JPS58215822A - Predictive encoder of voice signal - Google Patents

Abstract

PURPOSE:To simplify the constitution of a titled device, by constituting a predictive circuit for the encoder encoding a difference signal between an estimated instantaneous value and a present instantaneous value of a lattice type digital filter and an adder summing its outputs, for performing stable predictive processing to a voice signal. CONSTITUTION:A voice signal analyzing circuit obtains a characteristic parameter signal of a voice signal from the past sampling value of the voice signal and inputs the parameter signal to the predictive circuit to obtain the estimated predictive value to the voice signal. This predictive circuit is constituted of the lattice type digital filter 5c and the adder 5d obtaining the total sum of the tap outputs of the filter 5c. A decoded difference signal deltan'' inputted by the filter 5c is transmitted sequentially via adders 21-1-21-8, and the output is transmitted through delay circuits 22-8-22-1 and adders 23-8-23-2 alternately. The output of the adders 21-2-21-8 is applied to multipliers 24-2-24-8 to multiply the autocorrelation coefficient, which is applied to the adders 21-8-21-1 for making the predictive processing stable.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to a predictive encoding device for an audio signal that can efficiently encode an audio signal at a low bit rate.

[Technical backbone of invention]

An adaptive predictive coding device configured as shown in Figure 1 is known as a device for efficiently predictive coding an audio signal at a low bit rate.

This device inputs a sampled input speech signal to a speech analysis circuit 1 and extracts its characteristic parameter P.
The signal is led to a subtracter 3 via a delay circuit 2. The audio analysis circuit 1 extracts, for example, a parameter α indicating a vector envelope in a short time period of an audio signal, a characteristic parameter such as pitch frequency fp1 power, etc. from a plurality of past sample values of the audio signal. The feature parameter signal is supplied to an output circuit 4, which will be described later, and is also supplied to a prediction filter 5a forming a prediction circuit 5.

On the other hand, the voice signal Sn delayed through the delay circuit 2
is led to subtractor 3. The prediction circuit 5 inputs the feature/flat meter signal and estimates the predicted instantaneous value Sn for the audio signal and the audio signal Sn.
O difference signal δ. is required. This difference signal δ. is encoded by the encoding circuit 6, resulting in an encoded difference signal δ. ' is supplied to the output circuit 4 as the characteristic parameter signal P
It is configured to be multiplexed and transmitted. Therefore, the prediction circuit 5 receives the encoded difference signal δ. ' is decoded via the decoding circuit 7 to produce a decoded difference signal δ. 〃 and the estimate/\ constant instantaneous value (predicted value) Sn are added by an adder 5b,
The output rn is provided to the prediction filter 6hK. The prediction filter 5a uses the output γ1 and the feature J? as described later. The estimated instantaneous value Sn is determined according to the parameter signal P.

Now, a receiving and reproducing circuit that receives the predictively encoded and transmitted signal, decodes it, and reproduces the signal is configured as shown in FIG. 2, for example. That is, the received signal is encoded by the distribution circuit 8 into an encoded difference signal δ. ' and a feature parameter signal P, and this feature parameter signal P is supplied to a prediction filter 9a of a prediction circuit 9. Also, the encoded difference signal δ. ' is decoded via the decoding circuit 10 and supplied to the prediction circuit 9 as a decoded difference signal δn#. This prediction circuit 9 generates the decoded difference signal δ. 〃 is guided to the adder 9b and added to the prediction signal sn output from the △ prediction filter 9a,
It is output as a reproduction output signal Rn. This output signal Rn
corresponds to the sampled value sn of the audio signal.

The above output signal Rn is input to the prediction filter 9a, and a prediction signal sn is created according to the characteristic parameters and the Δ signal P.

A digital transmission system for audio signals consisting of a pair of such a predictive coding device and a receiving/reproducing circuit is capable of transmitting audio signals with low bit rate and good sound quality, as detailed in Japanese Patent Publication No. 50-21203. Predictive coding transmission can be performed.

By the way, the prediction filter 5& constituting the prediction circuit 6 predicts the pitch period of the audio signal and the instantaneous value from a plurality of past sandal values.
When performing linear prediction of the instantaneous value from sample values, the prediction filter 5a is generally constructed as shown in FIG. That is, the prediction filters 5h and 9m are of a so-called transformer-caesal type, and include a plurality of stages (eight stages) of delay circuits 11.
~+ l I Decoded difference signal δ sequentially passed through J-2 to Horn-8. ′ is output from each tap by the coefficient unit 12 12~
12-8, the characteristic parameters 41-2 are multiplied by coefficients αl and α8 to α8 based on the characteristic parameter 41-2, and these outputs are added and combined through an adder 13 and output.

[Problems with leisure technology]

However, in the conventional prediction circuit 5 in which the prediction filter 5a is configured using a transformer type filter as described above, the transfer function of the prediction filter 5a is
In the case of (2), the signal γ is obtained from the decoded difference signal δ□〃. The transfer function of the prediction circuit 5゜9 itself that creates 1/(]-F(
Z)). However, such a transfer function is not always stable due to its finite calculation accuracy, and there is a risk of oscillation. Also α・(
The amount of calculation required to calculate the parameters is large. In order to solve these problems and construct a stable prediction circuit 5゜9, it is necessary to increase the number of taps mentioned above to improve performance and accuracy, which ultimately leads to an increase in the scale of the device and This resulted in an increase in power consumption, a rise in device costs, etc., and had problems with practicality.

