otemachi, Chiyoda-ku, Tokyo No. 1 Nippon Telegraph and Telephone Corporation F term (reference) 5D045 DA11 5J064 AA01 BA01 BC08 BC16 BD01 5K041 AA01 CC01 EE38 FF27 9A001 EE04 KK43

Claims (9)

[Claims] An encoding method for encoding an audio signal on a frame-by-frame basis, comprising: a plurality of layers having a deformation function and a quantizer; an error signal calculator between each layer; Of the input signal x (f)
The signal deformed by g ₁ () is input, and a signal g ₁ ＾ (x) (f) reconstructed by quantization code and inverse quantization is output. The first-stage error signal calculator calculates the input signal x ( error d ₁ (f) (= x (f) −g ₁の) between signal f _{1 and} signal g ₁ ＾ (x) (f) reconstructed in the first hierarchy
(x) (f)), and the quantizer of the next layer receives as input the signal transformed by the transformation function g ₂ () into the error signal d ₁ (f),
The signal g ₂ ＾ (d ₁ ) (f) reconstructed by the inverse quantization is output, and the error signal calculator of the second stage outputs the signal g ₁ reconstructed in the first layer.
The error d ₂ (f) (= x (f) −) between the sum of ＾ (x) (f) and the signal g ₂ ＾ (d ₁ ) (f) reconstructed in the second hierarchy and the input signal x (f) (G ₁ ＾ (x) (f) +
g ₂ ＾ (d ₁ ) (f))), and similarly, sequentially performs multistage quantization of the upper layer-deformed error, and outputs a quantization code of each layer. Hierarchical coding method for signals. 2. An encoding method for encoding an audio signal on a frame basis, comprising: a plurality of hierarchies having a deformation function and a quantizer; and an error signal calculator between the hierarchies. Receives the input signal x (f), outputs a signal x ＾ (f) reconstructed by quantization code and inverse quantization, and the first-stage error signal calculator calculates the input signal x (f ) and x ^ (f) the transformation function g ₁ (the deformed signal) the error signal _{d 1 (f) (= x} (f) -g
₁ (x ＾ (f))), and the quantizer of the next layer receives d ₁ (f) as an input, and the signal d ₁ ＾ (f ), And the second-stage error signal calculator calculates the transformation function g as x ＾ (f) and d ₁ ＾ (f).
₂ The error d ₂ (f) between the signal transformed in () and the input signal x (f) (= x
(f) −g ₂ (x ＾ (f), d ₁ ＾ (f))), and similarly, the higher-layer-deformed error is sequentially quantized in multiple stages, and the quantization code of each layer And a hierarchical encoding method of the acoustic signal. 3. The hierarchical encoding method of an acoustic signal according to claim 1, wherein an object to be quantized is a transform coefficient in a frequency domain. 4. A method according to claim 1, wherein the quantization target is a signal in the time domain. 5. The hierarchical coding method of an acoustic signal according to claim 1, wherein the transformation function (g ₁ (), g ₂ (),...) Is a constant of 1 or less. A hierarchical encoding method of an acoustic signal, characterized in that: 6. A method according to claim 3, wherein the transformation function (g ₁ (), g ₂ (),...) Is a frequency-dependent weight function. Encoding method of the audio signal to be encoded. 7. A method according to claim 4, wherein the transform function (g ₁ (), g ₂ (),...) Uses a filter whose frequency characteristic changes. Hierarchical encoding method for audio signals. 8. A method for decoding an audio signal on a frame basis, wherein the quantization code of each layer generated by the audio signal hierarchical encoding method according to claim 1 is provided. An inverse quantization process of inputting an error detection code for the quantization code, inversely quantizing the quantization code of each layer and outputting an inversely quantized signal, and determining whether or not an error exists in the quantization code of each layer based on the error detection code. And reconstructing an audio signal by adding an inverse quantization signal to a quantization code in which no error is detected, and decoding the audio signal hierarchically. 9. The hierarchical decoding method of an audio signal according to claim 8, further comprising means for correcting a volume by applying a correction constant to the audio signal output from the step of reconstructing the audio signal. A method for hierarchical decoding of an acoustic signal.