JPH0511452B2 - - Google Patents

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a speech encrypting device, and particularly to playback in an analog encrypting device that attempts to polarize an input audio signal by representing it with a spectral envelope parameter and a residual waveform. Concerning improving sound quality.

[Conventional technology]

Analog privacy devices that attempt to conceal the contents of a call by expressing the input audio signal as a spectral envelope parameter and a residual waveform and transmitting this instead of transmitting the input audio signal as it is from the transmitting side to the receiving side are various types. It is becoming widely used in the operational field.

This type of privacy device has high secrecy and can secure communication content that can withstand use even on relatively poor lines, so it is being used particularly in fields such as shortwave communication privacy.

In such a secret speech device, the spectral envelope parameter that indicates the macroscopic spectral distribution of the input audio signal is LPC (Linear Prediction).
The residual waveform as a microscopic spectral distribution, that is, the sound source information, is expressed by modeling the sound source using voiced/unvoiced information and pitch period, and the voiced/unvoiced information and pitch period are The normal operating format is to transmit periodic information instead of the sound source to deal with the problem of line transmission capacity.

[Problem that the invention seeks to solve]

However, the above-mentioned conventional secret communication device of this type has the following problems.

In other words, in order to perform confidential communication under limited line transmission capacity, the residual waveform is transmitted by representing the pitch period and information regarding voiced/unvoiced, so errors in pitch period extraction and voiced/unvoiced determination may occur. Deterioration of sound quality due to such factors is unavoidable, and if a method of transmitting the residual waveform is used to eliminate this problem, the required line capacity will increase significantly.

SUMMARY OF THE INVENTION An object of the present invention is to eliminate the above-mentioned drawbacks and to provide a confidential communication device that can essentially eliminate deterioration in sound quality without increasing line capacity.

[Means for solving problems]

The device of the present invention is a secret speech device that decomposes an input audio signal into a spectral envelope parameter based on LPC coefficients representing the spectral envelope and a residual waveform from which the spectral envelope is removed, and transmits the decomposed signal.
in-band frequency multiplexing means for converting into frequency domain data of frequency bands independently assigned in a preset occupied frequency band and transmitting the spectral envelope parameter and the residual waveform as an in-band frequency multiplexed signal; The residual waveform in the in-band frequency multiplexing means is represented by a plurality of impulse trains each having an amplitude proportional to its maximum value and whose time position corresponds to the time position of each local maximum value of the residual waveform. a transmitting side comprising an impulse train forming means expressed as , a demultiplexing means for demultiplexing a frequency multiplexed signal by the in-band frequency multiplexing means, and a residual waveform from the impulse train obtained by the demultiplexing means. and a residual waveform forming means for reproducing the spectral envelope parameter obtained by the demultiplexing means as a filter coefficient of an LPC synthesis filter, and a residual waveform generated by the residual waveform forming means as a filter input to synthesize an input audio signal. and a receiving side equipped with an LPC combining means.

〔Example〕

Next, the present invention will be explained in detail with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of the transmitting side of an embodiment of the privacy device of the present invention, and FIG. 2 is a block diagram showing the configuration of the receiving side of the privacy device of the invention.

The transmitting side 1 shown in FIG. 1 includes an A/D converter 1
01, window processor 102, delay circuit 103, autocorrelation coefficient calculator 104, LPC analyzer 105, LSP
Analyzer 106, linear converter 107, interpolator 10
8. Frequency generator 109, LPC inverse filter 11
0, window processor 111, maximum value searcher 112, residual waveform normalizer 113, interpolator 114, μ converter 1
15, a frequency generator 116, a μ/Linear converter 117, a residual waveform pulse generator 118, a frequency generator 119, and an adder 120.

Further, the receiving side 2 shown in FIG. 2 includes an A/D converter 201, a sliding DFT (discrete
Fourier Transform) circuit 202, delay circuit 2
03, peak picking circuit 204, interpolator 20
5. Linear inverse converter 206, μ/Linear converter 2
07, ω/α converter 208, LPF (Low Pass
Filter) 209, FM detector 210, cross-correlation calculator 211, peak detector 212, pulse former 213, multiplier 214, LPC combiner 215,
It is configured to include a D/A converter 216.

First, the transmitting side 1 will be explained. The input audio signal is sampled by the A/D converter 101 at a sampling frequency of 8 KHz, and then quantized to 12 bits, and this quantized signal is supplied to the window processor 102 and the delay circuit 103. .

The window processor 102 processes the input quantized signal.
30 mSEC, that is, 240 samples each, are temporarily stored in the internal memory as a window time, and weight multiplication by a window function such as a Hamming function is performed every 5 mSEC. Therefore, 5mSEC is the analysis frame period.

The output of the window processor 102 is the autocorrelation coefficient calculator 1
04, whereby a 12th order autocorrelation coefficient is calculated for each analysis frame and sent to the LPC analyzer 105.

The LPC analyzer 105 uses a known LPC based on the 12th order autocorrelation coefficient input for each analysis frame.
Extract the α parameter as the 12th-order LPC coefficient through analytical means, and this 12th-order α parameter is
LSP analyzer 106 with an analysis frame period of 5mSEC
and LPC inverse filter 110.

The LSP analyzer 106 generates a 12th-order LSP using a known method using the input 12th-order α parameter.
(Line Spectrum Pairs) parameters are determined and supplied to the linear converter 107.

The 12th order LSP parameter output from the LSP analyzer 106 is a frequency domain parameter, and in this embodiment, it is extracted using a quantized signal sampled at a sampling frequency of 8KHz, so the frequency is It is distributed over the frequency range of 0 to 4KHz. The linear converter 107 converts this frequency distribution from 0 to
Bandwidth compression is achieved by linearly converting 4KHz into a predetermined allocated frequency band of 2 to 3KHz.

The output of the linear converter 107 is supplied to the interpolator 108, which interpolates the linear compression LSP parameter input input at an analysis frame period of 5 mSEC at a predetermined interpolation step of 8 KHz, 125 μSEC for each order, and uses the output as a frequency generator. 109. In this way, the frequency generator 109 has a frequency of 2 to 3 KHz.
The 12th-order LSP parameters compressed into the assigned frequency band are smoothed by interpolation for each order, and the values are updated and supplied for each analysis frame.

The frequency generator 109 has a function of generating 12 frequencies corresponding to each of the 12th-order LSP parameters, generates 12 such frequencies within the assigned frequency band of 2 to 3 KHz, and adds them. 120.

The spectral envelope parameters are thus converted into 12 time-varying frequencies distributed in the assigned frequency band of 2 to 3 KHz and supplied to the adder 120. Assigned frequency band 2 to spectral envelope parameter
~3kHz is determined based on the following operational conditions.

That is, in this embodiment, the occupied frequency band of the transmission line is about 0.3 to 3.4 KHz, of which the 2 to 3 KHz band is used as the spectral envelope parameter, and the 0.4 to 1.9 KHz band is used as the residual waveform component. Assigned. The content of such allocation can be arbitrarily set in consideration of the occupied frequency band, audio reproducibility, and other operating conditions.

Now, the delay circuit 103 delays the input quantized signal for a predetermined time, and then passes the input quantized signal to the LPC inverse filter 11.
Supply to 0. The delay time in this case is that the quantized signal input to the window processor 102 is
This approximately corresponds to the processing time until the LPC inverse filter 110 outputs the 5th to 12th order α parameters while being updated every 5mSEC. parameters will be provided.

The LPC inverse filter 110 is a filter that has frequency characteristics opposite to that of the LPC synthesis filter that synthesizes input audio signals by driving with modeled residual waveforms using LPC parameters as filter coefficients, and is updated every 5 mSEC. A quantized audio signal is input using the 12th-order α parameter as the filter coefficient, and a residual waveform is output, which is supplied to the window processor 111.

The window processor 111 cuts out the input residual waveform using a rectangular function with a window time of 20 mSEC, and then passes this to the maximum value searcher 112 and the residual waveform normalizer 113.
supply to. Therefore, 20 mSEC becomes the analysis frame period for the residual waveform.

The maximum value searcher 112 searches for the maximum value of the input residual waveform for each analysis frame of 20 mSEC and sequentially supplies this to the interpolator 114.

The interpolator 114 smoothes the data by interpolating between the input maximum values in predetermined interpolation increments, for example, 8 KHz and 125 μSEC, and transfers these data to the μ converter 1.
15.

When changing the frequency of the output of the interpolator 114 and sending it out to the transmission line, it is necessary to compress the large dynamic range of several tens of dB or more in the analysis frame. This compression must be performed with the aim of optimizing the S/N ratio in the transmission path and the signal deterioration due to compression. A μ converter 115 performs nonlinear compression using μ conversion, which allows compression to ensure the necessary S/N even at a low level, and supplies this to a frequency generator 116 and a μ/Linear converter 117.

The frequency generator 116 generates a frequency that changes within the assigned frequency band of 1.6 to 1.9 KHz in response to the magnitude of the maximum value of the input residual waveform, and adds this to the adder 1.
Supply 20. In this way, 1.6 to 1.6 KHz of the frequency band 0.4 to 1.9 KHz assigned to the residual waveform
1.9KHz is provided for the maximum value of the residual waveform every 20mSEC analysis frame period.

On the other hand, the maximum value after μ conversion supplied to the μ/Linear converter 117 is thereby restored to the state before conversion and is supplied to the residual waveform normalizer 113.

The residual waveform normalizer 113 normalizes the residual waveform input from the window processor 111 by using the maximum value of the residual waveform input in this way. This normalization is
Positive residual waveform within each analysis frame of 20mSEC,
The maximum values in both negative directions are normalized with respect to the maximum value, and these normalized maximum value sequences are supplied to the residual waveform pulse generator 118 for each analysis frame. These normalized local maximum value sequences are shown as impulse trains indicated by positive and negative lines shown in FIG. The residual waveform pulse generator 118 gives a bandwidth to each of these normalized maximum value sequences to generate residual waveform pulses. For example, one of them, indicated by a dotted line, is a negative residual waveform pulse. Providing such a bandwidth is done in the form of waveform generation with each normalized local maximum value sequence as the pulse center frequency, and the generated waveform in this case simulates the residual waveform and is possible Gaussian waveforms or cosine waveforms are recommended as they occupy as little bandwidth as possible.
A squared waveform etc. is used.

From a different perspective, the above-mentioned allocation of bandwidth is intended to reduce the occupied bandwidth. The number and bandwidth of the residual waveform pulses for each analysis frame period are set while compromising the contradictory conditions of wanting to set as many as possible for the purpose of input audio signal reproduction and, on the other hand, wanting to keep the occupied frequency band as narrow as possible. be done. The residual waveform pulse thus formed is a pseudo residual waveform in which the waveform structure approximates the residual waveform and each local maximum value including the maximum value is expressed by a normalized value. and then converted to frequency domain data. In this conversion to frequency domain data, a frequency band of 0.4 to 1.5 KHz is assigned, and the conversion is performed in a conversion format that preserves the time position of each pulse included in the residual waveform pulse and the pulse size. .

The frequency generator 119 further performs FM modulation using the residual waveform pulse as a modulating wave with (1.5-0.4)/2 KHz as a carrier wave, and sends the modulated wave to the adder 120.

In this way, the residual waveform consists of a residual waveform pulse due to the normalized maximum value in the frequency band of 0.4 to 1.5KHz, and a residual waveform pulse due to the normalized maximum value in the frequency band of 0.4 to 1.5KHz.
It will be expressed by the maximum value of the residual waveform in the 1.9KHz frequency band. Generally speaking, handling it as frequency domain data in this way
Processing on the receiving side is also often convenient.

The adder 120 adds the input frequency-domain spectral envelope parameters and data regarding the residual waveform in a predetermined manner, and supplies the sum as a transmission signal to the receiving side 2 via the transmission line 1201.

On the receiving side 2, the transmission signal input to the A/D converter 201 is sampled at a sampling frequency of 8KHz, quantized with a predetermined number of bits, and further
30mSEC Hamming windowing is performed every 5mSEC, and data for each analysis frame period of 5mSEC is supplied to the sliding DFT circuit 202 and the delay circuit 203.

The sliding DFT circuit 202 inputs
The shift rate that presets the processing for each analysis frame of 125μSEC, in this example, 8KHz,
Sliding DFT that processes data in as small processing increments as possible while shifting in steps of 125μSEC
The input data is discretely Fourier transformed while performing
That is, 5 mSEC worth of data regarding frequency components of 1.6 to 3 KHz is outputted and supplied to the peak picking circuit 202.

The peak picking circuit 204 performs peak picking (peak picking) to detect the maximum value of input data.
The maximum value of the residual waveform that exists as a maximum value included in the 1.6-1.9KHz band and the 12th-order LSP parameter that exists as 12 maximum values included in the 2-3KHz band are detected. The respective corresponding frequencies are determined and these are fed to the interpolator 205 in turn for each analysis frame.

The interpolator 205 interpolates the maximum value of the input residual signal and the spectral envelope parameter at predetermined analysis increments of 8 KHz and 125 μSEC between analysis frames and supplies them to the linear inverse transformer 206.

In addition, in this example, the amount of calculation is reduced, 125μSEC
The output of the peak picking circuit 204 is output every 5 mSEC in consideration of fluctuations in the value, etc., but when outputting this every 125 μSEC, the interpolator 205 becomes unnecessary.

The linear inverse transformer 206 performs linear inverse transform of the 12th-order LSP parameter as the input spectrum envelope parameter. This linear inverse transformation is 0
~4KHz has been compressed to 2~3KHz and is then inversely converted back to 0~4KHz and restored. Furthermore, the maximum value of the residual waveform is also compressed by μ transformation, but this is μ/
The signal is supplied to a linear converter 207 and subjected to μ conversion, and then supplied to a multiplier 214.

The 12th-order LSP parameters subjected to the linear inverse transformation are supplied to the ω/α converter 208, which converts the LSP parameters (ω) into α parameters and supplies them to the LPC synthesizer 215.

The LPC synthesizer 215 receives from the ω/α converter 208 the 12th-order α parameter updated every 125μSEC as a filter coefficient of an all-pole digital filter, and also receives the residual waveform from the adder 214 and is driven. However, this residual waveform is reproduced as follows.

That is, the output of the A/D converter 201 is also supplied to the delay circuit 203;
shall be given a delay time so that the maximum value of the residual waveform output from the μ/Linear converter 207 and the residual waveform pulse train output from the pulse former 213 have the same frame period, and this output is supplied to the LPF 209.

The LPF209 extracts only the frequency band of 0.4 to 1.5KHz, that is, the FM modulated residual pulse.
It is supplied to the FM detector 210.

The FM detector 210 receives the input signal in this way.
FM detection is performed to detect a residual waveform pulse shown in FIG.

The cross-correlation calculator 211 stores in advance a 1-level normalized value of the residual waveform pulse shown in FIG. After confirming that it is a waveform pulse, it is supplied to the peak detector 212.

The peak detector 212 thus detects the peak value of the input output of the cross-correlation calculator 211 and supplies it to the pulse former 213 .

The pulse former 212 generates time domain data of an impulse train of a level normalized to the maximum value at the time position indicated by the vertical line of the residual waveform shown in FIG. 1, and supplies this to the multiplier 214.

The multiplier 214 multiplies the normalized level of the input impulse train by the maximum level, generates a residual waveform expressed by the actual level impulse train for each analysis frame, and supplies this to the LPC synthesizer 215.

The LPC synthesizer 215 uses the input 12th-order α parameter as a filter coefficient, is driven by the residual waveform, regenerates digital audio, and converts it into D/
A converter 216 is supplied.

The D/A converter 216 converts the digital input into analog, removes unnecessary high-frequency components using an LPF, and sends it out as an output audio signal.

In this way, confidential communication based on transmission of residual waveforms can be performed without increasing line capacity.

In this example, the residual waveform is expressed by a plurality of impulse trains that have amplitudes proportional to the maximum value of the residual waveform for each analysis frame and are located at positions that match the time positions of the residual waveforms. It is clear that other similar representations, such as multi-pulses formed by substantially the same impulse train with limited pulse intervals, can be implemented in substantially the same way.

Further, the α parameter, the analysis order of the LSP parameter, etc. may be set arbitrarily.

〔Effect of the invention〕

As explained above, according to the present invention, in a confidential communication device that decomposes an input audio signal into spectral envelope parameters and residual waveforms and transmits the signals, the input audio signals are converted into frequency domain data of frequency bands assigned independently from each other. The spectral envelope parameter and the residual waveform are transmitted, and the residual waveform has an amplitude proportional to its maximum value, and a plurality of impulse trains each having a time position corresponding to a time position of each local maximum value of the residual waveform. By carrying out secret communication using a means of representing the number by , it is possible to realize a secret communication device that can significantly improve the deterioration of sound quality without increasing the line capacity.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of an embodiment of the transmitting side of the secure communication device of the present invention, and FIG. 2 is a block diagram showing the configuration of an embodiment of the receiving side of the secure communication device of the present invention. 1...Sending side, 2...Receiving side, 101...A/
D converter, 102... Window processor, 103...
Delay circuit, 104...Autocorrelation coefficient calculator, 10
5...LPC analyzer, 106...LPC analyzer, 1
07...Linear converter, 108...Interpolator, 109
...Frequency generator, 110...LPC inverse filter,
111... Window processing device, 112... Maximum value search device,
113...Residual waveform normalizer, 114...Interpolator, 115...μ converter, 116...Frequency generator, 117...μ/Linear converter, 118...
Residual waveform pulse generator, 119... Frequency generator, 120... Adder, 201... D/A converter, 202... Sliding DFT circuit, 2
03...Delay circuit, 204...Peak picking circuit, 205...Interpolator, 206...Linear inverse transformer, 207...μ/Linear converter, 208...
ω/α converter, 209……LPF, 210……
FM detector, 211... Cross correlation calculator, 212
... Peak detector, 213 ... Pulse former, 2
14... Multiplier, 215... LPC synthesizer, 21
6...D/A converter.

Claims (1)

[Claims] 1 In a confidential communication device that decomposes and transmits an input audio signal into a spectral envelope parameter based on LPC coefficients expressing its spectral envelope and a residual waveform from which the spectral envelope has been removed, allocation is made independently of each other in a preset occupied frequency band. an in-band frequency multiplexing means for converting the spectral envelope parameter and the residual waveform into frequency domain data of a frequency band in which the spectral envelope parameter and the residual waveform are transmitted as an in-band frequency multiplexing signal; impulse train forming means for representing the impulse train as represented by a plurality of impulse trains each having an amplitude proportional to the value and whose time position corresponds to the time position of each local maximum value of the residual waveform. a demultiplexing means for demultiplexing the frequency multiplexed signal by the in-band frequency multiplexing means; a residual waveforming means for reproducing a residual waveform from the impulse train obtained by the demultiplexing means; The spectral envelope parameter obtained by the separating means is used as a filter coefficient of an LPC synthesis filter, and the residual waveform produced by the residual waveforming means is used as a filter input to synthesize an input audio signal.
1. A private conversation device comprising: a receiving side equipped with LPC synthesis means;