Furthermore, in the reception and regeneration circuit, it is impossible to directly configure a so-called Chi-Col type speech synthesizer in place of the prediction filter 5a shown in FIG. 3.

[Purpose of the invention]

The present invention has been made in consideration of these circumstances, and its purpose is to provide a highly practical audio signal prediction method that is inexpensive, has a simple configuration, and is capable of stably performing prediction processing on audio signals. An object of the present invention is to provide an encoding device W.

[Summary of the invention]

The present invention uses a lattice-type prediction circuit to predict an estimated instantaneous value of the audio signal using characteristic parameters of the audio signal, and encodes a difference signal between the estimated instantaneous value and the current instantaneous value. It consists of a digital filter and an adder that calculates the sum of the tap outputs of this filter! This is a characteristic feature.

〔Effect of the invention〕

Thus, according to the present invention, signal filtering processing is performed using the partial autocorrelation coefficient without performing feedback addition processing using the prediction circuit shown in FIG.
Simply setting the partial autocorrelation coefficient to the condition that Ik,l<1 results in stable operation without causing oscillation.

Therefore, it is possible to always stably and efficiently predictively encode an audio signal at a low bit rate, and the apparatus can be configured to be small, simple, and inexpensive.

Therefore, tremendous practical advantages and effects can be achieved.

[Embodiments of the invention]

Hereinafter, one embodiment of the present invention will be described with reference to the drawings.

The device of the present invention has the same basic configuration as the conventional device shown in FIG. 1, except that the prediction circuit 5 is composed of a lattice digital filter 5C and an adder 5d as shown in FIG. This is a characteristic feature. Thus, the lattice type digital filter 5C receives the decoded difference signal δ. 〃 is sequentially transmitted via adders 21=, 2L2 to 21-8, and the outputs are transmitted to delay circuits 22-8 to 22, .
22= and adders 23-8 to 23-2 alternately. And each adder 21-2.
The outputs of 2L, ~2No8 are respectively multiplier 242+
”-3 to 24-8 to the partial autocorrelation coefficient kl+ from 2 to
is multiplied by 7, and then the adder 234.2 L3~23-
8, and the delay circuit 22-
Each delay output of 8 to 22-2.2L1 is output from a multiplier 25-8 to
25. , 25-.

to 1 to partial autocorrelation coefficient through? After being multiplied by +k11, the signals are supplied to the adders 21-8 to 2L2.2L, respectively.

However, such a lattice type digital filter 5
Each of the multipliers 25-, 254-- which is the tap output of C
The respective outputs of 25-8 are supplied to the adder 5d, the sum of which is calculated and outputted as the estimated instantaneous value Sn0.

Each tap output of lattice type digital filter 5C
The adder 5d that calculates the sum of H-1·kI obtains an output of △, and since this output SnO is expressed as △=Sn, it is configured using the above-mentioned prediction filter 5a and adder 5b. This results in a signal exactly similar to that of a prediction circuit realized by a feedstock loop. However, the above relationship is obtained from the fact that the relationship holds true when γn is converted by 2.

Thus, if such a prediction circuit 5 is configured by a lattice type digital filter 5C and an adder 5d, the partial autocorrelation coefficient k of the digital filter 5C can be expressed as tk, I
By simply setting <i, it is possible to easily construct a stable system without oscillation. Therefore, there is no need to increase the scale of the device configuration in order to improve the calculation accuracy as in the past, and it is possible to easily realize a stable device at low cost, and it is possible to use a low bit rate, high efficiency, and good prediction code for audio signals. It becomes possible to Therefore, its practical advantages are enormous.

Note that the present invention is not limited to the above embodiments. For example, the number of stages of the lattice digital filter 5C forming the prediction circuits y and s may be arbitrarily determined according to the specifications. Further, each partial autocorrelation coefficient k of the lattice type digital filter 5C may be given in accordance with the characteristic parameter of the audio signal after satisfying the condition IkiI<1. In short, the present invention can be implemented with various modifications without departing from the gist thereof.

[Brief explanation of drawings]

FIG. 1 is a schematic block diagram of a conventional predictive encoding device, FIG. 2 is a block diagram of a receiving and reproducing circuit, FIG. 3 is a block diagram of a predictive filter in the conventional device, and FIG. 4 is a device according to an embodiment of the present invention. FIG. 3 is a configuration diagram of a prediction circuit used for the prediction circuit. J... Voice analysis circuit, 2... Delay circuit, 3... Subtraction circuit, 5... Prediction circuit, 5. m...Prediction filter, 5b...Adder, 5C...Lattice type digital filter, 5d...Adder, 6...Encoding circuit, 7
...Decoding circuit. Takehiko Suzue, Patent Attorney on behalf of the applicant Figure 1 Figure 2 Figure 183

Claims (1)

[Claims] an audio signal analysis circuit that obtains a feature/parameter signal of the audio signal from past sample values of the audio signal; a prediction circuit that inputs the feature parameter signal and estimates and obtains an estimated instantaneous value for the audio signal; an encoding circuit that encodes a difference signal between the estimated instantaneous value and the current instantaneous value of the estimated audio signal; What is claimed is: 1. A predictive coding device for an audio signal, comprising: