JPH0670751B2 - Pole zero analyzer with estimation of phase characteristics - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a pole-zero analyzer that estimates not only amplitude characteristics but also phase characteristics when approximating the spectrum of a signal such as voice. .

(Prior Art) In the fields of speech synthesis and speech recognition, it is important to extract the values of parameters that approximate the spectrum of speech. Furthermore, it may be necessary to extract the values of parameters that approximate the spectrum of general signals.

A pole or pole zero is often used as a parameter for approximating the spectrum of a signal such as voice (hereinafter, simply referred to as a signal). This is because poles and zeros have characteristics such as clear physical meanings and are advantageous in application. Of these, it is known that the analysis by a model using only pole parameters (all-pole model) can be easily performed by a linear prediction method or the like, but in order to approximate the spectrum more accurately with relatively few parameters, the pole-zero Parametric analysis is required.

The first conventional example of this pole-zero analysis is described in, for example, Jay. Dee. Markel and A. H. Gray (JD Markel and
AHGray's book "Linear Prediction of Speech (Translated by Suzuki)" (Lin
Ear Predic-tion of Speech) Section 11.4. This uses homomorphic deconvolution. That is, the minimum phase approximation of the impulse response of the model is obtained from the low time component of the cepstrum of the voice segment, and the pole-zero parameter value is obtained from this value.

On the other hand, as a second example of extracting the value of the pole-zero parameter, sea. tea. Maris and Earl. A. "The Youth of Second Order Information in the Aproximation," published in 226-238, pages 226 to 238 of IEEE Transaction 24 (ASSP-24), No. 3, by Roberts (CTMullis and RARoberts). of
Discrete Time Linear Systems (The Us
e of Second-Order Infor-mation in the Approximatio
n of Discrete-Time Linear Systems) ”.

This method is based on the principle that the pole-zero parameter value that minimizes the value of the square integral of the difference between the signal and the amplitude spectrum of the model gives the optimum approximation.

This is obtained by solving a simultaneous equation in which the autocorrelation value of the input signal and the value determined from the impulse response associated with it are used as the coefficients of the equation, and the coefficients of the transfer function of the pole-zero model are the unknowns.

The autocorrelation value and the value of the impulse response can be easily obtained by, for example, a high-order linear prediction method. Alternatively, as in the case of the first conventional example, it is also possible to obtain by homomorphic deconvolution.

(Problems to be Solved by the Invention) According to these pole-zero analysis, the amplitude spectrum of a signal such as voice can be accurately approximated with a small number of parameters. However, since both models assume the minimum phase system, the phase component of the spectrum is not approximated. Since the voice generation process cannot be always promoted as a minimum phase system, this point may be a problem when applied to voice synthesis.

In the first conventional example, if a parameter called a complex cepstrum is used instead of the cepstrum, approximation including the phase characteristic is possible, but the amount of calculation required to extract this parameter value is necessary for simple cepstrum analysis. In addition to the huge amount of computation, it is not easy to obtain pole-zero parameter values from the complex cepstrum.

An object of the present invention is to obtain a pole-zero parameter value that accurately approximates a phase characteristic that is considered to affect the quality of synthesized speech when applied to speech synthesis or the like, with a pole that can be stably obtained with a relatively small amount of calculation. It is to provide a zero analyzer.

(Means for Solving the Problems) The pole-zero analyzer with the estimation of the phase characteristic according to the present invention,
The spectrum of a signal such as a voice has means for performing pole-zero analysis of a type approximated by a pole-zero model of the minimum phase system, and further, using the pole-zero parameter value extracted by the means for performing the pole-zero analysis, Means for inversely filtering a signal such as voice to calculate a residual signal, means for calculating a z-transformed value of the residual signal in the vicinity of a zero point indicated by the extracted zero parameter, and its z-transformed value And means for converting the absolute value of the zero parameter into its inverse.

