JPS61252600A - Lsp type pattern matching vocoder - 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 (Field of Industrial Application) The present invention relates to an LSP pattern matching vocoder that converts an audio signal into a low-speed code string.

(Prior art) A spectral envelope that most closely resembles the spectral envelope of an input audio signal is selected by comparing it with a standard pattern obtained by analyzing audio materials in advance, and this is selected as a spectral envelope that is most similar to the spectral envelope of an input audio signal. Pattern matching vocoders, which reproduce the waveform of an input audio signal by transmitting information as well as sound source information such as pitch period and sound intensity, from the analysis side to the synthesis side are well known these days. LSP coefficients are used as analysis and synthesis parameters on the analysis and synthesis sides of a pattern matching vocoder.
The fMP type bataso pattern matching vocoder is well known.

This LSP coefficient is used as a parameter representing the resonance characteristics of the vocal tract, along with the rock shape prediction coefficient and the PA C0R (partial autocorrelation) coefficient. It is well known that the number of transmitted breasts is a parameter depending on the spectral frequency.

This more precise LSP coefficient is a parameter expressed in the frequency domain, and unlike the α parameter, etc., which is expressed in the time domain, it is a quantity that can be handled more intuitively, and can be synthesized with a small amount of information. The LSF coefficients are used as analysis and synthesis parameters to determine the transfer function of the vocal tract filter, and are used to analyze input speech signals. The LSP type vocoder that performs synthesis is also configured to have the above-mentioned characteristics.

The LSP type pattern matching vocoder that uses this LSP type vocoder uses the LSP coefficients analyzed by an LSP analyzer and the standard pattern regarding the distribution content of the standard LSF coefficients of the voice obtained by LSP analysis of the voice rent in advance. The system selects the closest standard pattern that maximizes the similarity between the two, and transmits this to the synthesis side along with the sound source information to synthesize the input audio signal.The spectral envelope is approximately 10 bits. Recently, it has become well known as a method that allows analysis and synthesis with a low amount of information.It can be easily achieved by adding pattern matching and decoding functions to the j5 and LPC vocoder.

The LSP vocoder in such an LSPg pattern matching vocoder is usually LPC (LinearPr
LSP from the LPG coefficient obtained by the LPG coefficient (Linear prediction coefficient) analyzer
The LSF coefficients are obtained by inducing the coefficients.

Now, as a unit for pattern matching, either a physical unit that focuses on the physical characteristics of the sound, such as the spectral envelope of the input voice, or a linguistic unit that focuses on the linguistic imitation of the voice, can be used. The most efficient pattern matching vocoder configuration is selected according to the Uchidera, and standard patterns are selected by pattern matching using these units as a matching measure. There is a method that deals with gambling elements.

Therefore, for example, a LaP pattern matching vocoder (or one that has the functions of an LPC vocoder), the matching unit is a physical unit, and the selection method is It can be said that it is desirable to use the parameter spatial distance.

Parameter space 1'14m1 used as a pattern matching measure in LSP type pattern matching vocoder
mc, LSP coefficient 4LPC, PARCOR4fi can be regarded as a space vector, and in order to select the standard pattern closest to the LSP coefficient of the input audio signal by comparing the magnitude using the distance between these space vectors as a measure. used. The distance between LSP coefficients, which are space vectors, is the spectral distance Dff shown in (1) 2 below.
It is indicated by ij.

l π Dij= −/ (8i(−−83((4)” dω
−・−−−−(1)π ・(1) Equation can also be converted into an approximate equation as shown in Equation (2) below.

In equations (1) and (2), the number s S l ('4-8 J (- is the frame i and The logarithmic spectrum of j, pk(1), pal is the Nth-order LSP coefficient in frames i and j, %Wk is the Nth-order I, 8P spectral diameter.

The order of the L8F coefficients corresponds to the order of an all-pole digital filter for configuring the vocal tract filter to be realized by the L8F coefficients, and in an N-order all-pole digital filter, the order of N individual rock spectrum ω
1. ω goose, ω3...shows ωN. Also, the Nth L
The SP spectral sensitivity Wk indicates the degree of spectral change caused by a minute change in the N-th order L8F coefficient, and the L8F frequency spectral sensitivity determined in accordance with the LSP frequency is usually used.

Now, in order to select the standard pattern that is most similar to the spectral color glaze of the human-powered back door signal from among the pre-registered semi-patterns, the spectral distance calculation using equation (1) is performed using all the standard patterns over all frames of the input audio signal. However, since the amount of calculation is extremely large, the approximate equation (2) is generally used to measure the so-called simple spectral distance. This is the spatial feature vector of the analyzed input audio signal of order N
The inner product of the SP coefficient pJU and the spatial feature vector pH1l registered in the standard pattern is calculated for each order of LSF coefficient, and the weighting percentage set in advance for each LSF frequency corresponding to the order of the LSP coefficient is calculated. Simple spectral distance measurement is performed by multiplying by Wk.

(Problems to be Solved by the Invention) This type of conventional LSPfi pattern matching vocoder uses the LSP frequency spectrum sensitivity corresponding to the I8P8r number as the weighting coefficient Wk shown in equation (2). ,
LSP frequency spectrum sensitivity with 8r frequency rf! 1
．． Since it differs depending on the number interval, if a standard pattern is selected using the spectral distance measured using such a spectrum m degree as the measure for pattern matching, the sound to be synthesized is 7! ! The disadvantage is that I often deteriorate significantly.

The object of the present invention is to eliminate the above-mentioned drawbacks by providing a system with the ability to preselect a small number of standard patterns and directly compare the difference between the selected standard patterns and the spectral envelope of the input audio signal. , LSP that can significantly improve sound quality deterioration
An object of the present invention is to provide a fi pattern matching vocoder.

(Means for Solving the Problems) The vocoder of the present invention is capable of processing L8P (Lin
e8pectrum Pa1r) In an LSP type pattern matching vocoder that synthesizes an input audio signal by comparing a standard pattern created in consideration of the coefficient distribution with a pattern related to LSP coefficients obtained by LSP analysis of the input audio signal, 1. of the standard pattern. A small number of standard patterns are preliminarily selected by measuring the spectral distance based on the weighted inner product of the SP coefficient and the L8F coefficient of the input audio signal, and the selected li! Calculate the cisvector envelope of the quasi-pattern, and use the calculated spectral envelope and input sound /! The present invention is configured to include means for measuring a difference from a spectrum envelope obtained by analyzing one signal, and selecting a quasi-pattern for which the measured difference is the minimum as a representative pattern.

(Example) Next, the present invention will be described in detail with reference to the drawings.

Fig. 1 (4) is a block diagram showing the first embodiment of the present invention, Fig. 1 (5) is a block diagram showing the structure of the analysis side, and Fig. 1 (b) is a block diagram showing the structure of the synthesis side. .

The analytical diagram 1 shown in FIG. 1 (5) shows that L P F (Low
Pa5sFilter) l l, A/D Coy Parter 12. Window function processor 13. Autocorrelation coefficient meter 14
, LPC analyzer 15. Voiced/unvoiced/silent discriminator 16.
Pitch extractor 17, LSP analyzer 18. Spectral distance measuring instrument 19. Standard pattern memory 20. The synthesis 111!12 shown in FIG. Pattern decoder 25° standard pattern memo 926. LbP synthesizer 27. Variable gain multiplier 1M unit 28. Switcher 29. Pulse generator 30° noise generator 31. It is configured to include a D/mu converter 32 and an 0LPP 33.

In FIG. 1 (5), the human voice signal input via the input line 111 is filtered for frequency components in a predetermined analysis band by the LPFII, and the output 2-in 112
The signal is sent to the A/D converter 12 via the A/D converter 12, where it is digitized with a predetermined number of bits, and then sent to the window function processor 13 via the output line 121 as a quantized audio signal.

The window function processor 13 processes 30m5EC of the input audio signal.
Window function processing is performed in which each frame is multiplied by a Hamming function, and this window function processing is repeated at a cycle of 10m5EC, which is defined as the basic frame cycle.

The audio waveform data of the input audio signal subjected to the window function processing is sent to the autocorrelation coefficient measuring device 14 via the output line 131 for each basic frame.

An autocorrelation coefficient meter 61j from 14 manufacturers measures the autocorrelation coefficient at each delay time within the required delay time using a multiplication circuit or the like using the input audio waveform data, and calculates the autocorrelation coefficient 91.
LPC analyzer 15 outputs numerical data via 2-in 151
In addition, it is sent to the voiced/unvoiced/silent discriminator 16 and the pitch extractor 17 via the output line 152, and
The autocorrelation coefficient at the delay time is determined and sent to the encoder 23 via the output line 153 as short-time audio power data for each basic frame.

The voiced/unvoiced/silent discriminator 16 uses the input autocorrelation coefficient data to determine whether the audio signal included in each basic frame is voiced, unvoiced, or completely silent, and determines whether it is voiced/unvoiced/silent. It is sent as data to the encoder 23 via the output line 161, and the pitch extractor 17 uses the input autocorrelation coefficient data to extract the pitch data of the audio signal included in each basic frame and outputs it. It is sent to the encoder 23 via the 2-in 171.

The LPC analyzer 15 constitutes a variable-length frame LSP analysis circuit together with an LSP analyzer 18 (to be described later). Based on the voiced/unvoiced/silent discrimination data received via the line 162, the frame is divided into voiced sections corresponding to voiced and unvoiced, and other silent sections, and a variable length transmission frame is set in advance for each of these two sections. is set. in this case,
The LPG analyzer 15 uses the autocorrelation coefficient of each input frame to calculate linear prediction coefficients up to a predetermined order (in the case of this embodiment, up to the 10th order) using the well-known Levinson method, and outputs the linear prediction coefficients to the output line. 154 to the LSP analyzer 18, and the LSP analyzer 18 converts the linear prediction coefficients into 10th-order L8F coefficients by a higher-order equation method using Newton's iterative method, and further converts the linear prediction coefficients into 10th-order L8F coefficients for each basic frame. This LSP8P coefficient with a constant period is converted into a variable frame length with a variable period by an optimal approximation method using an approximation function set in advance while using information from voiced/unvoiced/silent discrimination data input via the output line 162. do.

In addition, as pre-processing for such LSP analysis, pre-emphasis processing is performed to pre-emphasize high-frequency components in the waveform region using the first-order difference of the waveform in order to emphasize the high-frequency components of the input audio data. , and Lag window processing using Lag@ numbers in the autocorrelation coefficient region, but these preprocessings are performed using LSP8
This is done by widening the minimum interval of the P coefficients and increasing the stability of the LAF synthesis-270 all-pole digital filter in synthesis 1112, which will be described later, by reducing the LSP quantization sensitivity in the low range.

Now, the 10th-order LSF coefficient obtained in this way is sent to the spectral distance meter side unit 19 via the output line 181. In addition, from the LSP analyzer 18, frame change rate information obtained by expanding or contracting the basic frame length when forming a variable length frame, so-called repeat bit data, is output to an output line 182.
The data is sent to the encoder 23 via the encoder 23.

The lO-order LSP coefficients sent from the LAF analyzer 18 to the spectral distance measuring device 19 via the output line 181 are:
The spectral distance measuring device 19 calculates the so-called simple spectral distance using the approximate equation of equation (2). ) K corresponding lO-order LSP coefficients and the % imitation vector of the standard pattern registered in the standard pattern memory 20, that is, pk(j) K
The inner product with the corresponding standard 10th-order LSP coefficient is first calculated as yL as shown in equation (2), and this inner product is multiplied by the frequency spectral sensitivity Wk as a weighting factor to calculate the 10th-order LSP coefficient from the 1st-order LSP coefficient. Standard pattern memo for each variable length frame of the input voice signal, up to the LSP coefficient! The spectral distance Dij is determined, and the pattern with the smallest spectral distance Dij is selected as the standard pattern for each variable length frame. Such standard patterns are sequentially sent to the encoder 23 via the output line 191 as standard pattern designation code data that designates the standard pattern registration address code in the standard pattern memory 20.

In the case of this embodiment, the standard pattern registered and stored in the Nukahaya pattern memory 20 is created in advance by a separate input process as follows.
There is no problem even if this is created in advance using the vocoder according to this embodiment.

First, pre-saturation such as removal of silent sections, removal of unnecessary adjacent frames, and voiced, silent, and silent assistance is performed using a method such as LPC analysis using preset audio data.

In this case, the frame period is 10 m8 EC, and a tag code is given to each frame to indicate whether it belongs to voiced, unvoiced, unvoiced, or voiced/unvoiced boundary sound. Next, silent frames are removed and the remaining frames are separated into voiced and unvoiced frames, and at this time, boundary sounds are included in either or both of the voiced and unvoiced frames.

Furthermore, frames that are close in time and have a small spectral distance are removed, thus reducing the number of required samples. The patterns are classified by distance and registered and stored as standard patterns.

In the case of this example, the standard pattern method described above is 10-dimensional L
Assume that the space υ of 8F coefficients consists of N(- patterns,
For each of these N patterns, the spectral distance is measured using the following formula, and the number of N patterns that has a preset spectral distance range value #dBj is determined, and the number of patterns Mi = (i =1.2...
After determining the pattern PL with the maximum Mi among . is registered as a standard pattern, and this operation is repeated until there are no more patterns including the space UK, and the pattern is registered as a standard pattern.

Further, the contents stored in each frequency spectrum sensitivity memory 21 are determined as follows.

The spectral sensitivity of the second (Kth) element Pk of the LSP obtained by actually measuring the audio material using equation (1) is as follows (
3) Required by Ni.

/(ΔPk)”dB”/? Jian......(
3) In equation (2), ΔPk is a minute change in Pk, and 1.
8i (→ in this case P,,P,,...Pk...
...spectral envelope Wts 8j (
b) is P, ,P, ,...Pk+ΔPk...
...The spectrum envelope obtained from PL is used.

Therefore, according to equation (3), when ΔPk is a preset value θ radian, the LSP frequency spectrum sensitivity for each frequency of the 10th-order LSF coefficient can be obtained.

In pattern matching, the spectral distance between the LAF analysis data of the input audio signal and the standard pattern is calculated using equation (2) using the frequency spectral sensitivity obtained as a weighting factor, and the spectral distance is calculated from the one with the minimum spectral distance to the smallest one. A desired number of standard patterns are searched for each variable length frame in order of size, and these five * quasi-pattern data (10th order 1.8P) are
and standard pattern designation code data are output to the standard pattern selector 22 via the output line 191.

Standard pattern selection 622 is the most important part of the invention 64
), its detailed operation will be described later, but generally it has the following functions. The ms pattern selector 22 calculates a spectral envelope from a desired number of standard patterns preselected by the spectral distance measuring device 19, and calculates the spectral envelope from the spectral distance measuring device 19,
Calculating the difference in spectral envelope calculated from the edge shape prediction coefficients supplied by the LPC analyzer via the L8F8P device,
Standard pattern designation code data corresponding to the standard pattern with the minimum difference is output to the encoder 23.

The encoder 23 encodes each data thus supplied in a preset code format and transmits it to the transmission path 2.
31 to the combining side 2.

In the synthesis 114112, each pole encoded information input via the transmission line 231 is decoded and a standard pattern is specified.

Code data is input to the pattern decoder 25 via the input line 251, repeat bit data is input to the L8F8P unit 27 via the input line 271, and short-time audio power data is input to the input line 281 to the variable gain amplifier 28, voiced/unvoiced. /
Silence discrimination data and pitch data are supplied to switch 29 and pulse generator 30 via human power lines 291 and 301, respectively.

The pattern decoder 25 extracts the standard pattern specified by the input standard pattern designation code data from the standard pattern memory 26 via the output line 261 and sends it to the LSF synthesizer 27 via the output line 252. I
The I semi-pattern memory 26 is almost the same as the standard pattern memory 2o in the analysis $111), and the data supplied to the LSP synthesizer 27 by the pattern decoder 25 is the input audio signal as a result of pattern matching in the analysis. L8 according to a standard pattern selected for each variable length frame corresponding to the content of
This is an F8F coefficient, that is, an LSF frequency sequence.

The LSP synthesizer 27 restores the variable length frame containing the LSF coefficient sequence inputted in this manner into the original basic 7 frame units using the repeat bit data received via the input line 271, and uses an approximation function to set this in advance. Then, the LSP coefficients are interpolated at the sampling interval of the input audio signal, that is, at the sampling period in the synthesis@1 window@number processor 14.

The L8F coefficients for each basic frame subjected to interpolation processing in this manner are supplied as filter coefficients of a 10th-order LSP speech synthesis digital filter based on an all-pole model.

The LSP voice synthesis digital filter performs calculations as a voice synthesis digital filter using the input filter coefficients and the sound source excitation power input from the variable gain amplifier 28 via the output line 282, and outputs synthesized voice in digital format. and output this to the output line 272t-
The above-mentioned sound source excitation power, which is transmitted to the L)/A converter 32 via the input audio signal, corresponds to so-called residual power obtained by removing the spectral envelope component from the input audio signal. Necessary sound source information is added together with the L8F coefficient as an envelope component, and this is generated as follows.

Short-term audio power data for each basic frame input via input line 281 is supplied to variable gain amplifier 28 .

On the other hand, the pulse generator 30 receives pitch data via an input line 301, generates a pulse with a preset frequency corresponding to the pitch data as a pitch pulse, and sends this to the switch 29 via an output line 302. do.

The switch 29 receives voiced/
When the unvoiced/silent discrimination data specifies voiced, the pitch pulse described above is selected, and when unvoiced or silent, the white noise output from the noise generator 31 is switched to be input via the output line 311. Do the following. Pulse generator 3 selectively output by switch 29
0 or the output of the noise generating fF31 is output on the output line 29.
2 to the variable gain amplifier 28, where it is variably amplified so as to receive weighting corresponding to the magnitude of the short-time voice power data input through the input line 281, and is sent to the output line 282 as sound source excitation power. be done.

ζThe digital synthesized voice signal outputted from the LSP synthesizer 27 is then converted into an analog signal by the D/A converter 32, filtered for a required band by the LPF 33, and sent to the output line 331 as a synthesized voice signal. .

In this way, analysis 1 synthesis of input audio signals can be easily performed through pattern matching using spectral distances measured using LSP frequency interval spectral sensitivities as weighting factors.

Next, the operation of the standard pattern selector 22 will be explained in detail. FIG. 2 is a block diagram for explaining the operation of the standard pattern selector 22 in detail.

The desired number of standard pattern data preselected by the spectral distance measuring device 19 is sent to ω/
The standard pattern designation code data is supplied to the α converter 40 and the 2-bell memory 41, respectively.

In addition, the standard pattern data is based on the measurement results with the spectral distance measuring device 19, and in order from the one with the minimum distance,
The distance is output as the distance increases. Standard pattern data input to the ω/α converter 40 microprocessor in the order in which the % spectral distance increases is recorded at a predetermined address. The ω/α converter also uses the recorded standard pattern data (
The 10th order LSP) is converted into a 10th order α parameter, and the conversion results are output to the α parameter memory 42 in the order in which the spectral distance increases. Note that the method for converting the L8F coefficient into the α parameter is as follows. Itakura et al. have shown that the LSP coefficient is ωi in (3) 2 below. (Voice Study Group Material 879-46, Formula 10) % Formula % Therefore, the LSP is converted into a sea parameter according to the steps ① to ② below.

■Pp(Z)=(1-Z-”Xl 2cos a+1Z
-'+Z-") (1-2cosm4Z-1+Z-")
-----(1-2cos 0)16z-"+Z-"
)=(1-Z-凰)(1+PtZ-'+P1Z-''-1
"...+P1@Z-") (6) However, pt=p
t・P snow=P...P,=P.

Teh ”) PihaPplZ)/ (1-Z-”)
Coefficient Qp(Z) when expanded = (1+Z-') (1-2cos ω
,Z-"+Z-") (1-2cos#5z-1+z
=), -----(1-2c08 m, z-t+z-s
)=(1+Z-”)(1+qlZ-”+qSnow Z-”4”
+qto+Z-”) (7) However, qt=Qt・
Qs=q...qs=(1・adashi qi is Qt)/
Coefficient when expanding (1+Z-1)"(2+(Pt+qt)Z-'+(P*+ql)Z-
"+...+=Ding(P,.+qt.)Z-" ) -'Z-" ((P
t −qt )Z−”+(Pg−qs)Z−”+”・+
(Plo Qto)Z-”)=1+4(Pt+qt
)Z-”+ 2 ((Ps+qs)-(Pt Qt)
) Z-S + -... (8) The coefficient of Kz-i is the α parameter αi.

Again in Figure 2, the spectral distance measuring device 19°LSP
The linear prediction coefficients (ai, i=1.2...10) supplied to the LPC analyzer 15 via the analyzer 18 are input to the spectral envelope calculator 43. The spectral envelope calculator 43 is a microprocessor that calculates discrete spectral envelope data Fi(N) (N=o, 1..
..., 101) is calculated. This method was co-authored by Messrs. Saito and Nakata, 1. Fundamentals of Speech Information Processing, Ohmsha, 1930.
Chapter 7 1 Spectrum Estimation” page 96 of November 2006. The calculated Fi(N) is the output line 43.
1t- to the spectral envelope memory 44.

The spectral envelope memory 44 records the Bi(N) and outputs it to the spectral envelope difference calculator 45 as required. α
The parameter memory outputs the spectral distance minus 0 button α parameter in the spectral distance measuring device 19 to the spectral envelope calculator 43 from time to time. The spectral envelope calculator 43 calculates discrete spectral envelope data (') (N) (
N=o, l,...101) (however, j=1
, 2, . The spectral envelope difference calculator 45 calculates the following spectral distance DI from ]Fi(N) and P/'2N) and outputs it to the minimum distance pattern searcher 46.

The minimum distance pattern searcher 46 becomes m1n(DI)! (
The order in which the α parameters are supplied is determined, and the data j is output to the label memory 41. Label memory 41 has data jK
System vector distance measuring device 19] Encoder 23 converts the standard pattern designation code data of the standard pattern with the smallest spectral distance among the preselected standard patterns.
Output to.

Although the number of patterns to be preselected has been described as a desired number, it may be a measured number or a variable number. If the number is variable, the LSP obtained by analyzing the input audio signal
The number of preliminary selection patterns can be determined using the minimum interval of parameters and the spectral distance trend in preliminary selection.

On the analysis side in each of the above embodiments, LSF analyzer 1B
The LSP coefficients obtained by
Although the coefficients are analyzed and extracted in a variable length frame, the variable length frame may be a fixed length frame if desired.

In addition, the pre-emphasis processing and Lag function processing performed as pre-processing of the L8F8F coefficients are performed as desired, taking into account the characteristics of the input audio signal to be analyzed and synthesized, the contents of the speech synthesis digital filter, and the distribution of the number of data bits. It is clear that there is a choice to be made with or without.

Furthermore, in each of the above-mentioned examples, analysis and synthesis are performed using the 10th order L8F coefficient, but L8F
It is obvious that the coefficients can be implemented in other orders as well.

(Effects of the Invention) As explained above, according to the present invention, in the %LaP pattern matching vocoder, the spectral distance between the LSP coefficient of the standard pattern and the LAF coefficient of the input audio signal is
The spectral sensitivity of the coefficient is calculated, multiple standard pattern candidates are preselected, and from the preselected standard pattern candidates, the best standard pattern is calculated based on the spectral distance calculated through the actual spectral envelope data. By selecting , it solves the problem that the optimal standard pattern is not necessarily selected because the LSP frequency spectrum sensitivity differs depending on the LSP frequency interval, and also minimizes the increase in the amount of computation by limiting the pattern candidates by preliminary selection. It has the effect of limiting

[Brief explanation of the drawing]

Figure 1 is a block diagram showing the structure of the analysis side (5) and the synthesis side of the LaP pattern matching vocoder according to the first embodiment of the present invention, and Figure 2 is a block diagram showing the configuration of the analysis side (5) and synthesis side of the LaP pattern matching vocoder according to the first embodiment of the present invention. FIG. 2 is a block diagram for explaining in detail the important standard pattern selector 22. FIG. l... Analysis side, 2... Synthesis side, 11.
...LPF. 12...Mu/D converter, 13...
Window function processor, 14...Autocorrelation coefficient measuring device,
15... LPC analyzer, 16... Voiced/unvoiced/focused discriminator, 17... Pitch extractor,
18... LSP analyzer, 19... Spectral distance measuring device, 20... Standard PA/Y memory, 21... Frequency interval spectral sensitivity memory,
22...Standard pattern selector, 23...
Encoder, 24... Decoder, 25...
Pattern decoder, 26...Standard pattern memory, 2
7...LSP synthesizer, 28...Variable gain amplifier, 29...Switcher, 30...
Pulse generator, 31...Noise generator, 32...
...D/A converter, 33...LPF,
40...ω/α converter, 41... Label memory, 42... Parameter memory, 43.
... Spectrum envelope calculator, 44 ... Spectrum envelope memory, 45 ... Spectrum envelope difference calculator, 46 ... Minimum distance pattern search device. Agent Patent Attorney Susumu Uchihara1,-,',:,y- ゝ-11,,-

Claims (1)

[Claims] LSP (Line Spectrum Pa) of audio materials
ir) In an LSP amount pattern matching vocoder that synthesizes an input audio signal by comparing a standard pattern regarding coefficient distribution with a pattern regarding LSP coefficients obtained by LSP analysis of an input audio signal, the LSP coefficients of the standard pattern and the means for preliminarily selecting a small number of standard patterns by measuring a spectral distance by a weighted inner product with an LSP coefficient of an input audio signal; and analyzing a spectral envelope signal calculated from the preselected standard patterns and the input audio signal. and means for selecting a standard pattern using a spectral distance calculated from the spectral envelope signal obtained by the LSP pattern matching vocoder.