JPS63244100A - Voice analyzer and voice synthesizer - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] [Table of Contents] Overview Industrial Field of Application Prior Art (i) Waveform Coding System (ii) Analysis and Synthesis System Problems to be Solved by the Invention Means for Solving the Problems ( i) First invention (i+) Second invention effect (i) First invention (ii) Second invention embodiment ■, Correspondence between the embodiment and FIG. 1 (i) First invention (:i) Second Invention ■, Structure and operation of embodiments (i) Speech analysis device (i-1) Structure of speech analysis device (i-2) Operation of speech analysis device (11) Speech synthesis device (ii-1) Speech synthesis device Configuration (ii-2) Operation of speech synthesis device■, Summary of embodiments■0 Variations of the invention Effects of the invention [Summary] A speech analysis device that selectively determines a plurality of parameters to determine a sound source waveform. By modeling the input audio signal, performing an autoregressive moving average analysis on the input audio signal based on the modeled sound source waveform, and performing audio analysis by determining the optimal parameters according to the error at that time, the amount of information about the audio can be increased. Efficiently compressed.

In addition, the speech synthesis device is a voice synthesizer, and according to various parameters obtained by such autoregressive moving average analysis,
By modeling the sound source waveform based on the parameters introduced for modeling the sound source waveform and performing speech synthesis based on the multiple parameters obtained through autoregressive moving average analysis, high-quality synthesized speech can be created. can get.

[Industrial application field]

The present invention relates to a speech analysis device and a speech synthesis device, and particularly to a so-called A-b-S (Analysis-by-S).
The parameters of the vocal cord sound source waveform model are determined so that the mean square error is minimized by adopting the method of
The present invention relates to a speech analysis device and a speech synthesis device that analyze speech and synthesize speech by combining it with ARMA (hereinafter referred to as ARMA).

[Conventional technology]

Various methods have been proposed for the recognition, transmission, storage, etc. of speech in order to compress the amount of speech-related information as much as possible and to make it possible to reproduce high-quality speech from the speech-related information. has been done. In this case, it is desired to increase the compression ratio of the amount of information related to speech and to reproduce speech with rich naturalness.

For example, ADPC
There is a ``waveform encoding method'' that encodes the waveform of a voice such as M as it is, and as an alternative to this, there is a narrowly defined ``analysis and synthesis method'' using a vocoder (VOCODBR). These methods will be explained separately.

(i) Waveform Coding The ``waveform coding method'' performs linear prediction analysis on an audio signal, obtains linear prediction coefficients and prediction errors, and then quantizes the prediction errors. Furthermore, in the case of reproduction, the quantized prediction error is driven by a filter using the linear prediction coefficient obtained by analysis. The distortion of the reproduced audio due to this waveform encoding method is due to the quantization of prediction errors, and high quality reproduced audio can be obtained.

However, the amount of information is, for example, 16 kbps ~
64 kbps, and the amount of information regarding audio is quite large. In other words, the compression rate of this "waveform encoding method" is not very high, and speech recognition, transmission, storage, etc. are not efficient.

(ii) In the ``Analysis and Synthesis Method,'' when a human voice is analyzed, information is compressed separately into frequency spectrum envelope information and sound source information of the voice. Therefore, we model the speech generation mechanism and focus on the sound source signal and the acoustic filter characteristics of the articulator.

For example, the acoustic filter is a linear prediction filter, the sound source signal of a voice is a periodic impulse train, and the sound source signal of an unvoiced sound is white noise. According to this, for example,
Speech is expressed by voiced/unvoiced sound discrimination information, pitch frequency, amplitude information, and linear prediction coefficients regarding a two-period sound source. In other words, it can be considered that the prediction error is modeled, and the audio information is processed at, for example, 1.2 kbps to 9 kbps.
, 6 kbps.

However, the quality of speech synthesized by this analysis and synthesis method is considerably lower than that of the above-mentioned "waveform encoding method."

[Problem that the invention seeks to solve]

As described above, even when analyzing or synthesizing speech using the above-mentioned "waveform encoding method" and "analysis/synthesis method," the amount of information regarding speech is large, or the sound quality after synthesis is insufficient. Therefore, there has been a need for a speech analysis and synthesis method that is as high quality as a waveform encoding method and that can compress information as well as an analysis and synthesis method.