(Operation) In the present invention, first, pole-zero analysis of a signal is performed based on a model assuming a minimum phase as in the conventional pole-zero analysis, and then the signal is inversely filtered as post-processing to obtain z-transform. The position of the true zero point is determined by.

According to the pole-zero parameter obtained assuming the minimum phase,
Since the amplitude spectrum characteristic of the signal should be well approximated, it is possible to extract the pole-zero parameter that accurately approximates both amplitude and phase by re-analyzing the phase characteristic. In this case, the absolute value of the pole of the model must be 1 or less due to the conditions of stability and causality, so if the phase characteristics of the signal cannot be approximated by the minimum phase, only the position of the zero point is not optimal. become.

The zero points of systems having the same amplitude characteristic but different phase characteristics have the property that the argument is the same and the absolute values thereof are inverse to each other. Therefore, the problem of re-analyzing the phase characteristic of the pole-zero model that approximates the amplitude characteristic assuming the minimum phase is that the zero point obtained as being inside the unit circle actually exists outside the unit circle. If it is determined that it exists outside, it is solved by replacing the zero point with one having the same argument and the reciprocal absolute value.

By the way, since the analysis is performed based on the model assuming the minimum phase, the parameters obtained here have both poles and zeros inside the unit circle on the z plane. Therefore, there always exists an inverse system of this model, that is, a stable causal system in which the poles of the model are replaced by zeros and the zeros are replaced by poles. When the signal is inversely filtered by the inverse system, the pole component of the original signal is canceled at the zero of the inverse system, and the zero component existing inside the unit circle is canceled by the pole of the inverse system. But,
If the original signal had a zero component outside the unit circle, it would not be canceled by the poles of the inverse system, but the inverse filtered signal (so-called residual signal).
Has a pole at the position of the zero of the model of the minimum phase inside the unit circle, and has a zero at the position where the argument is the same and the absolute value is the reciprocal.

Therefore, if the z-transform of this residual signal is obtained in the vicinity of the zero point obtained by the minimum phase model, if the original signal zero point is inside the unit circle, it is canceled by inverse filtering as described above. Therefore, the z-transform of the residual signal in the vicinity thereof has a smooth characteristic, and since the pole of the inverse system remains outside the unit circle, the z-transform of the residual signal in the vicinity thereof
The transformation has a sharp peak. From this, it is possible to determine whether the position of the original zero point of the signal is inside or outside the unit circle. At this time, the value of z for z conversion needs to be different from the position where the zero point exists inside the unit circle.

In this way, the zero point determined to exist outside the unit circle is
By replacing the argument with the absolute value reciprocal, the value of the pole-zero parameter that approximates the phase characteristic with high accuracy can be obtained.

(Example) Next, the Example of this invention is described with reference to drawings.

FIG. 1 is a block diagram of an embodiment of the present invention. First,
The operation of each of the main components will be described. In the figure, reference numeral 2 is a pole-zero analysis circuit, which performs pole-zero analysis based on a model assuming a minimum phase as in the conventional case.

Reference numeral 3 denotes an inverse filter, which calculates a coefficient of a transfer function in which a pole of a pole zero parameter value obtained by assuming a minimum phase is a zero point and a pole is the zero point, and an input signal is filtered to generate a residual signal. .

Reference numeral 4 is a z conversion circuit, which performs z conversion of the residual signal in the vicinity of the zero point extracted by the minimum phase model. The value of z at the time of this conversion is an intermediate value between the absolute value of the zero point and 1 and the deviation angle is the same as that of the zero point, and there are two types that differ by a slight angle. Needed for each. The difference in the declination between the two types of z values can be considered to correspond to the frequency width (half the band width) expected to change, for example, by 3 dB from the value at the resonance point.

A resonance determination circuit 5 determines whether or not a polar component exists in the residual signal based on the results of two types of z conversion in the z conversion circuit. If the z-transform value differs by 3 dB (corresponding to the above example), it is determined that the residual signal includes a polar component, and if it is smaller than that, the polar component is not included. To determine.

Reference numeral 6 denotes a pole-zero determining circuit, which, when it is determined that a pole component exists in the residual signal based on the determination result of the resonance determining circuit 5, determines the absolute value of the zero point corresponding to the minimum phase model. Replace the value with the reciprocal and output as the final pole-zero parameter value. The pole parameter value obtained by the minimum phase model is output as it is.

Next, the overall operation will be described along the flow of signals.

A signal is input from the signal line 11 to the input buffer 7, and signal data of a preset time length (so-called analysis frame data) is sent to the pole-zero analysis circuit 2 via the signal line 12 and converted into a minimum phase model. A pole-zero analysis based on it is performed.
The candidates for pole-zero parameter values obtained as a result of the analysis are signal lines 14
Is sent to and stored in the pole-zero buffer 8 via.

When a pole-zero parameter candidate value based on the minimum phase model is sent from the pole-zero buffer 8 to the inverse filter 3 via the signal line 15, the coefficient of the inverse filter is calculated as described above, and the input buffer 7 outputs the signal line 13 The signal in the analysis frame sent via the signal is inversely filtered and sent via the signal line 18 to the residual buffer 9 for storage.

Next, the zero-zero candidate value obtained based on the minimum phase model is sent from the pole-zero buffer 8 to the z conversion circuit 4 via the signal line 16, and sent from the residual buffer 9 via the signal line 19. The z-transformed value of the residual signal is calculated as described above and sent to the resonance determination circuit 5 via the signal line 20.

The resonance determination circuit 5 determines whether or not a pole component exists in the residual signal based on the z-transformed value as described above, and the determination result is sent to the pole / zero determination circuit 6 via the signal line 21. Sent.

In the pole-zero determination circuit 6, among the pole-zero parameter candidate values sent from the pole-zero buffer 8 via the signal line 17, the pole parameter value is directly output to the signal line 22. On the other hand, for the zero parameter value, when a polar component is detected in the residual signal based on the determination result from the resonance determination circuit 5, the absolute value data of the corresponding zero parameter value is replaced with the reciprocal as described above. Is output as the correct zero parameter value. If the determination result does not indicate the presence of a pole component in the residual signal, the candidate value sent from the pole-zero buffer is output as it is.

1 is a control circuit, which includes signal lines 23, 24, 25, 26, 27, 28, 29 and
Control signals are sent to the input buffer 7, the inverse filter 3, the pole-zero analysis circuit 2, the pole-zero buffer 8, the residual buffer 9, the z conversion circuit 4, the pole-zero determination circuit 6 and the resonance determination circuit 5 via 30, respectively. The initialization of each part is instructed for each frame cycle, and a synchronization signal for data transmission / reception is sent to establish synchronization.

(Effects of the Invention) As described above, according to the present invention, as compared with the conventional pole-zero analysis, the calculation amount of inverse filtering and z-transform is increased, the phase characteristic is also included, and more accurate. A pole-zero parameter value is obtained that can approximate the spectrum.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of the present invention. In the figure, 1 is a control circuit, 2 is a pole-zero analysis circuit, 3 is an inverse filter, 4 is a z conversion circuit, 5 is a resonance determination circuit, 6 is a pole-zero determination circuit, 7 is an input buffer, 8 is a pole-zero buffer, 9
Represents the residual buffer, respectively.

Claims (1)

[Claims] 1. A pole-zero analysis apparatus of a type having means for approximating a spectrum of a signal such as voice with a pole-zero model of a minimum phase system, wherein the pole-zero parameter value extracted by the means for performing the pole-zero analysis is used. Means for calculating a residual signal by inversely filtering a signal such as the voice using the means, means for calculating a z-transformed value of the residual signal in the vicinity of a zero point indicated by the extracted zero parameter, and its z And a means for replacing the absolute value of the zero parameter with its reciprocal number based on the value of the transformation, a polar-zero analysis apparatus with phase characteristic estimation.