JPH0946233A - Sound encoding method/device and sound decoding method/ device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To transmit a wide band sound signal by transmitting a spectrum envelope information parameter and a sound source information parameter to a first frequency band signal and transmitting only the spectrum envelope information parameter to a second frequency band signal. SOLUTION: A transmission orthogonal mirror filter 60 divides an input sound signal a4 into a low band component c4 and a high band component b4. The component c4 is encoding-processed by a highly efficient sound encoder 62 and a sound information parameter e4 is outputted. On the other hand, the component b4 is processed by a linear analyzer 61 and a linear prediction coefficient d4 for representing spectrum envelope is outputted. The coefficient d4 from the analyzer 61 and the parameter e4 from the encoder 62 are encoded and are made into frames by an encoder 63. Then, a data string f4 being the result is transmitted to a transmission line. Thus, the wide band sound signal can be encoded and transmitted as well in mobile communication taken place by using a radio line limited in information transmission speed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice encoding / decoding method for encoding / decoding a voice signal in a telephone, and more particularly to a voice encoding / decoding method suitable for encoding a wideband voice signal, and further to this. The present invention relates to a voice encoding / decoding device using the.

[0002]

2. Description of the Related Art Generally, in a conventional telephone device, communication voice is transmitted in a normal telephone band (300 to 3400 Hz).
However, it is expected that the intelligibility will be improved in the recent telephone conversations under high background noise.

In order to improve the intelligibility in a call under such high background noise, the band of the voice signal to be transmitted is wide band (50 to 50 Hz) from the telephone band signal (300 to 3400 Hz).
Up to 7000 Hz) is effective. In addition, as a method of encoding such a wideband speech signal, for example, TTC
So-called SB-ADP standardized as standard JT-G722
CM (Sub Band‐Adaptive Differencial Pulse Code Mod
There is something called ulation).

[0004]

The above-mentioned prior art SB
The encoding speed by one ADPCM is usually 48 to 64 kbp
However, when applying this to mobile radio communication, there were the following problems. That is, in mobile radio communication, in order to effectively use the frequency, it is necessary to make the frequency bandwidth occupied by one channel as narrow as possible, which limits the coding rate of voice information. As an example, the voice coding rate (including error correction) in the Japanese car telephone system is 11.
2 kbps and half rate 5.6 kbps are used. However, the above-mentioned SB-ADPCM has a voice coding rate of 48% as compared with the voice coding rate of mobile radio communication.
Since it is as high as ˜64 kbps, it is not appropriate to use it as it is in a mobile radio communication system.

Therefore, in view of the above-mentioned prior art, the present invention makes it possible to code and transmit a wideband voice signal even in mobile communication performed using a wireless line having a limited information transmission rate. The present invention relates to a voice coding system and a device therefor, that is, a wideband voice signal coding process capable of improving the intelligibility of a call under high background noise is realized at a low speed applicable to mobile radio communication. It is an object of the present invention to provide a speech encoding / decoding method capable of performing the above, and further to provide a speech encoding device and a speech decoding device using such a method.

[0006]

According to the present invention, after dividing an input voice signal to be transmitted into a first frequency band signal and a second frequency band signal, a high frequency signal is added to the first frequency band signal. Performing an efficient speech coding process, transmitting the spectrum envelope information parameter and the sound source information parameter, which are the outputs, and extracting and transmitting only the spectrum envelope information parameter for the second frequency band signal. A featured speech coding method is disclosed.

Further, the present invention is a method for decoding a voice signal transmitted by being encoded by the voice encoding method described above, wherein the spectral envelope information parameter and the sound source information parameter of the transmitted voice signal are For the first frequency band signal, the first frequency band signal is decoded by a high-efficiency speech decoding process from the spectrum envelope information parameter and the sound source information parameter corresponding to the signal, and for the second frequency band signal, Decoding is performed using the spectral envelope information parameter corresponding to the signal and the pseudo sound source parameter estimated from the sound source information parameter of the first frequency band signal, and further, the two voice signals generated by these decoding are added. Disclosed is a speech decoding method characterized by synthesizing reproduced speech signals.

Furthermore, the present invention provides a first audio signal to be transmitted.
Band divider for dividing the frequency band signal into the second frequency band signal and the second frequency band signal, and the high efficiency speech coding processing of the first frequency band signal divided by the band divider, and the resulting spectrum envelope information. High-efficiency speech encoder that outputs parameters and excitation information parameters, and linear prediction that performs linear prediction analysis on the second frequency band signal divided by the band divider, and outputs only the spectrum envelope information parameter that is the result. An analyzer, a spectrum envelope information parameter and a sound source information parameter of the first frequency band signal output from the high-efficiency speech coder, and a second frequency band signal output from the linear prediction analyzer. Speech coding, characterized in that it has a coder for coding a spectrum envelope information parameter, and multiplexing these and sending them to a transmission path. To disclose the location.

Further, the present invention is a device for decoding a voice signal encoded and transmitted by the voice encoding device described above, wherein the multiplexed signal sent to a transmission line is received, and A separation circuit for separating and extracting a spectrum envelope information parameter and a sound source information parameter of a first frequency band signal and a spectrum envelope information parameter of the second frequency band signal, and the first circuit from the separation circuit.
A high-efficiency speech decoder for reproducing the first frequency band signal from the spectrum envelope information parameter and the sound source information parameter of the frequency band signal by a high-efficiency speech decoding process,
A parameter obtained based on the sound source information parameter of the first frequency band signal together with the spectrum envelope information parameter of the second frequency band signal from the separation circuit is used as the sound source information parameter for the second frequency band signal. A frequency component restoration circuit for reproducing the second frequency band signal by high-efficiency audio decoding processing and a band synthesis for combining the reproduced voice signal by adding the reproduced first and second frequency band signals. Disclosed is a speech decoding device characterized by comprising:

[0010]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the accompanying drawings. First, FIG.
1 shows an overall configuration of an apparatus that implements mobile communication (telephone) with a mobile body by encoding a wideband voice signal by the voice encoding method of the present invention and using a wired communication or a wireless line. Has been done. Note that this device is basically composed of a wideband speech coding device 100 and a wideband speech decoding device 200, as indicated by the alternate long and short dash line in the figure.

First, the wideband speech coding apparatus 100 shown in FIG.
Is a transmission quadrature mirror filter that constitutes the band divider 60, a linear prediction analyzer 61, and a high performance speech encoder 62.
And an encoder 63. Furthermore, in the wideband speech coding apparatus 100 having such a configuration, the signal is input by the microphone of the handset, and then 7 kHz.
Wideband input audio signal a sampled at 16 kHz and quantized with 16 bits after being band-limited by
4 is a low-frequency component c due to the transmission quadrature mirror filter 60.
4 (0 to 4 kHz) and the high frequency component b4 (4 to 7 kHz). Thereafter, the divided low frequency component c4 of the input voice signal is encoded by the general high efficiency voice encoder 62 in a frame unit of, for example, 40 ms length, whereby the voice information parameter e4 is output. To be done. The configuration example of the high-efficiency speech encoder 62 will be described later in detail.

On the other hand, the high-frequency component b4 of the divided input speech signal is processed by the linear prediction analyzer 61 in a frame unit of, for example, 40 ms, and the linear prediction coefficient d4 for expressing the spectrum envelope is output. It should be noted that in the speech coding processing for ordinary telephone band speech, the analysis order is 8 to 1
A linear prediction analyzer 6 having a degree of 0 is used.
When a high frequency component is encoded with 1, the resolution is low due to human auditory characteristics, so it is sufficient to transmit information to the extent that the outline of the spectral envelope can be roughly understood. Therefore, the division order here Is about the second order.

The high-frequency component linear prediction coefficient d4 from the linear prediction analyzer 61 and the low-frequency component speech information parameter e4 from the high-efficiency speech encoder 62 are further encoded and encoded by the encoder 63. The data is framed, and the resulting data string f4 is sent to the transmission path.

Next, details of a specific configuration of the high efficiency speech encoder 62 will be described. FIG. 2 shows an example of a specific configuration of the high efficiency speech encoder 62.
In this figure, the low-frequency component c4 of the input speech signal is first processed by the linear prediction splitter 11 in 40 ms frame units and in the analysis order 10, and as a result, the linear prediction coefficient b1 which is the spectral envelope information. Is output. Then, the spectrum envelope parameter quantizer 12
Quantizes the output linear prediction coefficient b1, and
The quantized linear prediction coefficient c1 is output. The spectral envelope parameter dequantizer 13 dequantizes the input quantized linear prediction coefficient c1 to reproduce the linear prediction coefficient d1. Further, in this high-efficiency speech encoder 62, the frame power calculator / normalizer 26 calculates the frame power of the input speech signal from the low-frequency component c4 of the input speech signal, thereby normalizing the input speech signal, Accordingly, the frame power information x1 and the normalized input audio signal w1 are output.

Subsequently, the local decoder 25, as is clear from the figure, has an adaptive codebook 14, a noise codebook 15, a gain codebook 16, multipliers 17 and 18, an adder 19, and a linear prediction filter. Finally, the reproduction signal n1 is finally combined and output in units of subframes having a length of 10 ms, for example. This local decoder 2
The internal configuration of 5 will be described. First, the adaptive codebook 1
4 is a waveform pattern having periodicity of the sound source signal, for example, 256 types (vector number = 256, vector dimension =
80) are accumulated, and similarly, the noise codebook 15 accumulates, for example, 512 types of waveform patterns having noise characteristics of the excitation signal (vector number: 512, vector dimension: 80). Further, the gain codebook 16 sets, for example, 128 types of gain sets by which the adaptive codebook 14 and the noise codebook 15 are multiplied (the number of vectors: 128, the dimension of vector:
2) It is accumulating. In addition, the high efficiency speech encoder 62
Further includes a hearing weighting filter 21 and a distance calculator 2.
2, a codebook search controller 23, and an encoder 24.

In such a configuration, the above adaptive codebook 1
4 and the random codebook 15, according to the control signals h1 and i1 output from the codebook search controller 23, respectively, a shape vector e1 indicating the shape of the signal waveform, an index q1 indicating the number of the vector, and a signal waveform Shape vector f1 representing the shape and index r representing the vector number
Outputs 1. Here, q1 corresponds to the pitch period.
Further, the gain codebook 16 has the gain g1 for the multiplier 17 of the adaptive codebook 14 and the multiplier 18 of the noise codebook 15 according to the control signal j1 output from the codebook search controller 23, as described above. The gain u1 and the index s1 are output. As a result, the multipliers 17 and 18 multiply the shape vectors e1 and f1 by the gains g1 and u1, respectively, so that the gain-adjusted shape vectors k1 and l1 are obtained.
Is output. Further, the adder 19 adds the shape vector k1 and the shape vector l1 and outputs a sound source signal m1.

The linear prediction synthesis filter 20 having the excitation signal m1 as an input, the linear prediction coefficient d1 dequantized by the spectrum envelope parameter dequantizer 13 is used.
, The spectral envelope characteristic is added to the sound source signal, and the reproduced signal n1 is synthesized and output. Further, the perceptual weighting filter 21 uses the linear prediction coefficient b1 from the linear prediction divider 11 and the input signal w1 normalized by the frame power calculator / normalizer 26 and the local decoder 2
5, the reproduced signal n1 which is the output of FIG. 5 is perceptually weighted, thereby outputting the perceptually weighted input signal o1 and the local decoder output v1. The distance calculator 22 calculates these distances p1 from the perceptually weighted input signal o1 and the local decoder output v1. The codebook search controller 23 uses the calculated distance p1 as a human power, and outputs the vector that minimizes the distance p so as to output the local decoder 25.
The adaptive codebook 14, the noise codebook 15, and the gain codebook 16 are controlled.

Quantized linear prediction, which is a transmission parameter output from the spectrum envelope parameter quantizer 12, adaptive codebook 14, noise codebook 15, gain codebook 16, and frame power calculator / normalizer 26. Coefficient c1,
Indexes q1, r1, s1 and frame power information x
1 is encoded and framed by the encoder 24,
As a result, the transmission data string t1 is obtained. The transmission data string t1 output from the encoder 24 corresponds to the voice information parameter e4 in FIG.

Next, the wideband speech decoding apparatus 200 of the present invention.
Will be described. The schematic configuration of the wideband speech decoding apparatus 200 is shown in FIG. 1, and the detailed configuration thereof is shown in FIG. 3 attached hereto. As is clear from these figures, this wideband speech decoding device 200
Is composed of a separation circuit 70, a high-efficiency speech decoder 71, a high-frequency component restoration circuit 72, and a transmission quadrature mirror filter which constitutes a band synthesizer 82.

Here, this wideband speech decoding apparatus 200 is used.
Will be described in detail with reference to FIG. The high frequency component restoration circuit 72 includes a gain correction circuit 73, an adaptive codebook 75, a gain correction circuit 77, a noise generator 78, two multipliers 76 and 79, an adder 80, and a linear prediction synthesis filter 81. , Gain correction circuit 83, and multiplier 8
And 4. The separation circuit 70 outputs six types of signals, b5, c5, d5, e5, f5, and r5,
In addition, signals g5, s5, t from the high-efficiency speech decoder 71
5 is output.

In the wideband speech decoding apparatus 200 having such a configuration, first, the data string a5 transmitted by the wired or wireless line is separated and restored by the separation circuit 70, so that the linear prediction coefficient b5 of the high frequency component and low The linear prediction coefficient d5 of the band component, the index c5 corresponding to the adaptive codebook which is the excitation information, the index e5 corresponding to the noise codebook, the index f5 corresponding to the gain codebook, and the frame power information r5 are reproduced. The signals are, in this order, the signal d4 of FIG. 1 and the signals c1, q1, r1, s of FIG.
1 and x1, which are decoded signals, of which the signals d5, c5, e5, f5, and r5 are input, and the high-efficiency audio decoder 71 reproduces the low frequency component and outputs this as the signal g5. To do. Note that this high-efficiency speech decoder 71
Is generally used for the decoding processing of telephone band voice, and its detailed configuration is shown in FIG. 4 attached.

That is, the high efficiency speech decoder 7 of FIG.
1, the adaptive codebook 31, the noise codebook 32, and the gain codebook 33 store data of similar contents to those shown by reference numerals 14, 15, and 16 in FIG. 2, respectively. , And inputs the indexes c5, e5, and f5, and outputs the shape vectors f2 and g2 and the gains j2 and k2. Further, the multipliers 34 and 35 respectively multiply the above-described shape vector and the gain, that is, the shape vector f2.
And gain j2 (f2 × j2), shape vector g2 and gain k
2 (g2 × k2) is multiplied, and the gain-adjusted shape vectors h2 and i2 are output. On the other hand, the adder 39
Outputs the sound source signal 12 by adding the above shape vectors h2 and i2. The processing up to this point has been performed in units of subframes (10 ms) in the high-efficiency speech decoder 71, and thereafter, the processing up to the post filter 38 is performed in frames (40 m).
s) is executed in units.

The linear predictive synthesis filter 36 uses the quantized linear predictive coefficient d5 to determine the spectrum envelope characteristic of the sound source signal l.
2 and the reproduced signal m2 is synthesized. After that, the multiplier 37 multiplies the reproduction signal m2 by the frame power information r5, and outputs the reproduction signal o2 to which the frame power information is added. And the post filter 3
8 uses the quantized linear prediction coefficient d5 to reproduce signal o2
The post-filtering process for improving the perceptual quality is performed to output the low-frequency component g5 which is the reproduction output of the high-efficiency audio decoder 71. In addition, the gain j2 output from the gain codebook 33 of the high efficiency speech decoder 71,
k2 is the high frequency component restoration circuit 7 in FIG.
2 are input as signals s5 and t5.

Returning to FIG. 3 again, the wideband speech decoding device 2
The high frequency component restoring circuit 72 that composes 00 is a high-frequency component linear prediction coefficient b5 from the separating circuit 70, an adaptive code length index c5 of the low-frequency component, and the high-efficiency speech encoder 71. The generated gain codebook output s5,
At t5, further, the frame power information r5 is used to reproduce the high frequency component and output it as the signal x5. In other words, in other words, by using the spectral envelope information parameter corresponding to the high frequency band signal and the pseudo sound source parameter estimated from the sound source information parameter of the low frequency band signal, the high frequency band signal is reproduced. .

The structure of the high frequency component restoring circuit 72 having such a function will be described below. First, the adaptive codebook 75 receives the index c5 from the separating circuit 70 and receives the waveform shape vector having periodicity. Output j5. The gain correction circuit 73 multiplies the gain s5 of the adaptive codebook from the high efficiency speech encoder 71 by a coefficient and outputs a corrected coefficient h5. It should be noted that the coefficient used here is set to a value smaller than 1.0 in consideration of the low periodicity in the high range as compared with the low range. Then, the multiplier 76 multiplies the shape vector j5 by the corrected gain h5, and outputs the gain-adjusted shape vector k5.

On the other hand, the gain correction circuit 77 outputs a corrected gain l5 obtained by multiplying the gain t5 of the noise codebook from the high efficiency speech encoder 71 by a coefficient. Note that the coefficient used here is also 1.0 so that the high frequencies are not overemphasized.
Set to a smaller value. Then, the multiplier 79 multiplies the white noise m5 output from the noise generator 78 by the above-described corrected gain l5, and outputs the gain-adjusted shape vector n5. After that, the adder 80 adds the outputs k5 and n5 from the two multipliers 76 and 79, and outputs the sound source signal o5. So far,
This is processing in the sub-frame (10 ms) unit in the high frequency component restoring circuit 72, and the subsequent processing is executed in frame (40 ms) units. That is, the linear prediction synthesis filter 81 adds the spectrum envelope characteristic to the sound source signal o5 by using the quantized linear prediction coefficient b5, and thus synthesizes the reproduction signal v5.

Further, the high frequency component restoration circuit 72
Then, the gain correction circuit 83 outputs the corrected frame power information w5 by multiplying the frame power information r5 from the separation circuit 70 by a coefficient. The coefficient used here is set to a value smaller than 1.0 so that the high frequencies are not emphasized too much. Then, the multiplier 84 uses the reproduction signal v5 and the corrected frame power information w5.
And the reproduction signal x5 to which the frame power information is added is output. After that, as shown in FIG. 1, by the reception quadrature mirror filter that constitutes the band synthesizer 82, the reproduced low frequency component (0 to 4 kHz) g5 and high frequency component (4 to 7).
(kHz) component x5, and the reproduced wideband audio signal q5 is output.

Wideband speech coding apparatus 10 described in detail above.
0 and the speech information coding rate in the apparatus including the wideband speech decoding apparatus 200 are as shown in Table 1 below.

[Table 1] As is clear from Table 1, the audio information coding rate is 137 × 25 frames = 3.425 kbps. Further, if the overall coding rate including the transmission path coding is 5.6 kbps, the remaining 2.175 kbps can be used for countermeasures such as error correction.

In the apparatus according to the present invention described in detail above, since the high frequency component is reproduced approximately (pseudo) on the receiving side, some distortion occurs in the reproduced high frequency component. . This distortion becomes more noticeable when the background noise on the receiving side is small. Therefore, in an environment where the background noise is small, it can be said that it is desirable to output only the low-frequency component signal as a reproduction output rather than the wideband signal. Therefore, it is proposed to add a wideband reproduction signal / low range reproduction signal output switch 90 as shown in FIG. 5 to the wideband speech decoding apparatus (FIG. 3) of the present invention. In FIG. 5, the output switcher 90 receives the wideband reproduction signal q5 and the low-frequency reproduction signal g5 shown in FIG. 3, and outputs either one of the above two signals q5 and g5 by the control signal a6. It has such a switching function that its output is the reproduction signal b6.

Further, in the wide band reproduction signal / low range reproduction signal output switch 90, as a method of generating the control signal a6 which is information as to which signal is to be selected, for example, a manual type and an automatic type are considered. To be The former is generated, for example, by a listener manually operating a switch on the operation panel of the receiving device, and the latter is automatically generated based on the measurement result of the background noise power.

Further, in the above embodiment, of the divided low-frequency band and high-frequency band of the wideband speech input signal, particularly the high-frequency band, on the transmitting side, the spectrum thereof is determined by the linear prediction analyzer 61. It is assumed that only the envelope information parameter is extracted and transmitted, and on the receiving side, in the high frequency component restoration circuit 72, the sound source parameter corresponding to the high frequency band signal is pseudo-estimated from the sound source information parameter of the low frequency band signal. , Is used to reproduce high frequency band signals. However, the present invention is not limited to such an embodiment,
On the contrary to the above embodiment, on the transmission side, the linear prediction analyzer is used to extract and transmit only the spectrum envelope information parameter of the low frequency band signal, and on the reception side, in the frequency component restoration circuit, the low frequency It is also possible to artificially estimate the sound source parameter corresponding to the band signal from the sound source information parameter of the high frequency band signal, and use this to reproduce the low frequency band signal.

[0032]

As is clear from the above detailed description, according to the audio encoding / decoding method and the audio encoding / decoding apparatus using the same of the present invention, the encoding of the wideband audio signal is performed. The processing and decoding processing can be easily realized at a low speed applicable to mobile wireless communication or wired communication, which enables broadband audio signals to be transmitted even in mobile communication performed using a wireless line having a limited information transmission speed. This makes it possible to code and transmit, and thus it is possible to improve the intelligibility of a call under high background noise, which is an extremely technically excellent effect.

[Brief description of drawings]

FIG. 1 is a block diagram of a wideband speech encoding / decoding device for implementing a speech encoding / decoding method of the present invention.

FIG. 2 is a block diagram showing a detailed structure of a high-efficiency speech encoder of the wideband speech encoding / decoding apparatus shown in FIG.

3 is a block diagram showing a structure of a wideband speech decoding apparatus of the apparatus shown in FIG.

FIG. 4 is a block diagram showing details of a high-efficiency speech decoder of the wideband speech decoding device shown in FIG.

FIG. 5 is a block diagram showing a specific configuration of a wideband / lowband reproduction signal output switch according to a modified example of the present invention.

[Explanation of symbols]

 60 band divider 61 linear prediction analyzer 62 high-performance speech encoder 63 encoder 71 high-efficiency speech decoder 72 high-frequency component restoration circuit 82 band synthesizer 90 output switcher 100 wide-band speech encoder 200 wide-band speech decoder

Claims (14)

[Claims] 1. An input speech signal is divided into a first frequency band signal and a second frequency band signal, and then the first frequency band signal is subjected to high-efficiency speech coding processing and its output. The speech coding method is characterized in that the spectrum envelope information parameter and the sound source information parameter are transmitted, and for the second frequency band signal, only the spectrum envelope information parameter is extracted and transmitted. 2. The speech coding method according to claim 1, wherein the first frequency band signal is a low frequency band signal and the second frequency band signal is a high frequency band signal. Speech coding method. 3. The speech coding method according to claim 1, wherein the first frequency band signal is a high frequency band signal and the second frequency band signal is a low frequency band signal. Speech coding method. 4. A method for decoding a voice signal transmitted by being encoded by the voice encoding method according to claim 1, comprising a spectrum envelope information parameter and a sound source information parameter of the transmitted voice signal. Received the first
For the frequency band signal, the decoding is performed by the high-efficiency speech decoding process from the spectrum envelope information parameter and the sound source information parameter corresponding to the signal, and for the second frequency band signal, the spectrum envelope corresponding to the signal. Decoding is performed using the information parameter and the pseudo sound source parameter estimated from the sound source information parameter of the first frequency band signal, and further, the reproduced sound signal is obtained by adding the two sound signals generated by these decoding. A speech decoding method characterized by synthesizing. 5. The audio decoding method according to claim 4, wherein the synthesized reproduced audio signal and the audio signal generated by decoding the first frequency band signal can be switched and output. A voice decoding method characterized by. 6. The speech decoding method according to claim 4 or 5, wherein the first frequency band signal is a low frequency band signal, and the second frequency band signal is a high frequency band signal. Characteristic speech decoding method. 7. The speech decoding method according to claim 4 or 5, wherein the first frequency band signal is a high frequency band signal and the second frequency band signal is a low frequency band signal. Characteristic speech decoding method. 8. An audio signal is a first frequency band signal and a second frequency band signal.
And a first frequency band signal divided by the band divider for high efficiency speech coding, and outputs the resulting spectrum envelope information parameter and excitation information parameter. A high-efficiency speech coder, a linear prediction analyzer that performs linear prediction analysis on the second frequency band signal divided by the band divider, and outputs a spectrum envelope information parameter as a result, the high-efficiency speech Encoding the spectrum envelope information parameter and the excitation information parameter of the first frequency band signal output from the encoder, and the spectrum envelope information parameter of the second frequency band signal output from the linear prediction analyzer And a coder that multiplexes these and sends them to a transmission path. 9. The speech coding apparatus according to claim 8, wherein the first frequency band signal is a low frequency band signal, and the second frequency band signal is a high frequency band signal. Speech coding device. 10. The speech coding apparatus according to claim 8, wherein the first frequency band signal is a high frequency band signal, and the second frequency band signal is a low frequency band signal. Speech coding device. 11. A device for decoding a voice signal encoded and transmitted by the voice encoding device according to claim 8, wherein the device receives the multiplexed and transmitted signal on a transmission line, A separation circuit that separates and extracts a spectrum envelope information parameter and a sound source information parameter of the first frequency band signal and a spectrum envelope information parameter of the second frequency band signal; and the first frequency from the separation circuit. A high-efficiency speech decoder that reproduces the first frequency band signal by a high-efficiency speech decoding process from a spectrum envelope information parameter and a sound source information parameter of the band signal, and a spectrum envelope of the second frequency band signal from the separation circuit. Together with the information parameter, a parameter obtained based on the sound source information parameter of the first frequency band signal is set to the second frequency. By using as a sound source information parameter for a band signal, a frequency component restoration circuit that reproduces the second frequency band signal by high-efficiency audio decoding processing, and by adding the reproduced first and second frequency band signals. A voice decoding device comprising: a band synthesizer for synthesizing a reproduced voice signal. 12. The audio decoding device according to claim 11, wherein the output switching for switching and outputting the synthesized reproduced audio signal and the audio signal generated by decoding the first frequency band signal. A speech decoding apparatus characterized by comprising a receiver. 13. The speech decoding device according to claim 11, wherein the first frequency band signal is a low frequency band signal and the second frequency band signal is a high frequency band signal. A characteristic voice decoding device. 14. The speech decoding apparatus according to claim 11 or 12, wherein the first frequency band signal is a high frequency band signal and the second frequency band signal is a low frequency band signal. A characteristic voice decoding device.