In order to meet such demands, the present applicant has filed Japanese Patent Application Laid-Open No. 61-128299 (Japanese Patent Application No. 59-250133).
We have already proposed the ``Speech Analysis/Analysis and Synthesis Method''.

The technique proposed by this patent uses a vocal fold sound source waveform such as a Rosenberg waveform to model the sound source waveform, rather than approximating the sound source with a pulse and a noise signal. Here, the four parameters of pitch period, rise time, fall time, and amplitude for defining this vocal cord sound source waveform model are A
-b-3 method is used to find it.

In other words, when performing speech analysis or speech analysis and synthesis based on information modeling a sound source waveform, the sound source signal is driven by a sound source signal defined by at least four parameters: pitch period, rise time, fall time, and amplitude. It has a speech synthesis system that generates a speech signal using a linear prediction filter, which sequentially selects four types of parameters, and generates a synthesized speech signal obtained by the linear prediction filter and an input speech signal for the four selected parameters. , and optimize the four types of parameters so that the error between the synthesized voice signal and the input voice signal becomes smaller, and determine the four types of parameters. and speech analysis or speech analysis and synthesis based on the linear prediction coefficients.

However, even with such a technique, there are still problems in that the compression ratio of the amount of information is low when obtaining and compressing parameters related to speech, and the quality of synthesized speech is also low.

The present invention was created in view of these points, and includes a speech analysis device that has a high compression rate for speech-related information I[i], and a speech analysis device that can produce high-quality synthesized speech when performing speech synthesis based on the analysis results. The purpose of the present invention is to provide a speech synthesis device that can obtain the following.

[Means for solving problems]

As a means for solving the problems according to the present invention, "
There is a speech analysis device J and a "speech synthesis device."

"UNI" LL1 skin FIG. 1 (A) is a principle block diagram of the speech analysis device according to the first invention.

In the figure, parameter determining means 113 selectively determines a plurality of parameters necessary for modeling the sound source waveform,
A modeling parameter signal 111 representing the determined parameters is output.

The sound source waveform generating means 117 generates the modeling parameter signal 1
A sound source waveform is modeled according to a plurality of parameters represented by 11, and a sound source waveform signal 115 representing the modeled sound source waveform is output.

The autoregressive moving average analysis means 125 receives the input audio signal 119 and the sound source waveform signal 115 to be analyzed, performs an autoregressive moving average analysis to obtain an error, and supplies an error signal 121 representing the error to the parameter determination means 113. At the same time, it outputs an analysis parameter output signal 123 representing a parameter based on autoregressive moving average analysis.

Therefore, as a whole, the plurality of parameters of the parameter determining means 113 are changed in accordance with the error represented by the error signal 121.

FIG. 1(B) is a block diagram of the principle of the speech synthesis device according to the second invention.

In the figure, the sound source waveform generation means 135 generates a modeling parameter signal 131 representing a plurality of first parameters obtained for modeling the sound source waveform required when performing autoregressive moving average analysis on the audio signal to be analyzed. is received, the sound source waveform is modeled based on the first parameter, and a sound source waveform signal 133 representing the modeled sound source waveform is output.

The autoregressive moving average synthesis means 141 receives the parameter signal 137 representing the plurality of second parameters obtained in the autoregressive moving average analysis and the sound source waveform signal 133 from the sound source waveform generation means 135, and synthesizes the first parameter and Speech synthesis is performed based on the second parameter and a synthesized speech output signal 139 is output.

Therefore, the overall structure is such that the sound source waveform is modeled according to various parameters obtained by autoregressive moving average analysis, and speech synthesis is performed based on the autoregressive moving average analysis parameters.

(Function) The sound source waveform generating means 117 models the sound source waveform according to the plurality of parameters selectively determined by the parameter determining means 113.The sound source of the modeled sound source waveform Based on the waveform signal 115, the autoregressive moving average analysis means 125 performs an autoregressive moving average analysis on the input audio signal 119.

The error at that time is given to the parameter determining means 113, and the parameter determining means 113 optimizes the plurality of parameters.

As a result of parameter optimization by the parameter determining means 113, sound source waveform parameters are obtained from the parameter determining means 113, and autoregressive moving average parameters are obtained from the autoregressive moving average analyzing means 125.

