JPH0923195A - Sound signal band compressing/extending device, sound signal band compressing/transmitting system and sound signal reproducing system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a sound signal band compressing/extending device capable of processing a signal in an analog waveform state even in the case of using a system parameter for band compression and obtaining band-compressed transmission by an analog signal transmission system by executing A/D conversion and D/A conversion. SOLUTION: A linear predicting analyzer 103 obtains a predicting coefficient ai from a digital sound signal y (nΔt), a reverse filtering circuit 104 obtains a prediction residual signal x (nΔt) based upon the coefficient a and a pitch period extracting circuit 105 extracts a pitch period from the signal x (nΔt). A segmenting circuit 106 extracts an element signal for one pitch period or integer times the pitch period, a time base extending circuit 107 extends the time base of the extracted signal, prepares a base signal x' (nΔT) in a low frequency band and supplies the signal x' (nΔT) to an autoregressive system type linear predicting synthesizer 110, which obtains a narrow band signal w(nΔT) and transmits the signal w(nΔT) to the receiving side. The receiving side restores the received signal to a sound signal by reverse processing against the above-mentioned 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 device capable of band compression of a voice signal in the state of analog waveform,
In particular, the present invention relates to a voice signal band compression / expansion device suitable for analog transmission over a narrow band wireless transmission line.

[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 voice signals, the frequency bandwidth of voice signals, although there are individual differences, generally spans several kilohertz, and therefore transmission of this also requires a transmission system with a frequency bandwidth of several kilohertz. However, if the occupied frequency bandwidth can be compressed without impairing the clarity required for information transmission by voice, the cost required for the transmission system can be reduced.

Therefore, various voice signal band compression techniques have been conventionally proposed. As an example, a human vocalization mechanism 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 band compression of a voice signal is obtained by simulating an audio signal as a signal and extracting a system parameter by predictive analysis. The example is as follows.

IEICE Journal '85 / 5 Vol.J68-A No.
5 PP489-495 '"Residual-driven vocoder method using learning identification type spectral smoothing method (LI-RELP)" IEEE TLANSACTIONS ON COMMUNICATIONS, VOL.CON-23,
NO.12, DECEMBER1976PP1466-1474 `` The Residual-Exci
ted Linear Prediction Vokoder with Transmission Ra
teBelow 9.6 kbit / s ”(The Residual-Excited Linear Prediction Vocoder Wiz
Transmission Late Below 9.6 kilobits)

[0005]

The above-mentioned prior art does not consider the fact that the system parameters are obtained as digital numerical information, and has a problem in that it is applied to an analog signal transmission system. An object of the present invention is to perform band compression transmission by an analog signal transmission system by performing A / D conversion and D / A conversion, which can be processed in an analog waveform state, even though system parameters are used for band compression. It is an object of the present invention to provide an audio signal band compression / expansion device capable of obtaining the following. Another object of the present invention is to use an analog signal transmission system and to compress and transmit an occupied frequency band without impairing the intelligibility of a voice signal, and a source voice signal reproduced from the narrow band signal. The purpose is to provide a playback method.

[0006]

The above object is to embed spectrum information of a voice signal in a narrow band analog waveform in the form of autocorrelation. At this time, at the transmitting side, the sampling rate is reduced to transmit and receive. This is achieved by restoring the sampling rate on the side. As a result, system parameters can be transmitted in the state of analog waveforms, and as a result, the main part of the audio signal is transmitted with sufficient fidelity, and high-quality and highly efficient band compression can be obtained. Is.

More specifically, first, the main part of the audio signal is cut out in accordance with the pitch period or an integral multiple thereof, expanded on the time axis and sequentially connected to form a low frequency narrow band signal, which is as it is. An analog waveform is transmitted as a base signal. The transmission of the system parameter is then obtained by embedding it in the analog waveform in the form of autocorrelation information by supplying the basis signal to the autoregressive system using the system parameter.

[0008]

The lower limit frequency of the frequency band of the voice signal y (nΔt) to be transmitted is f _L and the upper limit frequency is f _m . Here, at _{Δt = 1 / 2f m, y} (nΔt) is a time n.DELTA.t
It is assumed that (n is an integer) represents the value of the audio signal. Now, taking as an example a case where a linear prediction coefficient is used as a system parameter, a linear prediction analysis is performed on a speech signal to obtain a linear prediction coefficient a _i (i = 0, 1, 2, ..., N).
-1) and the prediction residual signal x (nΔt). here,
x (nΔt) is the value of the prediction residual at time nΔt. From the prediction residual signal x (nΔt), a signal having a time length equal to the pitch period of the voice or an integral multiple thereof (hereinafter referred to as an element signal) is cut out and expanded to C (C: integer) times on the time axis,
The highest frequency is compressed to f _m / C and connected successively, and this is used as the base signal x ′ (nΔT). Where ΔT = C / 2f _m
It is.

Next, the basis signal x '(nΔT) is converted into a _i
Is applied to an autoregressive system having a regression coefficient to obtain an output signal w (nΔT). Since the autoregressive system is linear, this output signal w (nΔT) also does not include high frequency components above f _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 = C / 2f _m . Here, the audio signal y (n
Both Δt) and the output signal w (nΔT) have the same linear prediction coefficient a _i . However, the upper limit frequency f _m of the audio signal y (n.DELTA.t), since the upper limit frequency of the output signal w (n.DELTA.T) is fm / C, during a predictive sampling interval, a relationship of [Delta] T = ct.

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 consisting of a narrow band analog waveform.
Only by transmitting (nΔT), the original audio signal y (nΔT
The spectrum information of t) can be faithfully transmitted. 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.

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An audio signal band compression / expansion device according to the present invention will be described below in detail with reference to the illustrated embodiments. First, FIG. 1 is a block diagram showing a configuration of a transmitting side in an embodiment of an audio signal band compression / expansion apparatus according to the present invention. An audio signal y (t) to be transmitted is supplied to an input terminal 101, and first, A / D (analog / digital) converter 102
Is sampled and converted into a digital signal y (nΔt). Here, the signal y (t) is the value of the audio signal at the time t, and the signal y (nΔt) is the time n as described above.
It is the value of the audio signal at Δt (n = integer).

Here, the lower limit frequency f _L = 300 Hz and the upper limit frequency f _m = 40 of the frequency component of the original audio signal y (t).
00 Hz and the sampling time interval Δt is Δt = 1 / 2f _m = 125 μs (sampling frequency 8
KHz). Next, this digital voice signal y (nΔt)
Is regarded as an autoregressive signal, and the linear prediction coefficient a _i is used as a system parameter,

[0013]

[Equation 1]

It is defined as Here, it is considered that the first term on the right side represents the sound source signal due to vocal cord vibration or expiration in the human vocal mechanism, and the second term also represents the filtering action by the vocal tract. So, this A /
The audio signal y (nΔt) output from the D converter 102 is
The linear prediction (LP) analyzer 103 and the inverse filtering circuit 104 are supplied to the first linear prediction analyzer 103.
Then, the linear prediction coefficient a _i (i = 1, 2, 3, ..., N−
The estimated value of 1) is obtained.

On the other hand, in the inverse filtering circuit 104,
Using this linear prediction coefficient a _i , the following (Formula 2) is applied to the digital audio signal y (nΔt) consisting of a time series signal.
A prediction residual signal x (nΔt) is obtained by performing an arithmetic operation according to an equation, which constitutes a linear prediction system.

[0016]

[Equation 2]

Since the prediction residual signal x (nΔt) which is the output of the inverse filtering circuit 104 includes the pitch component of the voice, the pitch cycle extraction circuit 105 uses the integral multiple M · ΔTp (however, of the pitch cycle ΔTp). M is an integer of 1 or 2 or more), and the cutout circuit 106 cuts out an element signal of time length M · ΔTp from x (nΔt). This element signal is expanded by C times on the time axis in the time axis expansion circuit 107 to become the base signal x ′ (nΔT). Specifically, the time axis expansion circuit 107 reduces the sampling rate to 1 /
It is dropped to C. x (nΔt) → x '(nΔT) (where ΔT = CΔ
t) is converted.

FIG. 2 is a conceptual diagram showing a time axis expansion operation when M = 1. The prediction residual signal has a pitch period Δ
It has a periodic waveform that repeats at Tp. So x
If the element signal of time length M · ΔTp is cut out from the waveform of (nΔT), the sampling speed is reduced to 1 / C, the signals are arranged on the time axis, and if they are connected sequentially, the original waveform x
A waveform x ′ (nΔT obtained by analogically extending (nΔT) on the time axis.
T) is obtained. Also, when M ≠ 1, the same result is obtained. Here, if ΔT = CΔt and C = 5, the sample rate is reduced to ⅕, and the sampling time interval ΔT = 625 μs.

Next, the basis signal x '(nΔT) is supplied to a linear prediction (LP) synthesizer 110, where the linear prediction coefficient a _i (i =
1, 2, 3, ..., N-1) is used as a regression coefficient,
The base signal x ′ (nΔT) is subjected to the autoregressive system calculation according to the following (Equation 3) to obtain the narrowband time series signal w (nΔT).

[0020]

(Equation 3)

Next, in this way, the linear prediction synthesizer 1
Narrowband time series signal w (nΔT) obtained at the output of 10
Is supplied to a D / A (digital / analog) converter 111, restored to an analog waveform signal, and output terminal 112
Then, a narrow band analog signal w (t) is obtained. Therefore, looking at this narrow band analog signal w (t),
It 0 to F _m / C, i.e., has a frequency component of 0Hz～800Hz.

On the other hand, the frequency component of the original audio signal y (t) has the lower limit frequency f _L = 300 Hz and the upper limit frequency f _m = 4000 Hz, as described above. Therefore, according to this embodiment, C = 5. , The frequency range of 300 Hz to 4000 Hz is 1 / C, that is, band compression is performed in the frequency range of 0 Hz to 800 Hz. Thus, the narrow band analog signal w (t) obtained at the output terminal 112 is obtained.
Is put on a predetermined signal transmission system, such as a telephone line or a wireless channel, and transmitted to the receiving side.

Next, FIG. 3 is a block diagram showing the configuration of the receiving side in one embodiment of the audio signal band compression / expansion device according to the present invention. The narrow band analog signal w (transmitted from the transmitting side of FIG. 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).

Next, this time series digital signal w (n
ΔT) is applied to the linear prediction analyzer 203 and the inverse filtering circuit 204. First, in the linear prediction analyzer 203, linear prediction coefficients a _i (i = 1, 2,
, ..., N−1) values are restored.

On the other hand, in the inverse filtering circuit 204,
Using this linear prediction coefficient a _i , the following (Formula 4) is applied to the digital audio signal w (nΔT) consisting of a time series signal.
The reproduction basis signal x ′ (nΔT) composed of the prediction residual signal is obtained by performing the calculation according to the formula, and thus the linear prediction system is configured.

[0026]

(Equation 4)

Next, the element signal extraction circuit 205 extracts the element signal period (C · M · ΔTp) from the reproduction base signal x ′ (nΔT), and the time base compression circuit 20.
6, the time axis is compressed to 1 / C, and the driving signal x
(NΔT) is obtained. Specifically, the time-axis compression circuit 206 increases the sampling speed by C times, so that the cut-out element signals are once stored and x ′ (nΔT) → x (nΔt) (where Δt = 1 /
CΔT), and one element signal is repeatedly transmitted C times and transmitted. That is, regarding the drive signal x (nΔt), the reproduction base signal x ′ (nΔT), which is the source thereof, has the sampling rate increased by the time axis compression circuit 206 and can be returned to the signal having the original sampling frequency. There is. Therefore, the sampling time interval is 125
μs, and the frequency components thereof are f _{L to} f _m.
It has been returned to the range of (300 to 4000 Hz).

In the linear prediction synthesizer 210, the linear prediction coefficients a _i (i = 1, i = 1,
2, 3, ..., N-1) is used as a regression coefficient, the drive signal x (nΔt) is subjected to the autoregressive system operation according to the following (Equation 5), and a reproduced audio signal y composed of a time series signal is obtained. Obtain (nΔt).

[0029]

(Equation 5)

Then, the reproduced voice signal y (nΔt) obtained at the output of the linear prediction synthesizer 210 in this way is
Then, the analog audio signal y (t) is supplied to the D / A converter 211, restored to an analog waveform signal, and output to the output terminal 212.

Next, each element in the above embodiment will be described. First, the linear prediction analyzers 103 and 203 execute processing according to the algorithm shown in FIG.
The autocorrelation function of the speech signal Sn is calculated, and the coefficient a _i (i
= 1, 2, 3, ..., N-1). Although not particularly necessary for understanding the present invention, details of this linear predictive analyzer are described in, for example, “Computer Audio”, June 10, 1980 (published by Showaho Publishing Co., Ltd.). Refer to pages 43 to 50 of “Process” <Electronic Science Series>.

Next, the inverse filtering circuits 104, 20
The inverse filtering process by 4 is the coefficient a described above.
_i (i = 1, 2, 3, ..., N−1) is known, and a residual signal, for example, a signal x (nΔt) is calculated from this, according to the above equation (2). The calculation is performed. Further, the linear predictive synthesizers 110 and 210 perform arithmetic operations according to the above-mentioned (Equation 3), and have a function of synthesizing an audio signal using the residual signal by the processing shown in FIG. 5, for example. is there. Note that this linear predictive synthesizer is not particularly necessary for understanding the present invention,
For details, for example, refer to pages 50 to 53 of "Computer Audio Processing"<Electronic Science Series>, published by Koho Publishing Co., Ltd., June 10, 1980 (Showa 55).

Next, since the pitch period detection circuit 105 is one of the features of the present invention,
An example of the circuit configuration will be described in detail with reference to FIG. FIG.
Is an embodiment of a circuit that specifically realizes the pitch period extraction circuit 105. In FIG. 6, the prediction residual signal x (nΔ
In t), first, the low-pass filter 601 extracts only the low-frequency component. The cutoff frequency of the low-pass filter 601 is the upper limit of the predicted pitch frequency (about 500 Hz for a high-voiced female voice).
Is set to Therefore, the output of the low-pass filter 601 contains at least the pitch fundamental wave component. After lowering the sampling rate of this low frequency component to, for example, 1.0 KHz, it is added to the shift register 603, the input signal and each stage signal are multiplied by the multiplier group 602, and the autocorrelation function R is passed through the linear filter group 604. _n (n = 0 and 10,
11, 12, ..., 20) are obtained.

When the original signal has a periodicity of n samples, R _n / R ₀ = 1 holds. Therefore, among the R _n obtained by the above method, R _n
By obtaining R _n closest to the condition of = R ₀ , it is possible to know a desired pitch period or a time length that is an integral multiple thereof.
In FIG. 6, reference numeral 605 indicates the configuration of each filter of the linear filter group 604. This is “recurive
This is known as "autocorrelation method", and the details of its operation are described in, for example, November 3, 1981 (Showa 56).
Please refer to page 116 of "Basics of Speech Information Processing" published by Ohmsha Co., Ltd., by Sozo Saito and Kazuo Nakata.

The pitch period extraction method described above can be applied to the element signal extractor 205 on the receiving side of the present invention in exactly the same manner. The voice includes a voiced section in which a pitch exists and an unvoiced section in which a pitch does not exist. In the voiced section, the element signal can be cut out from the prediction residual signal and the waveform information can be transmitted almost completely by the above-mentioned method, but in the unvoiced section, the pitch period cannot be extracted and the cutout interval of the element signal can be reduced. Becomes indefinite. However, empirically, the prediction residual signal in the unvoiced section may be replaced with white noise, and even if the clipping interval is indefinite, the white noise property is maintained, so that no practical problem occurs.

The above-mentioned method can be applied to the element signal extraction on the receiving side, but there is a drawback that the linear prediction analysis time cannot be shortened because the received signal is originally narrowed in band. To overcome this, the transmitter side adds an appropriate marker signal corresponding to the element signal cut-out time and sends it out, and the receiver side detects this and cuts out the element signal in synchronization. . The marker signal can be realized, for example, by setting one frequency outside the transmission signal band and turning it on / off.

In the above embodiment, the audio signal y
(NΔt) is defined by the above equation (1), and the prediction coefficient a
It is stated that the prediction analysis is to obtain _i (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 It is not limited to this. In general, there are various known methods for identifying F (z ⁻¹ ) on the assumption that y (z) = x (z) / 1 + F (z ⁻¹ ) is satisfied by describing a voice signal in Z · conversion format. However, the predictive analysis in the present invention is
It includes all of them.

The linear prediction system in the present invention means all systems that obtain x (z) from y (z) by x (z) = {1 + F (z ^-1 )} y (z), Similarly, the autoregressive system means all systems that obtain y (z) from x (z) by y (z) = x (z) / 1 + F (z ⁻¹ ).

[0039]

According to the present invention, the system parameters used for analyzing and synthesizing a voice signal are embedded in a narrow band analog signal for transmission, so that in combination with the conversion of the sampling rate, It is possible to easily obtain a frequency band compression / expansion device for a voice signal that can be transmitted by a narrow band analog transmission system. Further, according to the present invention, the prediction residual signal undergoes expansion / compression on the time axis in synchronization with the pitch period, so that the waveform itself is transmitted almost completely faithfully, and is clear despite narrow band transmission. It is possible to easily obtain a high-quality voice transmission system and reproduction system without deterioration in quality. As described above, since narrow band transmission with high clarity is possible, the cost of the transmission line can be reduced and the limited resources, especially the radio frequency band can be effectively used.

By the way, in the digital transmission method, the parameter value is updated every frame period, and as a result, a voice discontinuity may occur due to the jump of the parameter value at the transition of the frame. According to this, since the analog waveform can be transmitted, the linear prediction coefficient also responds almost in real time, and therefore, there is no fear of discontinuity appearing in the voice.

[Brief description of drawings]

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

FIG. 2 is a conceptual diagram showing expansion / compression on a time axis in synchronization with a pitch cycle of a prediction residual signal according to the present invention.

FIG. 3 is a block diagram showing the configuration of the receiving side in one embodiment of the audio signal band compression / expansion device according to the present invention.

FIG. 4 is an explanatory diagram showing an example of a linear prediction analyzer according to the embodiment of the present invention.

FIG. 5 is an explanatory diagram showing an example of a linear prediction synthesizer according to the embodiment of the present invention.

FIG. 6 is a block diagram showing an example of a pitch period extraction circuit in the embodiment of the present invention.

[Explanation of symbols]

101, 201 ... Input terminals 102, 202 ... A / D (analog / digital) converters 103, 203 ... Linear predictive analyzers 104, 204 ... Inverse filtering circuits 105 ... Pitch cycle extraction circuits 106 ... Cutting circuits 107 ... Time axis Decompressor 110, 210 ... Autoregressive system type linear prediction synthesizer 111, 211 ... D / A (digital / analog) converter 112, 212 ... Output terminal 205 ... Element signal cutout circuit 206 ... Time axis compressor 601 ... Low-pass filter 602 ... Multiplier group 603 ... Shift register 604 ... Linear filter group 605 ... Linear filter

Claims (6)

[Claims] 1. A linear prediction analysis unit for extracting a system parameter from at least a voice signal to be transmitted, an inverse filtering unit for obtaining a prediction residual signal from the voice signal by using the system parameter, and the prediction on the transmitting side. Pitch period extraction means for extracting pitch synchronization or an integer multiple thereof from the residual signal, means for cutting out the extracted element signal of a time length corresponding to the pitch period or an integer multiple thereof from the prediction residual signal, The element side signal is expanded to an integral multiple on the time axis and sequentially connected to obtain a base signal, and a linear predictive synthesis means to obtain a narrow band signal from the base signal using the system parameters are provided. Includes at least linear predictive analysis means for extracting a system parameter from the narrowband signal, and a base parameter from the narrowband signal using the system parameter. An inverse filtering means for obtaining a signal, an element signal for cutting out an element signal from the base signal, a cutting out means, a means for compressing the element signal to an integer fraction on a time axis, and sequentially connecting to obtain a drive signal, And a voice signal band compression / decompression device, which is provided with a linear predictive synthesizing means for obtaining a voice signal from the drive signal using the system parameters. 2. The voice signal band compression / decompression device according to claim 1, wherein the transmitting side removes high-frequency components from the prediction residual signal to obtain low-frequency components of the prediction residual signal, An audio signal band compression / expansion device comprising: a pitch cycle extraction means for calculating an autocorrelation function of a low frequency component of the prediction residual signal to extract a pitch cycle or an integer multiple cycle thereof. 3. The audio signal band compression / decompression device according to claim 1 or 2, wherein the receiving side includes filter means for removing a high frequency component from the base signal to obtain a low frequency component of the base signal. It is characterized by further comprising an element signal cutting-out means for calculating an autocorrelation function of a low-frequency component of the base signal to extract a pitch cycle or an integer multiple cycle thereof, and cutting out an element signal having a time length equal to the cycle from the base signal. Audio signal band compression / expansion device. 4. The audio signal band compression / decompression device according to claim 1, wherein the transmitting side is provided with a marker signal adding means for indicating the element signal cutout time, and the receiving side detects the marker signal. An audio signal band compression / expansion device is provided with element signal cutout means for cutting out an element signal from the base signal. 5. A voice signal is sampled to obtain a sampled signal,
A system parameter indicating the characteristic of the speech signal is extracted from the sampled signal, a prediction residual signal is generated from the sampled signal using the extracted system parameter, and at least a required component of the prediction residual signal and the above In the audio signal transmission method for transmitting system parameter information, in the prediction residual signal, an element signal having a time length corresponding to the pitch cycle or an integral multiple thereof is cut out and expanded to an integral multiple on the time axis. A band compression transmission system for a voice signal, characterized in that the above system parameters are combined with a signal obtained by successive connection in the form of autocorrelation, and the combined signal is converted into an analog waveform for transmission. 6. A voice signal reproduction method for receiving a signal including at least a required component of a prediction residual signal of a voice signal and information of system parameters, and reproducing the voice signal from the received signal, wherein the received signal is Is an analog waveform, the signal is sampled, then the system parameter is extracted, a prediction residual signal is generated from the signal using the extracted system parameter, and the prediction residual signal is generated on the time axis. A reproduction method of an audio signal, characterized in that a reproduced audio signal is obtained by synthesizing the above-mentioned system parameters in a form of autocorrelation with a signal obtained by compression and sequentially connecting.