JPH09171400A - Sound signal band compression transmission method, sound signal reproducing method and sound signal band compressing/expanding device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To bury also a pitch waveform of a voiced section and a system parameter of a voiceless section being substantial information incorporated in a sound signal into an analog waveform and to obtain band compression transmission by an analog transmission signal system. SOLUTION: In the voiced section of the sound signal, the pitch waveform is segmented by a pitch detector 103, a buffer memory 105 and a voiced/ voiceless switcher 112, then is expanded on a time base, and a frequency band is compressed and transmitted. In the voiceless section, the system parameter is obtained by a linear predictive analyzer 106, and a narrow-band base signal consisting of a repeat waveform of a fixed period is generated from an inverse filter 107, a power detector 108 and a base signal generator 109 to synthesize waveform as a drive signal by a linear predictive synthesizer 110. Then, both are transmitted to a reception side as a narrow band analog signal through voiced/voiceless switcher 112. The reception side restores it to the sound signal by inverse processing.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a band compression transmission method capable of band compression of a voice signal in an analog waveform state, and particularly a voice signal band compression transmission method suitable for analog transmission in a narrow band wireless transmission line. The present invention also relates to an audio signal reproducing method and an audio signal band compression / expansion device.

[0002]

2. Description of the Related Art In recent years, the utilization rate of wireless transmission lines has been increasing, but on the other hand, the radio frequency band is a finite resource. Therefore, compression of the occupied frequency band is not only for cost reduction, There is also a strong demand for effective use of resources.

Looking at the transmission of a voice signal, the frequency bandwidth of the voice signal generally extends over several kilohertz although there are individual differences. Therefore, for transmission of this, a transmission system having a frequency bandwidth of several kilohertz is also used. , But without sacrificing the intelligibility necessary for voice communication here,
If the occupied frequency bandwidth can be compressed, the cost required for the transmission system and the frequency bandwidth can be reduced.

Therefore, various voice signal band compression techniques have been conventionally proposed. As an example, a human voice function is regarded as a kind of autoregressive system and a voice signal is generated by this autoregressive system. There is known a technique in which the band compression of the voice signal is obtained by simulating as described above and extracting the system parameter by the predictive analysis. The example is as follows.

Nikkei Electronics, 2.12, PP58-7
5,1973 "New speech analysis and synthesis method" PARCO
R ””, Proc. ICASSP 85, PP937-940,1985 “Code-excited
linear prediction (CELP) "

[0006]

However, although the above-mentioned conventional technique transmits the extracted system parameters, the system parameters are obtained as digital numerical information, which is very effective in reducing the bit rate of the digital transmission system. However, there was a problem in the application to the analog signal transmission system.

An object of the present invention is to obtain the result of band compression processing as an analog waveform, enable band compression transmission by an analog signal transmission system, use the analog signal transmission system, and improve the clarity of a voice signal. The original frequency band signal is reproduced from the narrow band signal by compressing and transmitting the occupied frequency band without loss.

[0008]

SUMMARY OF THE INVENTION In the voiced sound section of a voice signal, the above-mentioned object is to expand the waveform corresponding to one pitch or one pitch or more of the voice signal on the time axis on the transmitting side to compress the frequency band. Converted into an analog waveform and transmitted, the receiving side compresses this on the time axis to reproduce the original speech waveform, and in the unvoiced section, the transmitting side converts the spectrum information of the speech signal into a narrow-band analog waveform. It is achieved by embedding it in a frame and transmitting it, and extracting the spectrum information on the receiving side to reproduce the voice signal.

As a result, the information of the main part of the audio signal is transmitted in a state of analog waveform with sufficient fidelity, high quality,
In addition, highly efficient band compression can be obtained. That is, first,
In the voiced sound section of the voice signal, a waveform having a time length of one pitch cycle or more is cut out, expanded on the time axis, and sequentially connected to form a low frequency narrow band signal, which is transmitted as it is as an analog waveform. In the unvoiced section, a system parameter is extracted by linear prediction analysis, and a narrowband base signal is supplied to a waveform synthesis circuit using the system parameter so that the system parameter is embedded in an analog waveform and transmitted. To do.

A voice signal is roughly divided into a vowel or voiced sound section and a consonant or unvoiced sound section, each of which has the following remarkable features.

First, the voiced sound section is composed of a repetitive waveform of a constant cycle called a pitch cycle, and energy concentration is observed in a relatively low frequency range, and even if the high frequency part is lacked, a decrease in clarity occurs. Absent.

On the other hand, although the periodicity is not recognized in the waveform of the unvoiced section, when linear prediction analysis is performed, concentration of spectral energy called formant is observed on the frequency axis, and the prediction residual signal becomes white noise. It will be close. Speech information is mainly formant-dependent, and even if the prediction residual signal is replaced with other white noise, the clarity is not significantly degraded.

Therefore, effective band compression is executed for each of the voiced sound section and the unvoiced sound section based on the characteristics of the voice signal as described above.

First, in the voiced sound section, a signal having a time length d times (d ≧ 1) the pitch cycle (hereinafter referred to as an element signal) is cut out and expanded to C (C: integer) times the pitch cycle on the time axis. To do. When the upper limit frequency of the original signal is fm, the upper limit frequency of the obtained signal is compressed to fmd / C. Further, paying attention to the fact that the intelligibility is not lowered even if the high frequency part is removed in the voiced sound section, the upper limit frequency of the signal obtained by limiting the entire band to 1 / m (m> 1) is fmd / mC. Is compressed to. Practically, for example, if d = 1, C = 5, m = 2, then d
/ M · C = 1/10, and a high compression rate of 1/10 can be obtained.

Since there is no pitch in the unvoiced section, different measures are required. Audio signal y (n to be transmitted
The upper limit frequency of Δt) is fm. Where Δt = 1 /
In 2 fm, y (n · Δt) represents the value of the audio signal at time n · Δt (n is an integer). Now, a case of using a linear prediction coefficient as a system parameter will be described as an example. A linear prediction analysis is performed on a speech signal to obtain a linear prediction coefficient ai (i = 0, 1, 2, ..., N-1) and a prediction residual signal. Get x (nΔt). x (nΔt) can be regarded as almost white noise. Here, the base signal x ′ as in the following equation 1
Consider (nΔT).

[0016]

[Equation 1]

The base signal x '(nΔT) has a period of 1 / fo
Is a repetitive waveform of.

This base signal x '(nΔT) is Ao, Φ
o, Ak, Bk (k = 1, 2, ..., K) total 2 (K +
1) Since there are 1 degrees of freedom, the average power of the prediction residual signal and the system parameter ai (i
, 1, 2, ..., N−1) can be used to carry the system parameter information. A case where an autoregressive system is used as an example of the waveform synthesis circuit will be described.
ΔT) is applied to an autoregressive system having ai (i = 1, 2, ..., N−1) as a regression coefficient, and an output signal w (nΔ
T) is obtained.

Since the autoregressive system is linear, this output signal w (nΔT) also does not include high frequency components above fm · d / m · C. Then, this output signal w (nΔ
T) is the value of the output signal at time nΔT (n is an integer), and ΔT = mC / 2fmd.

The voice signal y (nΔt) and the output signal w (nΔT) both have the same linear prediction coefficient ai. However, the upper limit frequency of the audio signal y (nΔt) is fm.
Since the output signal w (nΔT) has the upper limit frequency fmd / mC, ΔT = mC during the prediction sampling interval.
There is a relationship of / d · Δt.

As described above, since the voice signal y (nΔt) and the output signal w (nΔT) both have the same linear prediction coefficient a i, the output signal w having a narrow-band analog waveform.
Only by transmitting (nΔT), the original audio signal y (nΔt)
It is possible to faithfully transmit the spectrum information of the.

However, the spectrum information mentioned here is information in the form of a linear prediction coefficient (system parameter), and is not the frequency spectrum itself. This frequency spectrum itself is generated by the drive signal and the autoregressive system on the receiving side.

The above waveform synthesizing circuit is not necessarily limited to an autoregressive system, and includes all circuits that modulate the amplitude and phase of a sine wave forming a base signal with system parameter information.

[0024]

BEST MODE FOR CARRYING OUT THE INVENTION A voice signal band compression / expansion device according to the present invention will be described in detail below with reference to an embodiment shown in the drawings.

FIG. 1 is a block diagram showing the configuration of the transmission side of the audio signal band compression / expansion device according to the present invention when m = 1.

The audio signal y (t) to be transmitted is input terminal 1
01, first A / D (analog / digital)
The digital signal y (n is sampled by the converter 102.
Δt). Here, the signal y (t) is the value of the audio signal at the time t, and the signal y (nΔt) is the value of the audio signal at the time nΔt (n = integer) as described above.

Here, it is assumed that the upper limit frequency fm of the frequency component of the original audio signal y (t) is 4,000 Hz and the sampling time interval Δt is Δt = 1/2 fm = 125 μs (sampling frequency 8 kHz).

In the voiced sound section, the pitch detection circuit 103 detects the pitch period and the voice signal is stored in the buffer memory 105. The capacity of the buffer memory 105 is controlled by the output of the pitch detection circuit 103, and is set to one pitch or more. When the recording in the buffer memory 105 reaches the limit of the capacity, the recording is resumed by returning to the head address again, so that the buffer memory 105 always holds a waveform of one pitch or more. Buffer memory 10
5 is read at a sampling frequency of 1600 Hz, passes through a voiced / unvoiced switching circuit 112, and an upper limit frequency of 8 by a D / A (digital / analog) converter 113.
It becomes a narrow band signal of 00 Hz and is sent out from the output terminal 114.

The voiced / unvoiced switching circuit 112 is controlled by the pitch detection circuit 103.

FIG. 2 is a conceptual diagram showing time-axis expansion when C = 5.

The voiced voice signal has a periodic waveform which repeats at a pitch period ΔTp. Therefore, if the element signal of time length ΔTp is cut out from the waveform of y (nΔt), the sampling rate is reduced to 1 / C and they are arranged on the time axis, and they are connected successively, the original waveform y (nΔt) is obtained. A waveform y '(n [Delta] T) is obtained by analogically extending y on the time axis.

Next, returning to FIG. 1 again, the voice signal y (nΔt) is regarded as an autoregressive signal in the unvoiced section, and the linear prediction coefficient ai is used as the system parameter, and is defined as Equation 2.

[0033]

[Equation 2]

Here, the first term on the right side represents the sound source signal due to the exhalation in the human vocal mechanism, and can be regarded as almost white noise in the unvoiced section. The second term is also considered to represent the filtering action by the vocal tract. , This A /
The audio signal y (nΔt) output from the D converter 102 is
It is supplied to the linear prediction (LP) analyzer 106 and the inverse filtering circuit 107. First, in one linear prediction analyzer 106, the linear prediction coefficient ai (i = 1, 2, 3, ..., N−1).
Is estimated.

On the other hand, in the inverse filtering circuit 107,
Using this linear prediction coefficient ai, a digital speech signal y (nΔt) consisting of a time-series signal is subjected to calculation by the following equation 3 to obtain a prediction residual signal x (nΔt), This constitutes a linear prediction system.

[0036]

(Equation 3)

The power detection circuit 108 calculates the average power of x (nΔt), and the base signal generator 109 generates a base signal x '(nΔT) having a power proportional to the power. The sampling frequency of the base signal is 1600 Hz
Then, the sampling period is ΔT = 625 μS. The base signal x ′ (nΔT) is expressed by Equation 4.

[0038]

(Equation 4)

The base signal has a period of 1 thus constructed.
It is a repetitive waveform of / fo = 10 ms, and its upper limit frequency is 800 Hz.

Next, the basis signal x ′ (nΔT) is supplied to the linear prediction (LP) synthesizer 110, where the linear prediction coefficient ai = found by the linear prediction analyzer 106.
1, 2, 3, ..., N-1) is used as a regression coefficient, and the base signal x '(n [Delta] T) is subjected to the following autoregressive system operation according to Formula 5 to obtain a narrowband time series signal w (n [Delta] T). .

[0041]

(Equation 5)

Generally, the parameters ai = 1, 2, 3, ..., N-1) obtained when a linear prediction analysis is performed on a voice signal vary from moment to moment, but in order to maintain the clarity of the voice, it is 10 ms. It is said that the value should be updated about once.

Since the repetition period of the base signal x '(nΔT) is 10 ms, ai = 1, 2,
By updating the values of 3, ..., N-1), it is possible to follow the temporal change of the audio signal almost faithfully. Next, the narrowband time series signal w (nΔT) obtained at the output of the linear prediction synthesizer 110 in this manner is used for the voiced / unvoiced switching circuit 11
D / A (digital / analog) converter 113 via 2
To the output terminal 1
The narrowband time series signal w (t) is obtained at 14.

Therefore, the narrow band time series signal w (t) is composed of frequency components of 0 to 800 Hz.

On the other hand, the frequency component of the original audio signal y (t) has the upper limit frequency fm = 4000 Hz as described above. Therefore, according to this embodiment, the frequency range of 4000 Hz is band-compressed into the frequency range of 800 Hz. Will be.

The narrowband analog signal w (t) thus obtained at the output terminal 114 is put on a predetermined signal transmission system, such as a telephone line or a radio channel, and transmitted to the receiving side.

Next, FIG. 3 is a block diagram showing the configuration of the receiving side when the audio signal band compression / expansion device according to the present invention is set to m = 1. The narrow band analog signal transmitted from the transmitting side of FIG. w (t) is supplied to an input terminal 201, first sampled by an A / D (analog / digital) converter 202, and converted into a time-series digital signal w (nΔT).

In the pitch detection circuit 203, w (nΔT)
The periodicity included in is identified and the period is detected.

In the voiced sound section, the capacity of the buffer memory 204 is set according to the detected cycle, and w (nΔ
At T), the waveform for one pitch is always retained while being rewritten cyclically. The contents of the buffer memory 204 are endlessly read at a sampling frequency of 8 KHz, passed through a voiced / unvoiced switching circuit 212, and then D / A.
(Digital / Analog) An analog voice waveform is formed by the converter 213 and is output from the output terminal 214. In the unvoiced section, the time-series digital signal w (nΔT) is applied to the linear prediction analyzer 207 and the inverse filtering circuit 208. First, in the linear prediction analyzer 207, the linear prediction coefficient ai (i = 1, i = 1, 1) by the linear prediction analysis. 2, 3, ..., N-
Restore the value of 1).

On the other hand, in the inverse filtering circuit 208,
Using this linear prediction coefficient ai, the digital audio signal w (nΔT) consisting of a time-series signal is subjected to the operation of the following equation 6 to obtain a reproduction base signal x ′ consisting of a prediction residual signal.
(NΔT) is obtained, which constitutes a linear prediction system.

[0051]

(Equation 6)

In the power detection circuit 209, the above-mentioned x '(n
The average power of ΔT) is detected, and the white noise generator 210 generates white noise x (nΔt) having a power proportional to the average power.

In the linear prediction synthesizer 211, the linear prediction coefficient ai (i = 1, 1 obtained by the linear prediction analyzer 207).
2, 3, ..., N-1) is used as a regression coefficient, and the white noise x (nΔt) is used as a drive signal to perform autoregressive system operation according to the following equation 7 to reproduce audio signal y (time-series signal). We obtain nΔt).

[0054]

(Equation 7)

In this way, the linear prediction synthesizer 2
The reproduced voice signal y (nΔt) obtained at the output of 11 is supplied to the D / A converter 213 via the voiced / unvoiced switching circuit 212, restored to an analog waveform signal, and output terminal 21
The analog voice signal y (t) is obtained at 4.

The drive signal in the unvoiced section corresponds to the turbulence of the air generated when the exhaled air passes through the narrow space formed by the articulatory organs in the process of producing the voice, and as described above, the drive signal is the white noise. Although it can be approximated, further speaking, a prediction residual signal, that is, a sound source signal is sampled from a real voice and prepared as a sample sound source, and an optimum one is selected according to each case, and a drive signal is selected. If used as, the sound quality can be further improved.

Next, each element in the above embodiment will be described. First, the linear prediction analyzers 106 and 207 are
For example, processing is performed according to the algorithm shown in FIG. 6, the autocorrelation function of the speech signal Sn is calculated, and the coefficient ai
It has a function of determining (i = 1, 2, 3, ..., N-1). That is, the audio signal S by the autocorrelator
n is delayed by the delay element Z ~ ¹ and the original speech signal is subjected to correlation calculation, and the short-term autocorrelation numbers R ₀ , R ₁ , R ₂ , ... Of each data are calculated.
Ask for. The outputs R ₀ , R ₁ , R ₂ , ... Of the autocorrelator are put into the following simultaneous linear equations to obtain linear prediction coefficients a ₁ , a ₂ ,
a _3, ... we seek.

In order to understand the present invention, details of this linear prediction analyzer are described in, for example, June 10, 1980 (Showa 55), published by Kobo Publishing Co., Ltd., "Computer Speech Processing"<Electronic Science. Series> pp. 43-50.

Further, in the case of the linear prediction analyzer 207, the basis signal is known and is a set of sinusoidal waveforms.
The amplitude of each sine wave may be obtained by FFT analysis, and the coefficient ai (i = 1, 2, 3, ..., N-1) may be estimated from this.

Next, the inverse filtering circuits 107, 20
The inverse filtering process of 8 is a process of calculating a residual signal, for example, x (nΔt) from the coefficient ai (i = 1, 2, 3, ..., N−1) described above in advance. Then, the calculation is performed according to the above-described equation (3).

Further, the linear predictive synthesizers 110 and 211 perform arithmetic operations according to the above equation 5, and for example, FIG.
By the processing shown in (1), it has a function of synthesizing an audio signal using a drive signal. That is, it is possible to obtain a synthesized speech signal yn by weighting synthesizing speech signals delayed by the delay element Z ~ ¹ coefficients _{a 1, a 2, a 3} , ... in.

The details are described, for example, on pages 50 to 53 of "Computer Audio Processing"<Electronic Science Series>, published by Koho Publishing Co., Ltd., on June 10, 1980 (Showa 55). ing.

Further, the linear predictive synthesizer 110 may be configured as a modulation circuit that modulates the amplitude and phase of each sine wave forming the base signal according to the system parameter.

In the embodiment described above, d = 1, m = 1,
When C = 5, the compression rate was 1/5, and it was shown that the voice with the upper limit frequency of 4 kHz is compressed to the upper limit frequency of 800 Hz.

Next, an embodiment in which d = 1, m = 2, C = 5 and the compression rate is 1/10 will be described.

FIG. 4 is a block diagram showing the configuration of the transmitting side in the embodiment when m = 2 of the audio signal band compression / expansion device according to the present invention, which has almost the same configuration as FIG.
Since the operation of the blocks denoted by the same reference numerals in both figures is exactly the same, the description of those parts will be omitted.

In the voiced sound section, the pitch detection circuit 103 detects the pitch period and the voice signal is low-pass filtered.
In 04, the upper limit frequency was limited to 2000 Hz, the sampling frequency was lowered to 4 kHz, and the buffer memory 10
5 The capacity of the buffer memory 105 is controlled by the output of the pitch detection circuit 103, and is set to one pitch or more. When the recording in the buffer memory 105 reaches the capacity limit, the recording is resumed by returning to the head address again, so that the waveform of one pitch or more is held in the buffer memory 105. The contents of the buffer memory 105 are read at a sampling frequency of 800 Hz,
After passing through the unvoiced switching circuit 112, a D / A (digital / analog) converter 113 converts the signal into a narrow band signal having an upper limit frequency of 400 Hz, and the signal is sent from the output terminal 114.

The base signal x '(n generated in the unvoiced section
ΔT) is expressed by Equation 8.

[0069]

(Equation 8)

Therefore, the cycle is 1 / fo = 10 ms
However, the upper limit frequency is 400 Hz. Therefore, a linear prediction (LP) synthesizer 11 that performs linear prediction synthesis based on this
The 0 output also has a band of 0 to 400 Hz. The narrowband time-series signal w (nΔT) having a band of 0 to 400 Hz obtained in this way is supplied to the D / A (digital / analog) converter 113 via the voiced / unvoiced switching circuit 112 to be converted into an analog signal. Restored to a waveform signal, output terminal 1
A narrowband analog signal w (t) is obtained at 14.

Therefore, this narrow band analog signal w (t)
Regarding, it consists of frequency components from 0 to 400 Hz.

On the other hand, the frequency component of the original audio signal y (t) has the upper limit frequency fm = 4000 Hz as described above. Therefore, according to this embodiment, the frequency range of 0 to 4000 Hz is changed to the frequency range of 0 to 400 Hz. Bandwidth will be compressed.

In the voiced sound section, the pitch waveform is limited to 2 kHz or less, but for the reasons described above, the decrease in clarity is slight.

Next, FIG. 5 is a block diagram showing the structure of the receiving side in the embodiment when m = 2 of the audio signal band compression / expansion device according to the present invention, which has almost the same structure as FIG. The operations of the blocks denoted by the same reference numerals are exactly the same, so the description of those parts will be omitted.

In the voiced sound section, the capacity of the buffer memory 204 is set according to the detected cycle, and w (nΔ
While rewriting cyclically in T), the waveform for one pitch is always held. The contents of the buffer memory 204 are read endlessly at a sampling frequency of 4 kHz, and 0 is added between each sample, and then the low-pass filter 2
In 05, a component of 2 kHz or less is extracted and becomes a signal having a sampling frequency of 8 kHz, and the voiced / unvoiced switching circuit 2
D / A (digital / analog) converter 2 via 12
At 13, an analog voice waveform is formed and sent from the output terminal 214.

By the way, in the above embodiment, the audio signal y (nΔt) is defined by the above-mentioned equation 2 and the prediction coefficient ai
It is said that the prediction analysis is to obtain (i = 1, 2, 3, ..., N−1), but the present invention is not limited to this, and the prediction analysis process in the present invention is not limited to this. Not something.

Generally, a voice signal is described by the Z conversion method,
Number 9,

[0078]

[Equation 9]

Although various methods are known for identifying the equation 10 on the assumption that the above is true, the predictive analysis in the present invention includes all of them.

[0080]

(Equation 10)

The linear prediction system according to the present invention is expressed by

[0082]

[Equation 11]

Means all systems that obtain x (z) from y (z) by the autoregressive system.
It means any system that obtains y (z) from x (z) by.

As one example, one of the well-known ones is a partial autocorrelation coefficient (k parameter) km (m = 1, 2, ...).
..., N-1).

This is a linear prediction coefficient ai (i = 1, 2,
, N−1) and a recurrence relation such as Equation 21.

For example, March 2, 1988 (Showa 63)
5th See "Signal analysis and system identification", pages 55-57 and 72-76, published by Corona Inc.
The linear prediction coefficient ai (i = 1, 2, ..., N-1) is a simultaneous equation

[0087]

(Equation 12)

It is obtained as the root of.

Now, in general,

[0090]

(Equation 13)

If we put, the above equation 12 is given by

[0092]

[Equation 14]

Is obtained. Also, R _k is

[0094]

(Equation 15)

Therefore, if m−k is substituted for _k , R _k = R_ _k

[0096]

(Equation 16)

Substituting the equations 15 and 16 into the equation 14, the equation 17

[0098]

[Equation 17]

That is, equation 18 is obtained.

[0100]

(Equation 18)

Substituting for m-i instead of i, the following equation 19,

[0102]

[Equation 19]

[0103] sides people phase reduce the number 20 with the multiplied number 18 k _m on both sides of the number 19,

[0104]

(Equation 20)

That is, Equation 21 is obtained.

[0106]

(Equation 21)

Since the k parameter is always in the range of -1≤km≤ + 1, there is an advantage that the dynamic range of the signal whose waveform is synthesized by using the analog base signal falls within the limited range. As an example of the waveform synthesis algorithm,
When the composite waveform is represented by the equation 8,

[0108]

(Equation 22)

By making the correspondence as described above, the k parameter can be immediately restored by the FFT analysis on the receiving side. Also, bm
(M = 1, 2, ..., 6) is +1 corresponding to the polarity of the amplitude.
Alternatively, since the binary signal takes a value of -1, it can be used as a digital code for designating the optimum sample excitation signal.

The block diagrams of the transmitting side and the receiving side of the embodiment in which the K parameter is adopted and is restored by the FFT analysis on the receiving side are shown in FIG. 8 and FIG. 9 of the linear predictive synthesizer using the K parameter. An example is shown in FIG.

Next, when the time length of the phoneme in the voice section is examined, the vowel does not change much at any vowel, and the average is about 70 m.
It is said that the consonant changes in the range of 5 to 130 ms depending on its type. Vowel pitch period is 2
Since it is distributed in the range of 20 ms, C = 5 as in the above example.
When expanded on the time axis, the number of frames corresponding to one vowel phoneme is about 1 to 7. From such a small number of frames, it is technically very difficult to detect the periodicity and cut out the pitch waveform. In the consonant section, the frame period for transmitting parameters is 10m.
s, and in this case as well, the number of frames corresponding to consonant phonemes is about 1 to 13, which causes the same problem as in the case of vowels. The following method is proposed as a method for solving these problems.

First, in the silent section, Ao = 0 according to the general definition of Equation 22, but dare to set Ao> 0 and Ak
= Bk = 0 (m = 1, 2, ..., K). as a result,
Since the sine wave having the frequency f0 is transmitted even in the silent section, the receiving side can detect this and establish frame synchronization in advance. However, at the same time, Ak = Bk = 0 (m
, 1, 2, ..., K), and the reproduced voice output is kept at 0.

In the unvoiced sound section, it is possible to easily maintain the synchronization by continuing the frame synchronization as it is. Ao is kept at Ao> 0.

In the voiced sound section, the frame period is C times the original pitch period, and a constant value cannot be expected. Therefore, at the start of the voiced sound section, the value of the pitch period is transmitted using the frame of the formula 22 before the pitch waveform is transmitted. Therefore, Ao <0, that is, the phase of the frequency f ₀ is inverted only for the frame, and b
The value of the pitch period is transmitted using the value of _m (m = 1, 2, ..., N−1). On the receiving side, when phase inversion of the frequency f is detected, it is interpreted that b _m (m = 1, 2, ..., N−1) represents the pitch period, the division points of the pitch waveform are estimated in advance, and Search for an autocorrelation peak in the vicinity,
Frame synchronization can be easily maintained by determining the division points of the pitch waveform.

Further, by using this method, it is possible to eliminate the transmission delay at the rising time of the vowel. FIG.
It is a figure which shows the time relationship of the rising time of a vowel. FIG.
When the vowel rises from the consonant or the silent state at time T ₁ in _{No. 1} , there is a certain time delay until the pitch is detected and the rise of the vowel is recognized, so that the transmission of the pitch waveform is actually started. Becomes time T ₂ . Since the pitch waveform is expanded on the time axis, it is delayed at time T ₃ to receive the reproduced sound by receiving it.
When the lower limit of the pitch frequency is 50 Hz, the time difference between T ₂ and T ₃ is 80 ms when C = 5, which is a value that cannot be ignored in practice. According to the above method, the parameter information is transmitted until time T ₂ , and the pitch period information is transmitted immediately before the start point of the pitch waveform transmission. Then, first, a repetitive pulse train having a period equal to the pitch period is created, and if this is used as a drive signal and linear prediction (LP) synthesis is performed using the above parameters, a pseudo vowel waveform A ′ can be produced, and at least at time T ₂ . Gives a reproduced vowel.

In the description of the above embodiments, regarding reproduction of the voiced sound section on the receiving side, the received element waveform is compressed to 1 / C on the time axis, and this is repeated C times to reproduce the reproduced voice waveform. I'm supposed to get it. In this method, the resolution of the voice is maintained, but after the same waveform is regularly repeated C times, a rapid waveform change occurs, so that the naturalness of the sound quality of the reproduced voice is lost and a sense of discomfort occurs. . Therefore, it is possible to improve the sound quality by providing a means for smoothing changes in adjacent element waveforms. FIG. 12 shows an example thereof. In this embodiment, a waveform compressed on the time axis is repeatedly reproduced 2 · C times, weighting is performed to linearly change from 0 to 1 in the first half, and linearly changes from 1 to 0 in the second half. The weighting is performed and the results are added together to smooth the changes in adjacent element waveforms.

In the above description of the embodiment, the time length of the element signal cut out on the transmitting side is one pitch for simplification of the description, but this is not limited to one pitch, and if it is one pitch or more. , Theoretically, the pitch waveform can be reproduced on the receiving side. On the other hand, if the reproduced waveform for one pitch is directly connected as it is, it is unavoidable that discontinuity is actually generated at the connection point, which is one of the causes of sound quality deterioration.
This problem can be solved by setting the length of the element signal to 1 pitch or more. Although theoretically any number may be used as long as it is 1 pitch or more, the case where the length of the element signal is 1.4 pitch will be described as an example in FIG. Prior to reproducing the pitch waveform from the element signal on the receiving side, from 0 to 1 in advance by 0.4 pitch of the leading edge of the element signal (this value is the length of the element signal minus 1 pitch) A linearly changing weighting is applied, and a linearly changing weighting from 1 to 0 is applied to 0.4 pitches of the trailing edge. When the signals thus weighted are superposed by 0.4 pitches and sequentially connected, the discontinuity at the connection points is eliminated and good sound quality can be obtained. Since the length of the element signal is set to 1.4 pitches, if C = 5 on the transmitting side and the expansion rate on the time axis is set to 5 times, the frame length becomes 7 pitches. Therefore, the number of times the pitch waveform reproduced on the receiving side is repeatedly reproduced must be seven times. That is, in this case, the expansion ratio on the transmission side and the number of times of repeated reproduction on the reception side do not necessarily match. It is needless to say that the above-described weighting of the element signals can be performed on the receiving side, but the transmitting side may perform weighting in advance and then transmit, and the receiving side may simply perform superposition.

When the length of the element signal is one pitch or more, this can be utilized to transmit the frame synchronization signal. Following the above example, the element signal length is set to 1.4
Taking the case of the pitch as an example, the length of the waveform cut out from the original signal on the transmitting side is 1.3 pitch, and a total of 1.4 pitch is obtained by adding a synchronization pulse having a length of 0.1 pitch to this. It is sent as a length element signal. On the receiving side, the synchronization pulse is detected to establish frame synchronization, and 1.3 pitches of the pitch waveform are sent, and the pitch waveform is reproduced according to the above method.

[0119]

According to the present invention, the pitch waveform and system parameters used for analyzing and synthesizing the voice signal are embedded in the narrow band analog signal for transmission, so that transmission by the narrow band analog transmission system is performed. It is possible to easily obtain a frequency band compression / expansion device for audio signals capable of performing the above. Further, according to the present invention, the voice signal in the voiced section is
Since the waveform itself is expanded and compressed on the time axis in synchronization with the pitch cycle, the waveform itself is transmitted almost completely and faithfully, and in spite of narrow band transmission, there is no deterioration in clarity and high-quality voice transmission. And the regeneration can be easily obtained.

As described above, since narrow band transmission with high clarity is possible, the cost of the transmission line can be reduced, and limited resources, especially radio frequency band can be effectively used.

[Brief description of the drawings]

FIG. 1 is a block diagram of an audio signal band compression / expansion device according to the present invention.
It is a block diagram which shows the structure of the transmission side in the embodiment when it is set to 1.

FIG. 2 is a conceptual diagram showing expansion and compression on the time axis synchronized with the pitch cycle of the voice signal in the voiced sound section according to the present invention.

FIG. 3 is a block diagram of the audio signal band compression / expansion device according to the present invention, m =
It is a block diagram which shows the structure of the receiving side in the embodiment when it is set to 1.

FIG. 4 is a block diagram of an audio signal band compression / expansion device according to the present invention, m =
It is a block diagram which shows the structure of the transmission side in embodiment when it is set to 2.

FIG. 5: m = of the audio signal band compression / expansion device according to the present invention
It is a block diagram which shows the structure of the receiving side in embodiment when it is set to 2.

FIG. 6 is an explanatory diagram showing an example of a linear prediction analyzer according to the embodiment of the present invention.

FIG. 7 is an explanatory diagram showing an example of a linear prediction synthesizer that uses linear prediction coefficients according to the embodiment of the present invention.

FIG. 8 is a block diagram showing a configuration of a transmission side in the embodiment when the K parameter is adopted in the audio signal band compression / expansion device according to the present invention and the modulated wave of the base signal is made to correspond to this.

FIG. 9 is a block diagram showing the configuration of the receiving side in the embodiment when the K parameter is adopted in the audio signal band compression / expansion device according to the present invention and is restored by FFT analysis.

FIG. 10 is an explanatory diagram showing an example of a linear prediction synthesizer that uses K parameters according to the embodiment of the present invention.

FIG. 11 is an explanatory diagram showing a time relationship at the time of rising of a vowel in the embodiment of the present invention.

FIG. 12 is an explanatory diagram showing an example of smoothing a vowel waveform in the embodiment of the present invention.

FIG. 13 is an explanatory diagram showing an example of eliminating discontinuity of a connection point of a pitch waveform according to the embodiment of the present invention.

[Explanation of symbols]

101, 201 ... Input terminals, 102, 202 ... A / D
(Analog, digital) converter, 103, 203 ... Pitch detection circuit, 104, 205 ... Low-pass filter, 105,
204 ... Buffer memory, 106, 207 ... Linear prediction analyzer, 107, 208 ... Inverse filtering circuit, 10
8, 209 ... Average power detector, 109 ... Basis signal generator, 210 ... White noise generator, 110, 211 ... Autoregressive system type linear predictive synthesizer, 112, 212 ... Voiced / unvoiced section switch, 113 , 213 ... D / A (digital, analog) converter, 114, 214 ... Output terminal, 115 ... Modulator, 215 ... FFT analyzer.

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification number Office reference number FI technical display location H03M 7/30 G06F 15/66 330D

Claims (21)

[Claims] 1. A narrow band signal obtained by cutting out an element signal having a pitch period or a time length longer than that from a voice signal, expanding the element signal to an integral multiple of the pitch period on the time axis, and successively connecting the narrow band signals. A method for compressing and transmitting a voice signal band, which is characterized in that it is converted into an analog waveform and transmitted. 2. The voice signal band compression transmission method according to claim 1, wherein the element signal is formed by adding a synchronization pulse to a waveform having a pitch period cut out from the voice signal or a time length longer than the pitch period. Audio signal band compression transmission method. 3. A voice signal band compression transmission method according to claim 1, wherein a high frequency component of the element signal is removed and an occupied band is limited to a low frequency band. Transmission method. 4. A voice signal band compression transmission method according to any one of claims 1, 2 and 3, wherein a system parameter by linear prediction analysis showing characteristics of the voice signal in an unvoiced section of the voice signal. A method for compressing and transmitting a voice signal band, characterized in that a base signal of a constant cycle is used to transmit an output signal which is waveform-combined. 5. The voice signal band compression transmission method according to claim 4, wherein pitch period information is transmitted using a base signal immediately before the waveform transmission in the voiced sound section. . 6. The voice signal band compression transmission method according to claim 4, wherein in the unvoiced section of the voice signal, the best fit to the prediction residual signal is made from among a plurality of preset sampled source signals. A method for compressing and transmitting a voice signal band, which comprises: searching for an optimum sampled sound source signal to be performed, synthesizing information designating the optimum sound source signal into a transmission signal, and transmitting the synthesized signal. 7. The voice signal band compression transmission method according to claim 6, wherein in the unvoiced sound section of the voice signal, the absolute value of the amplitude of the harmonic component of the transmission signal is set to a linear predictive analysis system parameter indicating the characteristics of the voice. A method of band compression transmission of an audio signal, characterized in that optimum sample designation information is transmitted in correspondence with the polarity of the. 8. The voice signal band compression transmission method according to claim 4, wherein the voice signal band compression transmission method has the same period as the period of the base signal in the silent period of the voice signal. An audio signal band compression transmission method, characterized in that a sine wave signal having 9. A signal including an element waveform obtained by expanding an element signal having a pitch period of a voice signal or a time length longer than that into an integral multiple of the pitch period on the time axis, and compressing the element waveform on the time axis. Then, the pitch waveform is reproduced, and the audio signals are reproduced by sequentially connecting them to reproduce the audio signal. 10. The audio signal reproducing method according to claim 9, wherein a linearly varying weighting is added in advance to the leading edge and the trailing edge of the pitch waveform on the transmitting side or the receiving side, and a part of the adjacent pitch waveform is overlapped. An audio signal reproducing method characterized in that the discontinuity of connection points is also eliminated. 11. The audio signal reproducing method according to claim 9 or 10, wherein linearly varying weighting is performed on the entire reproduced waveform created by successively connecting pitch waveforms for one frame, and the weighted signals are added together. Method for smoothing changes in pitch waveforms adjacent to each other. 12. Claims 9, 10 and 1.
1. The audio signal reproducing method according to any one of 1, wherein the audio signal is reproduced by linear predictive synthesis using white noise as a driving signal using system parameters obtained by analyzing a received signal in a voiceless section. And audio signal reproduction method. 13. Claims 9, 10 and 11
In the voice signal reproducing method according to any one of the above, linear prediction synthesis using a system parameter obtained by analyzing a received signal in an unvoiced section and using an optimum sample sound source signal obtained by analyzing the received signal as a drive signal. A method for reproducing an audio signal, characterized in that an audio signal is reproduced by the method. 14. The voice signal reproducing method according to claim 12, wherein a pitch period and a linear prediction analysis system parameter are obtained from a base signal immediately before receiving a waveform in a voiced sound section, and a period equal to the pitch period is repeated. An audio signal reproducing method characterized in that an audio signal is reproduced by linear predictive synthesis using the pulse train as a drive signal and using the system parameters. 15. A pitch period extracting means for extracting at least a pitch period from a voice signal to be transmitted to a transmitting side, and a means for cutting out the extracted pitch period or an element signal having a time length longer than the pitch period from the voice signal. And means for expanding the cut-out element signal to an integral multiple of the pitch period on the time axis and successively connecting the element signals to obtain a first narrowband signal, and at least the element from the narrowband signal is provided on the receiving side. An audio signal band, comprising: an element signal cutting-out means for cutting out a signal; and a means for compressing the cut-out element signal on a time axis to reproduce a pitch waveform and successively connecting them to obtain an audio signal. Compression / decompression device. 16. The audio signal band compression / expansion device according to claim 15, wherein after the element signal is compressed on the time axis to reproduce the pitch waveform on the receiving side, changes in adjacent pitch waveforms are smoothed. An audio signal band compression / decompression device comprising means. 17. The audio signal band compression / decompression device according to claim 15 or 16, wherein the transmitting side is provided with means for removing a high frequency component of the element signal and limiting an occupied band to a low frequency range. An audio signal band compression / expansion device characterized by: 18. The voice signal band compression / decompression device according to claim 15, wherein the transmitting side includes means for detecting an unvoiced sound section of the audio signal, and the unvoiced sound section. Linear prediction analysis means for extracting a system parameter from a speech signal, inverse filtering means for obtaining a prediction residual signal from the speech signal using the system parameter, and power detection means for detecting an average power level of the prediction residual signal A base signal generating means for generating a base signal having a waveform that repeats at a constant cycle; and a second narrowband signal on which the average power level information of the prediction residual signal and system parameters are placed using the base signal as a drive signal. And a means for switching the first and second narrowband signals corresponding to the voiced sound section and the unvoiced sound section of the original voice. A voiced sound / unvoiced sound discrimination means for discriminating the first and second narrow band signals is provided on the receiving side, and average power level information and system parameters of the prediction residual signal are extracted from the second narrow band signal. Extracting means, white noise generating means for generating white noise having a magnitude proportional to the average power level information of the prediction residual signal, and a linear signal for obtaining an audio signal by using the white noise as a drive signal and the system parameter. An audio signal band compression / expansion device, comprising: predictive synthesis means. 19. The voice signal band compression / expansion device according to claim 18, wherein the base signal generating means on the transmitting side generates a base signal having a power proportional to the average power of the prediction residual signal detected by the power detecting means. An audio signal band compression / expansion device, which is characterized by: 20. The voice signal band compression / decompression device according to claim 18, wherein a plurality of sampled sound source signals as white noise are set in advance on the transmission side, and among the set signals, the prediction residual signal is the most significant. Means for searching an optimum sample sound source signal that is well matched and means for synthesizing information designating the optimum sample sound source signal into the waveform of the second narrow band signal are provided, and the second narrow band signal is provided on the receiving side. A means for extracting the optimum sample sound source signal designating information from the signal, and a white noise generating means for presetting a plurality of sample sound source signals as white noise and selecting the optimum sample sound source signal based on the optimum sample sound source signal designating information are provided. An audio signal band compression / expansion device characterized by the above. 21. The voice signal band compression / decompression device according to claim 20, wherein on the transmission side, a system parameter is designated as the absolute value of the amplitude of the harmonic component of the second narrowband signal, and an optimum sample sound source is designated as the polarity of the amplitude. Waveform synthesizing means for providing corresponding information is provided, and means for extracting the system parameter and the optimum sample signal designating information is provided on the receiving side from the absolute value and polarity of the amplitude of the harmonic component of the second narrowband signal. An audio signal band compression / expansion device characterized by the above.