Note that when the device of the present invention performs more specific operations, (
(corresponding to an embodiment), the plurality of parameters selected and optimized in the parameter determining means 113 are pitch period, glottis opening time.

Distortion of the sound source within the glottal opening section, time from glottal closure until the volume flow reaches its negative maximum value, slope of the volume flow waveform at the start of glottal opening, slope of the volume flow waveform just before glottal closure, and volume immediately after glottal closure. There are seven types of slopes of flow waveforms. The sound source waveform is modeled using these parameters.

In the present invention, the sound source waveform parameters and the autoregressive moving average parameters are obtained by optimizing the parameters in the parameter determining means 113, so that the amount of information related to speech is efficiently compressed.

Based on the sound source waveform parameters already obtained, the modeling parameter signal 131 models the sound source waveform.

The autoregressive moving average synthesis means 141 performs speech synthesis based on the sound source waveform obtained by this modeling and the already obtained autoregressive moving average variation.

Note that when the device of the present invention performs more specific operations, (
(corresponding to the embodiment), the sound source waveform parameters already obtained are pitch period, glottal opening time, distortion of the sound source within the glottal opening section, time from glottal closure until the volume flow reaches the negative maximum value, glottal opening. There are seven types: the slope of the volume flow waveform at the start, the slope of the volume flow waveform immediately before glottis closure, and the slope of the volume flow waveform immediately after glottis closure. The sound source waveform is modeled using these parameters.

In the present invention, high-quality synthesized speech can be obtained by performing speech synthesis using an autoregressive moving average based on the already obtained sound source waveform parameters and autoregressive moving average parameters.

(Example) Hereinafter, an example of the present invention will be described in detail based on the drawings.

FIG. 2 shows a speech analysis device according to an embodiment of the present invention, FIG. 3 shows a sound source waveform model used in the speech analysis device of FIG. 2, and FIG. 4 shows a speech analysis device according to an embodiment of the invention. A speech synthesis device is shown. FIG. 5 specifically shows the ARMA filter in the speech synthesizer shown in FIG. 4.

1. Correspondence between . and 1. Here, the correspondence between the embodiment of the present invention and FIG. 1 will be shown.

The modeling parameter signal III corresponds to the sound source waveform parameter signal 227.

The parameter determining means 113 includes the optimal parameter determining unit 2
23. This corresponds to the parameter selection section 225.

The sound source waveform signal 115 corresponds to the vocal cord sound source waveform signal 2F.

The sound source waveform generation means 117 includes the vocal cord sound source waveform generation section 215
corresponds to

Input audio signal 119 corresponds to input audio signal 211.

The error signal 121 corresponds to the error signal 221 from the ARMA analysis section 213.

Analysis parameter output signal 123 corresponds to ARMA parameter signal 219.

The autoregressive moving average analysis means 125 includes the ARMA analysis section 2
It corresponds to 13.

The ``work and ball'' modeling parameter signal 131 corresponds to the sound source waveform parameter signal 411.

The sound source waveform signal 133 corresponds to the vocal cord sound source waveform signal 415.

The sound source waveform generation means 135 includes the vocal cord sound source waveform generation section 413
corresponds to

The parameter signal 137 is the ARMA parameter signal 41
Corresponds to 7.

The synthesized speech output signal 1°39 corresponds to the synthesized speech signal 421.

The autoregressive moving average synthesis means 141 corresponds to the ARMA filter 419.

Examples of the present invention will be described below, assuming that the first invention and the second invention have the above-mentioned correspondence relationship.

(2) The present invention relates to a "speech analysis device" and a "speech synthesis device," so each case will be explained separately below.

(i) Voice 'I4+ First, the voice analysis device to which the first invention is applied will be described below. Here, it is assumed that what is being analyzed is speech uttered by a person.

(i-1)! In Fig. 2 of the father-in-law, the microphone 231 is used to collect the pronunciation that is the subject of analysis;
The sampled signal from 1 is supplied to an analog-to-digital (A/D) converter 233.

The input audio signal 211 obtained by being quantized and digitized in this A/D converter 233 is supplied to the ARMA analysis section 213 as an analysis target. This ARMA
The analysis unit 213 performs voice analysis based on the vocal cord sound source waveform signal 217 from the vocal cord sound source waveform generation unit 215, and
It outputs an ARMA parameter signal 219 representing the A parameter.

In the process of voice analysis, an error signal 221 representing an error between the input voice signal 211 and the vocal cord sound source waveform signal 217 is generated.
is generated and supplied to the optimal parameter determining section 223. The optimal parameter determining section 223 includes a parameter selecting section 225.
The parameters in step 3 are appropriately selected and switched. A sound source waveform parameter signal 227 representing the sound source waveform parameter selected by the parameter selection section 225 is output and supplied to the vocal cord sound source waveform generation section 215.

In the embodiment of the present invention having the above configuration (i-2), a vocal cord sound source waveform model is adopted instead of using an impulse as a periodic sound source to model the sound source for voice analysis. For example, there are various changes in the human voice, such as a clear voice and a melodious voice. This may be due to the influence of differences in sound sources, and it is difficult to obtain reasonable results when approximating the -temporal impulse. By using the vocal cord sound source waveform model,
Approximation can be further improved.

FIG. 3 shows an example vocal cord sound source waveform g(t) and its differential waveform #(1).

By the way, the modeled vocal cord sound source waveform has pitch period T, glottal opening time W, and sound source distortion S within the glottal opening section.
, the time from glottal closure until the volume flow reaches its negative maximum value, the slope A of the volume flow waveform at the start of glottal opening, the slope B of the volume flow waveform immediately before glottal closure, and the slope C of the volume flow waveform immediately after glottal closure. It can be expressed by seven types of parameters. Therefore, the differential waveform #(1) of the vocal cord sound source waveform is divided and expressed by time t.

■ If g<t≦R, g(t)-A-(2A+R,α) t/R+ (2A
+ R3α) tZ /R2・・・・・・(1) ■ If R<t≦W, gct”)=α(t-R) +(3B-2αF)(t-R)”/F”+ (2B=αF
)(t-R)' /F3... (2) ■ If W< t5W+D, g(t)=C-2(C-β)(t-W)/D+ (
C-β)(t-W)"/D" ...... (3) ■ If W+D<t≦T, #(1)=β ......(4)
becomes.

Here, α and β are expressed as follows.

α= (4AR+6FB)/(2R"-F")・
・・・・・・(5) β=CD/ (D-3(T-W)) ・・・・・・(
6) By the way, the glottis opening time W and the distortion S of the sound source within the glottis opening section are as follows: W=R+F (7)
S-(R-F)/(R+F)...(8
).

Therefore, in the configuration shown in FIG. 2, a predetermined pitch period T, glottal opening time W, distortion S of the sound source within the glottal opening section, time from glottal closure until the volume flow reaches the negative maximum value, glottis Appropriate initial values of seven parameters including the slope A of the volume flow waveform at the start of opening, the slope B of the volume flow waveform immediately before glottis closure, and the slope C of the volume flow waveform immediately after glottis closure are sent to the optimal parameter determination unit 223. Given.

The parameter selection unit 225 first selects these seven types of parameters using initial values, outputs them as a sound source waveform parameter signal 227, and provides the signal to the vocal cord sound source waveform generation unit 215.

The vocal fold and sound source waveform generation unit 215 uses these seven parameters (pitch period T, glottal opening time W, distortion S of the sound source within the glottal opening section, time from glottal closure until the volume flow reaches the negative maximum value, Slope A of the volume flow waveform at the start of glottis opening,
A vocal cord sound source waveform g(t) as shown in FIG. 3 is synthesized by integration based on the slope B of the volume flow waveform immediately before glottis closure and the slope C) of the volume flow waveform immediately after glottis closure. A vocal cord sound source waveform signal 217 representing a vocal cord sound source waveform g(t) as a result of the synthesis is provided to the ARMA analysis section 213.

Note that this vocal cord sound source waveform g(t) may include, as necessary,
Correction may be made taking into account so-called radiation characteristics.

The ARMA analysis unit 213 performs voice analysis based on the vocal cord sound source waveform signal 217 and the input voice signal 211.
Pseudo speech is synthesized according to the frequency spectrum envelope based on the A parameter, and compared with the vocal cord sound source waveform signal 217. The sound source waveform parameters and ARMA parameters are determined so that the error between these two signals is minimized.

Now, in the ARMA analysis performed by the ARMA analysis section 213, the audio signal s(n) is expressed as . . . (9).

Here, α1 is an AR parameter and βj is an MA parameter. p and q are respective prediction orders, g
(n) is the sound source signal, and e(n) is the prediction error signal. α
The parameters βj and βj are collectively referred to as ARMA parameters and indicate the frequency spectrum envelope, and these parameters are externally transmitted as the ARMA parameter signal 219 (for example,
In FIG. 4, the signal is provided to a speech synthesis device (to be described later).

In the ARMA analysis section 213, the error E (represented by the error signal 221) to be minimized is expressed as:

The error E obtained in this manner is supplied to the optimum parameter determining section 223.

In order to reduce this prediction error, the optimal parameter determining section 223 instructs the parameter selecting section 225 to change the parameters defining the vocal cord sound source waveform little by little. A parameter that takes a value different from the parameter is selected and output to the vocal cord sound source waveform generation section 215.

By the way, the selection of this parameter is, for example, A-b-S (Analysis-
by-Synthesis) method.

By repeating the above steps, the seven optimal parameters (pitch period T, glottal opening time W, sound source distortion S within the glottal opening section, time from glottal closure until the volume flow reaches its negative maximum value) are determined. , the slope A of the volume flow waveform at the start of glottis opening, the slope B of the volume flow waveform immediately before glottis closure, and the slope C) of the volume flow waveform immediately after glottis closure are determined. That is, the so-called A-b-
By using the S method, the above seven types of parameters are determined so that the mean square error in the time domain is minimized.

In this way, ARMA analysis is performed on the input speech signal 211, and as a result of the analysis, the ARMA parameter represented by the ARMA parameter signal 219 and the sound source waveform parameters (pitch period T, glottal opening) represented by the sound source waveform parameter signal 227 are obtained. Time W, distortion S of the sound source within the glottal opening section, time from glottal closure until the volume flow reaches the negative maximum value, slope A of the volume flow waveform at the start of glottal opening, slope of the volume flow waveform just before glottal closure B and 7 parameters of the slope C of the volume flow waveform immediately after glottis closure)
is obtained.

Since the input audio is represented by these ARMA parameters and sound source parameters, we use them as compressed information,
This means that the audio information has been compressed. Furthermore, for later speech synthesis, these multiple parameters may be stored in an external device (memory, etc.). Therefore, by using these parameters, it is possible to conversely perform speech synthesis using a speech synthesis device (described later) that is configured separately from the speech analysis device.

(ii) Speech synthesis device Next, a speech synthesis device to which the second invention is applied will be described below. This speech synthesis device is driven by parameters obtained through ARMA analysis by the above-mentioned sound source analysis device. Note that various parameters may be received via a communication line when performing speech synthesis, or may be stored in a ROM in advance.

(ii-1) The ``speech synthesis device'' shown in Figure 4 of ``Joining the Company'' is different from the ``speech analysis device'' shown in Figure 2, which has various parameters for speech synthesis.
shall be obtained directly from.

In FIG. 4, a vocal cord sound source waveform generation section 413 to which the sound source waveform parameter signal 411 is input is the same as the vocal cord sound source waveform generation section 215 shown in FIG. A sound source waveform is modeled based on the sound source waveform parameter represented by the sound source waveform parameter signal 411, and a vocal cord sound source waveform signal 415 obtained by the modeling is generated.

Also, the ARMA parameter signal 417 that receives the ARMA parameter signal 417 is
The filter 419 performs speech synthesis based on the vocal fold sound source waveform signal 415 from the vocal fold sound source waveform generation section 413 and outputs a synthesized speech signal 421 as the synthesis result.

By the way, the sound source waveform parameter signal 411 and the ARM
The sound source waveform parameters and ARMA parameters represented by the A parameter signal 417 are parameters obtained as an analysis result by the above-mentioned speech analysis device. Therefore, the sound source waveform parameters are pitch period T, glottal opening time W, sound source distortion S within the glottal opening section, time from glottal closure until the volume flow reaches the negative maximum value, and volume flow waveform at the start of glottal opening. It consists of seven parameters: slope A of the volume flow waveform immediately before glottis closure, slope B of the volume flow waveform immediately before glottis closure, and slope C of the volume flow waveform immediately after glottis closure.

FIG. 5 shows a detailed configuration of the ARMA filter 419 shown in FIG. 4. Here, the ARMA parameters (AR parameter α5 and MA parameter βj) represented by the ARMA parameter signal 417 are calculated by p coefficient multipliers 511 . ．．
511. ,...,5139.・・・・・・、5
11. Also, there are q other 513, . 5132.・・・
..., 513. are supplied to each. Here, p and q are the prediction orders.

Further, p delay elements 515+, 51 connected in series
5z,...,515. ,...,515
゜, and each delay element has a time element Z
It is a time delay element of unit time determined by . Output signals sequentially delayed by each delay element are commonly supplied to the coefficient multiplier 511 and the coefficient multiplier 513 of the corresponding order.

Vocal cord sound source waveform signal 41 from vocal cord sound source waveform generation section 413
5, and p coefficient multipliers 511. The output signal from .about.511.degree. Further, the adder 519 has q
Output signals of coefficient units 513+ to 513q are also supplied,
A synthesized audio signal 421 is output.

(ii-2)''-In the speech synthesis device with the configuration described above, the sound source waveform parameter signal from the optimal parameter determining unit 223 of the "speech analysis device" described above in "(i) Speech analysis device" 22
The sound source waveform parameters provided by 7 are first given to the vocal cord sound source waveform generation section 413. This vocal cord sound source waveform generation unit 413 has the same configuration and operation as the vocal cord sound source waveform generation unit 215 of the above-mentioned “speech analysis device”, and has the same configuration and operation as the vocal cord sound source waveform generation unit 215 of the “speech analysis device” described above.
Glottal opening time W, distortion of the sound source within the glottal opening section S, time from glottal closure until the volume flow reaches the negative maximum value, slope A of the volume flow waveform at the start of glottal opening, volume flow waveform just before glottal closure. and the slope C of the volume flow waveform immediately after glottal closure.
(7 types of parameters), the sound source waveform is modeled to generate a vocal cord sound source waveform g(t), which is output as a vocal cord sound source waveform signal 415.

On the other hand, p coefficient units 5111.511g, . . . , 511Q in the ARMA filter 419
,..., the coefficient of 511p is the AR parameter α
! (α1.αz, °9°I", Q'q, "---
−, αp), and the ARMA filter 41
The optimal coefficient for 9 is determined. Similarly, other q coefficient units 513. ．． 513g, 51
Each coefficient of 3q is also determined by the MA parameter β5 (β1
．． β2. . . . , βQ).

In this way, p coefficient units 5111 to 511 . and other q coefficient units 5131 to 513. The modeled vocal fold sound source waveform signal 415 (g(t)) is filtered in the vocal fold sound source waveform generation unit 413 based on each coefficient of .

Now, if the output signal of the adder 517 is S%17, then
The output signal 5SISI of the first delay element 5151 is the output signal 5SI7 of the adder 517 delayed by a unit time (determined by the time element Z). Further, the output signal ss+sz of the second delay element 515□ is the output signal ss+sz of the second delay element 515
The output signal 5sISI of I is delayed by a unit time. Similarly, the output signals of other delay elements are sequentially delayed by a unit time.

Therefore, the coefficient unit 511. and the output signal S of the other coefficient multiplier 513+! 1111 and S S13+ are α,・S
It is expressed as s1%1 and β2 ·5sts+. Further, the output signals SS+□ and SSt of the next coefficient multiplier 5118 and the other coefficient multiplier 513g. is expressed as α2・SSI□ and β2・ss+sz. Similarly, for the other coefficient units, the output signal from each delay element is multiplied by the coefficient and outputted.

These p coefficient units 511. ~511. The output signal is supplied to the adder 517 in the previous stage, and
3 is summed with the vocal cord source waveform (g(t)) signal 415 from 3.

Further, the output signals of the other q coefficient units 5131 to 513q are supplied to the subsequent stage coefficient supply 519, and the total sum is taken together with the output signals S and 17 of the adder 517. This adder 5
The voice signal synthesized by step 19 is output as a synthesized voice signal 421 of the voice synthesizer.

As described above, in the "speech analysis device" according to the embodiment of the present invention, the sound source waveform is modeled by selectively determining a plurality of modeling parameters, and the Audio analysis is performed by performing autoregressive moving average analysis on the input audio signal 211 based on the sound source waveform and determining optimal parameters according to the error at that time.

Further, in the "speech synthesis device", the vocal cord sound source waveform generation unit 413 models the sound source waveform according to the sound source waveform parameters obtained by autoregressive moving average analysis on the speech analysis device side. Furthermore, the AR parameters α, , M obtained by autoregressive moving average analysis on the voice analysis device side
An ARMA filter 419 performs speech synthesis in accordance with the A parameter β.

Since both are based on ARMA, the amount of information related to speech is efficiently compressed, and high-quality synthesized speech can be obtained.

Figure 6 shows the frequency spectrum of the original voice (ORIG), this method (GARMA), and the conventional methods (ARMA, GLP).
Figure 3 shows a comparison of the frequency spectral envelopes of nasalized vowels estimated by C, LPC).

Here, this method (GARMA) uses a pole-zero filter based on the ARMA model as an "acoustic filter" for analysis.
Furthermore, the vocal cord sound source waveform shown in FIG. 3 is used as the "sound source".

On the other hand, ARMA as a conventional method employs a pole-zero filter based on the ARMA model as an "acoustic filter" for analysis and a pulse as a "sound source." Furthermore, in the conventional GLPC, an all-powerful filter based on a linear prediction method is used as an "acoustic filter", and a vocal cord sound source waveform shown in FIG. 3 is used as a "sound source". Further, in the conventional LPG, an all-pole filter based on a linear prediction method is used as the "acoustic filter" and a pulse is used as the "sound source".

In this way, it can be seen that the present method (GARMA) approximates the spectral envelope of the original sound better than the conventional method, and is faithful to the original sound during reproduction.

By the way, the embodiments of the present invention are particularly effective for voiced sounds, and when analyzing an unvoiced sound part, for example, only that part can be analyzed using the conventional waveform encoding method and the method according to the embodiments of the present invention. The present invention can be implemented by combining this method with a conventionally used method.

■Form of the invention In the embodiment of the present invention described above, the "speech analysis device" and the "speech synthesis device" are integrally configured as a pair, but the invention is not limited to this. There isn't. That is, the speech analysis device and the speech synthesis device can be configured and used separately from each other. Therefore, for example,
Various parameters (sound source waveform parameters and AR
MA parameters) may be individually provided to each of a plurality of "speech synthesis devices" using a ROM or the like, and the speech synthesis may be performed individually in each speech synthesis device.

In addition, in "1. Correspondence between Examples and FIG. 1",
Although the correspondence between FIG. 1 and the present invention has been described, those skilled in the art will easily imagine that the present invention is not limited to this and that there are various modifications.

〔Effect of the invention〕

As described above, in the speech analysis device according to the present invention, the sound source waveform is modeled by selectively determining the sound source waveform modeling parameters, and the input speech signal is subjected to autoregressive moving average analysis based on the modeled sound source waveform. Then, voice analysis is performed by determining the optimal parameters according to the error at that time.

Furthermore, in the speech synthesis device according to the present invention, the sound source waveform is modeled based on the parameters introduced for modeling the sound source waveform in accordance with the various parameters obtained by such autoregressive moving average analysis. Speech synthesis is performed in response to a plurality of parameters obtained in regression moving average analysis.

Therefore, as a result of autoregressive moving average analysis and synthesis, the amount of information regarding speech is efficiently compressed and high-quality synthesized speech is obtained, which is extremely useful in practice.

[Brief explanation of drawings]

FIG. 1 is a block diagram of the principle of the present invention. FIG. 2 is a block diagram of the configuration of a speech analysis device according to an embodiment of the invention. FIG. 3 is an explanation of the sound source waveform used in the speech analysis device shown in FIG. 4 is a block diagram of the configuration of a speech synthesis device according to an embodiment of the present invention, and FIG. 5 is an ARM used in the speech synthesis device shown in FIG. 4.
FIG. 6 is a block diagram showing a specific configuration of the A filter. FIG. 6 is an explanatory diagram showing a comparison of frequency spectrum envelopes of audio between the method according to the embodiment of the present invention and the conventional method. In the figure, 111 is a modeling parameter signal, 113 is a parameter determining means, 115 is a sound source waveform signal, 117 is a sound source waveform generating means, 119 is an input audio signal, 121 is an error signal, 123 is an analysis parameter output signal, and 125 is a self Regression moving average analysis means, 131 is a modeling parameter signal, 133 is a sound source waveform signal, 135 is a sound source waveform generation means, 137 is a parameter signal, 139 is a synthesized audio output signal, 141 is an autoregressive moving average synthesis means, 211 is an input 213 is an ARMA analysis unit, 215 is a vocal cord sound source waveform generation unit, 217 is a vocal cord sound source waveform signal, 219 is an ARMA parameter signal, 223 is an optimal parameter determination unit, 225 is a parameter selection unit, 227 is a sound source waveform parameter signal, 411 is a sound source waveform parameter signal, 413 is a vocal cord sound source waveform generator, 415 is a vocal cord sound source waveform signal, 417 is an ARMA parameter signal, 419 is an ARMA filter, 421 is a synthesized speech signal, 5111 to 511. ．． 5131-513. is a coefficient unit, 515I to 515. is a delay element, and 517 and 519 are adders. Irregular← Sun Moon Principle Block Diagram 1 (A) 17 Deaths of Uncooked Memories Diagram 1 CB) ￥1 Color Poetry 1] To the Hanawa Diagram 2 Symbol Sound Left 1 Odd Simplified Explanation Guard gate Figure 3 Figure 4 Figure 6

Claims (6)

[Claims] (1) Parameter determining means (113) for selectively determining a plurality of parameters necessary for modeling a sound source waveform and outputting a modeling parameter signal (111) representing the determined parameters; and the modeling parameters. a sound source waveform generating means (117) that models a sound source waveform according to the plurality of parameters represented by the signal (111) and outputs a sound source waveform signal (115) representing the modeled sound source waveform; Upon receiving the input audio signal (119) to be analyzed and the sound source waveform signal (115), an autoregressive moving average analysis is performed to obtain an error, and an error signal (121) representing the error is supplied to the parameter determining means (113). and an autoregressive moving average analysis means (125) for outputting an analysis parameter output signal (123) representing a parameter resulting from the autoregressive moving average analysis, and according to the error represented by the error signal (121). A speech analysis device characterized in that the plurality of parameters of the parameter determining means (113) are changed to determine the optimum parameters. (2) The multiple parameters necessary for modeling the sound source waveform are the pitch period that defines the vocal fold sound source waveform model, the glottal opening time, the distortion of the sound source within the glottal opening section, and the negative maximum volume flow from glottal closure. Claim 1, characterized in that there are seven types of volume flow waveforms: the time to reach a value, the slope of the volume flow waveform at the start of glottal opening, the slope of the volume flow waveform immediately before glottal closure, and the slope of the volume flow waveform immediately after glottal closure. The speech analysis device according to item 1. (3) Optimization of the plurality of parameters in the parameter determining means (113) is characterized in that the plurality of parameters are changed in a direction in which the error represented by the error signal (121) is minimized. A speech analysis device according to claim 1. (4) The parameter represented by the analysis parameter output signal (123) is determined by the autoregressive moving average analysis means (1
25) The speech analysis device according to claim 1, wherein the speech analysis device is an autoregressive moving average parameter obtained by autoregressive moving average analysis in step 25). (5) Receive a modeling parameter signal (131) representing a plurality of first parameters obtained for modeling a sound source waveform required when performing an autoregressive moving average analysis of an audio signal to be analyzed; a sound source waveform generating means (135) that models a sound source waveform based on one parameter and outputs a sound source waveform signal (133) representing the modeled sound source waveform;
), a parameter signal (137) representing a plurality of second parameters obtained in the autoregressive moving average analysis, and a sound source waveform signal (133) from the sound source waveform generating means (135). A speech synthesis device comprising an autoregressive moving average synthesis means (141) that performs speech synthesis based on one parameter and a second parameter and outputs a synthesized speech output signal (139). . (6) The first parameter is the pitch period necessary to define the vocal fold sound source waveform model, the glottal opening time, the distortion of the sound source within the glottal opening section, the time from which the volume flow reaches the negative maximum value from glottal closure, There are seven parameters: the slope of the volume flow waveform at the start of glottis opening, the slope of the volume flow waveform just before glottis closure, and the slope of the volume flow waveform immediately after glottis closure, and the second parameter is an autoregressive moving average parameter. A speech synthesis device according to claim 5, characterized in that: