JPH08328598A - Sound coding/decoding device - Google Patents

Abstract

PURPOSE: To provide a sound coding/decoding device capable of correctly coding/decoding a higher quality sound compared with usual one by awaiting plural code books based on the sound data obtained under various noise environment, deciding an ambient noise of an input sound when the sound is coded really and putting the code books to its proper purpose. CONSTITUTION: This device is provided with a noise decision part 100 analyzing a digital sound signal obtained by an A/D converter 2 and deciding the noise environment of the input sound signal, plural coding side code books 11, 12, 13...1n beforehand storing plural sound source patterns based on the sound data obtained under different noise environment, a switching circuit 101 selecting the code book answering to the noise environment decided by the noise decision part 100 and connecting to a coding operation part 3, plural decoding side code books 21, 22, 23...2n of the same contents as the coding side respectively and the switching circuit 201 selecting the code book according to the data incorporated in a sound code obtained by the coding operation part 3 and connecting to a decoding operation part 4.

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 device, that is, a device for taking in voice, encoding it, decoding it and outputting it again as voice.

[0002]

2. Description of the Related Art A method of converting an analog voice waveform signal obtained by capturing a voice uttered by a person with a microphone into a code represented by a digital signal for transmission or recording is called voice encoding. On the other hand, converting the code obtained by voice coding into an analog waveform is called decoding, and if the analog voice waveform signal obtained by this is output from a speaker, it is reproduced as human-audible voice. .

[0003] The above-described speech signal coding / decoding method is widely used today, and one of them is a speech called a CELP (Code Excited Liner Prediction) method. The encoding system is promising for so-called digital telephones because it can transmit signals at a low bit rate, that is, with a smaller number of bits.

FIG. 8 is a block diagram showing a device configuration for realizing a conventional CELP speech coding system, that is, a structural example of a conventional CELP speech coding device. In FIG. 8, reference numeral 1 indicates a microphone, which takes in voice and outputs it as an analog voice waveform signal.

Reference numeral 2 is an A / D converter, which samples the analog voice waveform signal output from the microphone 1 at an appropriate sampling frequency and quantizes it.

Reference numeral 3 is an encoding operation unit, and details thereof are omitted, but correlation data by linear prediction and pitch prediction and its residual waveform are obtained from the quantized data obtained by the A / D converter 2. Although the input voice is encoded by these data, it is not possible to realize a low bit rate with the above residual waveform as it is, so one of a plurality of patterns prepared in advance as described later. Select to use the data that identifies it. The actual processing for this selection is to synthesize the data again in the encoding operation unit 3 and compare the result with the quantized data previously obtained by the A / D converter 2 to determine the optimum sound source. A method similar to pattern matching for selecting a pattern is adopted.

Reference numeral 10 is a coding side registration unit, and a plurality of sound source patterns necessary for selecting a sound source pattern by the coding operation unit 3 as described above are stored in advance as one codebook. It should be noted that each sound source pattern of the coding side codebook registered in advance in the coding side registration unit 10 is generally based on voice data sampled in an ideal environment with extremely low ambient noise. .

Reference numeral 4 is a decoding operation unit, which converts the voice code output from the encoding operation unit 3 into a digital voice signal. For this purpose, a codebook with the same contents as the coding side codebook registered in the coding side registration unit 10 is required, so a decoding side codebook with the same contents as the coding side codebook is registered. Decryption side registration unit 20
Is provided.

Reference numeral 5 is a D / A converter, which converts the digital voice signal obtained by the decoding operation section 4 into an analog voice waveform signal. This analog voice waveform signal is output from the speaker 6 and reproduced as voice that can be heard by a person.

In the conventional example shown in FIG. 8, the voice code output from the encoding operation unit 3 is immediately given to the decoding operation unit 4 and reproduced. However, in practice, for example, when the above-mentioned device is incorporated in a telephone, the portion on the encoding arithmetic unit 3 side and the portion on the decoding arithmetic unit 4 side are different telephones during a normal call, and A telephone line network is interposed between them. Further, when some storage means such as a memory IC, a flexible disk or the like is interposed between the encoding operation section 3 and the decoding operation section 4 in one telephone, it functions as so-called answering machine recording. It is possible.

[0011]

By the way, the conventional CELP
Since the voice coding method was realized by the above device, if the noise environment at the site where the voice is input to the microphone is different, the voice data originally collected in an ideal environment with extremely low ambient noise There is a possibility that processing on the encoding side that selects a sound source pattern from a codebook based on is not correctly performed.

The present invention has been made in view of such circumstances, and the first invention is to actually register a plurality of codebooks based on voice data obtained under various noise environments in advance, and Provide a speech coding / decoding device capable of more accurate and high-quality speech coding / decoding as compared with the conventional method by determining ambient noise of input speech at the time of speech coding and selectively using a codebook. The purpose is to

The second aspect of the present invention prepares a codebook in which a plurality of sound source patterns in which noise itself under various noise environments is used as data is stored, and the ambient noise of the input voice is judged at the time of actual voice coding, Speech coding that enables more accurate and high-quality speech coding / decoding than before by coding the input speech using the result of combining the corresponding sound source pattern and the basic sound source pattern -The purpose is to provide a decoding device.

[0014]

A speech coding / decoding apparatus according to the present invention includes an A / D converter for converting an input speech signal from an analog speech signal to a digital speech signal, and a plurality of sound source patterns registered in advance. The encoding side registration section, the encoding operation section for performing speech encoding of the digital audio signal obtained by the A / D conversion section using the sound source pattern registered in the encoding side registration section, and the encoding section. Selects and selects the sound source pattern data registered in the decoding side registration unit according to the decoding side registration unit that has registered the same contents as the registration contents of the side registration unit and the voice code obtained by the encoding calculation unit. The decoding operation unit that generates the data of the generated sound source pattern as a digital audio signal and the D / A conversion unit that converts the digital audio signal generated by the decoding operation unit into the analog audio signal are provided. The parts and decoding side registration unit, a plurality of sound sources pattern group obtained under different noise environments, each based on a group of audio data are pre-registered,
Furthermore, the digital speech signal obtained by the A / D conversion unit is analyzed to determine the noise environment of the input speech signal, and the sound source pattern group corresponding to the noise environment determined by the noise determination unit is encoded. According to the data included in the speech code obtained by the encoding operation unit, and the encoding side switching means selected from the side registration unit and connected to the encoding operation unit,
Decoding side switching means for selecting a sound source pattern group from the decoding side registration section and connecting it to the decoding calculation section is provided.

The voice encoding / decoding device according to the present invention is also provided with an A / D converter for converting an input voice signal from an analog voice signal to a digital voice signal, and an encoding side in which a plurality of sound source patterns are registered in advance. Registration unit, encoding operation unit that encodes the digital audio signal obtained by the A / D conversion unit using the sound source pattern registered in the encoding side registration unit, and registration of the encoding side registration unit The data of the sound source pattern registered in the decoding side registration unit is selected according to the decoding side registration unit that has registered the same contents as the contents and the voice code obtained by the encoding operation unit, and the selected sound source pattern of the selected sound source pattern is selected. It includes a decoding operation unit that generates data as a digital audio signal and a D / A conversion unit that converts the digital audio signal generated by the decoding operation unit into an analog audio signal. A plurality of ambient noise source patterns based on the ambient noise data obtained under a noisy environment and a plurality of basic source patterns based on the voice data obtained under a relatively low noise environment are registered in advance and decoded. In the side registration unit, the same contents as the registered contents of the basic sound source pattern in the encoding side registration unit are registered.In addition, the digital audio signal obtained by the A / D conversion unit is analyzed and noise of the input audio signal is registered. A noise determining unit that determines the environment, a switching unit that selects an environmental noise source pattern corresponding to the noise environment of the input speech signal determined by the noise determining unit from the encoding side registration unit, and an environmental noise that is selected by the switching unit. The present invention is characterized by comprising a synthesizing means for synthesizing a signal of a sound source pattern and a signal of each basic sound source pattern.

[0016]

In the speech coding / decoding apparatus of the present invention, the sound source pattern group connected by the coding side switching means.
The input speech signal is speech-encoded by the encoding operation unit using the (codebook), and the sound source pattern group (codebook) connected by the decoding side switching means is used to convert the speech code by the decoding operation unit. A digital audio signal is produced.

Further, in the speech encoding / decoding device of the present invention, the input speech signal is speech-encoded by the encoding operation section using the output of the synthesizing means, and the decoding side basic excitation pattern group (codebook). A digital voice signal is generated from the voice code by the decoding operation unit using only the voice code.

[0018]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below in detail with reference to the drawings showing the embodiments thereof. FIG. 1 is a block diagram showing a configuration example of a first embodiment of a speech encoding / decoding apparatus of the present invention, that is, a CELP speech encoding apparatus.

In FIG. 1, reference numeral 1 denotes a microphone, which takes in voice and outputs it as an analog voice waveform signal.

Reference numeral 2 is an A / D converter, which samples the analog voice waveform signal output from the microphone 1 at an appropriate sampling frequency and quantizes it.
The quantized data obtained by the A / D converter 2 is given to the encoding calculation unit 3 and the noise determination unit 100.

Although not described in detail, the encoding operation unit 3 obtains correlation data by linear prediction and pitch prediction from the quantized data obtained by the A / D converter 2 and obtains a residual waveform thereof. The input voice is encoded by these data, but in the present embodiment, it is prepared in advance in each of the n kinds of encoding side codebooks 11, 12, 13 ... 1n as described below. Note that these coding side codebooks 11,
12, 13 ... 1n are registered in the encoding side registration unit 10 in advance.

The coding side codebooks 11, 12, 13 ... 1n are
Each different noise environment, for example the coding side codebook (1) 11 is a basic codebook based on voice data sampled in an ideal environment where ambient noise is extremely low like the coding side codebook of the conventional example, The coding side codebook (2) 12 is a codebook based on the voice data collected in the crowd, and the coding side codebook (3) 13 is a codebook based on the voice data collected inside the car. , ..., Encoding codebook (n) 1n is a codebook based on the voice data collected in the station premises.

The quantized data output from the A / D converter 2 is given to the noise determination unit 100 as described above, and a feature is extracted by spectral analysis of the quantized data to extract the noise environment of the input voice, For example, the distinction between an environment where there is almost no noise, busy traffic, inside a car, ... The determination result by the noise determination unit 100 is output to the encoding calculation unit 3 and the switching circuit 101 as a signal (noise determination code) that specifies the noise environment of the input voice.

The switching circuit 101 is the noise determination unit described above.
In accordance with the noise judgment code given from 100, a corresponding one of a plurality of coding side codebooks 11, 12, 13 ... Connect to 3.

Therefore, the encoding operation section 3 outputs any one of the encoding side codebooks (1) 11 (or (2) 12, (3) 13 ... (n) 1n) connected by the switching circuit 101. The output signal is used to synthesize the output signal, and a sound source pattern is selected so that the synthesized speech is closest to the input speech. The data (for example, the codebook number and the pattern number) for identifying the sound source pattern selected in this way is combined with the previously obtained linear prediction and pitch prediction correlation data, and the noise judgment code to perform a decoding operation. It is output to the unit 4 as a voice code.

Reference numeral 4 is a decoding operation unit, which converts the voice code output from the encoding operation unit 3 into a digital voice signal. For this, the coding side codebook 1
Since a codebook having the same contents as 1, 12, 13 ... 1n is required, a plurality of decoding-side codebooks 21, 22, 23 ... 2n are registered in advance in the decoding operation unit 4 as well. A section 20 is provided. That is, the decoding side codebook (1) 21 is a basic codebook based on voice data sampled in an ideal environment where ambient noise is extremely low like the decoding side codebook of the conventional example. It has the same contents as Book (1) 11. The decoding side codebook (2) 22 is a codebook based on the audio data collected during the crowd, and has the same contents as the encoding side codebook (2) 12. Decoding side codebook (3) 23 is a codebook based on the audio data collected in the car.
(3) Same content as 13. And the decoding side codebook
(n) 2n is a codebook based on the voice data collected in the station yard, and has the same contents as the coding side codebook (n) 1n.

Reference numeral 201 is a switching circuit,
Any one of the decoding side codebooks 21, 22, 23, ... 2n registered in the encoding side registration unit 10 is connected to the decoding operation unit 4. Specifically, as described above, since the noise determination code determined by the noise determination unit 100 is also given from the encoding calculation unit 3 to the decoding calculation unit 4, the decoding calculation unit 4 switches the noise determination code. By giving it to the circuit 201, the switching circuit 201 selectively connects one decoding side codebook (1) 21 (or (2) 22, (3) 23 ... (n) 2n) to the decoding operation unit 4. To do.

Reference numeral 5 is a D / A converter, which converts the digital voice signal obtained by the decoding operation unit 4 into an analog voice waveform signal. This analog voice waveform signal is output from the speaker 6 and reproduced as voice that can be heard by a person.

In the embodiment of the present invention shown in FIG. 1, the speech code output from the encoding operation unit 3 is immediately given to the decoding operation unit 4 and reproduced. However, in actuality, for example, when the device of the present invention is incorporated in a telephone, the portion on the side of the encoding calculation unit 3 and the portion on the side of the decoding calculation unit 4 are different telephones during a normal call. A telephone line network is interposed between the two.
Further, when some storage means such as a memory IC, a flexible disk or the like is interposed between the encoding operation section 3 and the decoding operation section 4 in one telephone, it functions as so-called answering machine recording. It will be possible.

Next, regarding the operation of the apparatus of the present invention as described above, the flowchart of FIG. 2 showing the procedure of the encoding process,
This will be described with reference to the flowchart of FIG. 3 showing the procedure of noise determination processing by the noise determination unit 100 and the flowchart of FIG. 4 showing the procedure of decoding processing. In the following description, as an example, it is assumed that voice input to the microphone 1 is performed in a crowd.

When a voice is input to the microphone 1 from the outside (step S11), it is converted into an analog voice waveform and given to the A / D converter 2 (step S12). The A / D converter 2 quantizes the input analog voice waveform at a predetermined sampling frequency, and supplies the obtained quantized data to the encoding calculation unit 3 and the noise determination unit 100.

Here, first, the noise determination section 100 makes a noise determination (step S13). The specific processing procedure is shown in the flowchart of FIG. It should be noted that the noise determination unit 100 includes each codebook 11 (21), 12 (22), 13 (23).
The spectral characteristics of only noise in the noise environment corresponding to 1n (2n) are registered in advance.

First, the noise determination unit 100 extracts frame data for a certain period, for example, 5 seconds from the input speech (step S131), calculates the spectrum thereof (step S132), and based on the result, the spectrum Feature extraction is performed (step S133). Then, the noise determination unit 100 determines which of the pre-registered noise environments is the closest to the feature of the previously extracted spectrum (step S134). A noise determination code is output according to the determination result by the noise determination unit 100 (step S135) and is provided to the switching circuit 101 and the encoding calculation unit 3. Here, as described above, since the voice input to the microphone 1 is performed in the crowd, the noise determination unit 100 causes the switching circuit to output the noise determination code indicating that the noise environment is the crowd.
101 and the encoding calculation unit 3.

From the above, the encoding calculation unit 3 is the noise determination unit.
The noise determination code output by 100 is selected as it is (step S14), and the switching circuit 101 operates so as to connect the coding side codebook (2) 12 for crowding with the coding calculation unit 3. Therefore, the encoding calculation unit 3 performs the encoding process of the input voice by using each sound source pattern of the encoding side codebook (2) 12 for the crowd (step S15).

When determining which sound source pattern is the optimum, the coding operation unit 3 obtains the data (for example, the codebook number and the pattern number) for specifying the selected sound source pattern in advance by the linear prediction. , And outputs to the decoding operation unit 4 together with the correlation data by pitch prediction and the noise determination code.

The decoding operation unit 4 outputs the noise determination code in the data output from the encoding operation unit 3 to the switching circuit 201. As a result, the switching circuit 201
Is the codebook for crowding among the codebooks 21, 22, 23 ... 2n of the decryption side registered in advance in the decryption side registration unit 20.
(2) Since it operates so as to connect 22 to the decoding operation unit 4,
(Step S21), the decoding operation unit 4 uses the codebook (2) 22
Specified sound source pattern is selected from each sound source pattern of, and synthesized with other data and decoded (step
S22). As a result, the decoding calculation unit 4 outputs a digital voice waveform.

The digital voice waveform output from the decoding / calculation unit 4 is converted into an analog voice waveform by the D / A converter 5 (step S23) and output from the speaker 6 to be reproduced as human-audible voice. (Step S24).

Next, a second embodiment having the configuration shown in FIG. 1 will be described. In this second embodiment, the steps
In the noise determination process of S13, a plurality of noise environments are selected once and then finally reduced to one.

FIG. 5 is a flow chart showing the procedure of the noise judgment processing by the noise judgment unit 100 of the second embodiment. 5, steps S141, S142, and S143 are respectively shown in FIG.
The same process as steps S131, S132, and S133 of

When the spectral features are extracted in step S143, the noise determination unit 100 determines a plurality (m) of noise environment codes from the closest order, and outputs them to the encoding operation unit 3.

FIG. 6 is a flow chart showing the procedure for selecting the noise judgment code by the coding operation unit 3.

Since m noise determination codes are output by the processing of the noise determination unit 100 as described above, the encoding operation unit 3 first substitutes m for the parameter i (step S151).
, An optimal excitation pattern is selected from the codebook corresponding to the i-th rank noise determination code in the coding side codebooks 11, 12, 13 ... 1n (step S152). Next, the encoding calculation unit 3 sets the degree of approximation of this sound source pattern to the input speech as Xi, and compares it with the maximum degree of approximation Xmax in the past (step
S153), if Xi> Xmax, Xi is newly set to Xmax (step S155); otherwise, the parameter i is incremented by "1" (step S156).
. When the above steps S152 to S156 are repeated until the parameter i becomes "1" (step S157)
, The same processing is performed on all of the m noise determination codes determined previously.

As a result, the optimum sound source pattern is found corresponding to Xmax from any one of the codebooks corresponding to the m noise judgment codes previously judged, so that the coding operation unit 3 determines that Xmax is The noise determination code having the obtained rank is selected (step S158).

By performing the process of step S15 of FIG. 2 below, the encoding operation unit 3 performs encoding using the codebook corresponding to the selected noise determination code. Of course, the decoding process by the decoding calculation unit 4 is as shown in FIG.
It is exactly the same as the flowchart shown in FIG.

Next, a third embodiment will be described. FIG. 7 is a block diagram showing a configuration example for realizing the third embodiment of the voice encoding / decoding device of the present invention. Note that in FIG. 7, the same reference numerals as those in FIG. 1 referred to in the description of the above-described first embodiment indicate the same or corresponding portions.

In the third embodiment, as the coding side codebook, the basic codebook 30 and the environmental noise codebook 31 in which only various environmental noises are used as voice data are registered in the coding side registration unit 10 in advance. Has been done. Environmental noise codebook 31 includes a sound source pattern in which noise from crowds is used as voice data, a sound source pattern in which noise in the vehicle is used as voice data,
A sound source pattern and the like in which noise in the station yard is used as voice data is prepared.

Further, unlike the first embodiment shown in FIG. 1, the switching circuit of the third embodiment is different from that of the first embodiment.
101 is not provided, and the output of the environmental noise codebook 31 is configured to be given to the encoding operation unit 3 via the combining circuit 32. The output of the basic codebook 30 is also given to the synthesis circuit 32. That is, the output of the data of any sound source pattern of the basic codebook 30 and the output of the sound source pattern of any environmental noise selected by the noise determination code output from the noise determination unit 100 are synthesized by the synthesis circuit 32. It is configured to be provided to the encoding operation unit 3.

On the other hand, the decryption side registration unit 20 of the decryption calculation unit 4
Basic codebook with the same contents as Basic codebook 30
Only 40 are registered.

The operation of the third embodiment of the speech encoding / decoding apparatus of the present invention as described above is as follows.

Noise determination is performed by the noise determination unit 100, and one noise determination code is output according to the above first method.
This is the same as the embodiment. Here, as an example, it is assumed that voice input to the microphone 1 is performed in the crowd.

In this case, the noise judgment section 100 outputs the noise judgment code indicating that the noise environment of the input voice is crowded, so that the sound source pattern for crowdedness in the environmental noise codebook 31 is selected and synthesized. Output to the circuit 32. On the other hand, the encoding operation unit 3 sequentially outputs each sound source pattern in the basic codebook 30 to the synthesis circuit 32, and A / D converts the signal synthesized with the crowded sound source pattern output from the environmental noise codebook 31. Compared with the signal given from unit 2,
Select the optimal sound source pattern. As a result, the data specifying the sound source pattern of the basic codebook 30 (for example, the number of the sound source pattern) selected as a result is correlated with the previously obtained linear prediction and pitch prediction data, as in the first embodiment. And outputs to the decoding calculation unit 4. In addition,
In the third embodiment, the noise judgment code is not sent from the coding calculation unit 3 to the decoding calculation unit 4.

The decoding operation unit 4 first selects an excitation pattern from the basic codebook 40 in accordance with the data output from the encoding operation unit 3 as described above, synthesizes it with other data, and digitally synthesizes it. Generate a speech waveform.
The digital voice waveform generated by the decoding operation unit 4 is applied to the D / A converter 5 to be converted into an analog voice waveform, which is further output from the speaker 6 and reproduced as human-audible voice. .

As in the case of the second embodiment described above, also in the third embodiment, the noise judgment unit 100 obtains a plurality of judgment results, and a plurality of environments in the environment noise codebook 31 corresponding to them are obtained. It goes without saying that the optimum sound source pattern may be selected from the synthesis result of the synthesis circuit 32 of each noise source pattern and each sound source pattern in the basic codebook 30.

[0054]

As described in detail above, according to the speech coding / decoding apparatus of the present invention, a plurality of sound source pattern groups (codebooks) based on speech data obtained under various noise environments are generated. It is possible to perform more accurate and high-quality speech coding / decoding than before by preparing and using ambient sound of input speech at the time of actual speech coding and using different sound source pattern groups (codebooks). Become.

Further, according to the speech coding / decoding apparatus of the present invention, a sound source pattern group storing a plurality of sound source patterns in which noise itself under various noise environments is stored as data.
By preparing a (codebook), the ambient noise of the input speech is determined during actual speech coding, and the input speech is coded using the result of combining the sound source pattern corresponding to it and the basic sound source pattern. It is possible to perform more accurate and high-quality voice encoding / decoding than the conventional one.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration example of a first embodiment of a speech encoding / decoding device of the present invention.

FIG. 2 is a flowchart showing the procedure of the encoding process of the first embodiment of the audio encoding / decoding device of the present invention.

FIG. 3 is a flowchart showing a procedure of noise determination processing of the first embodiment of the speech encoding / decoding apparatus of the present invention.

FIG. 4 is a flowchart showing a procedure of a decoding process of the first embodiment of the audio encoding / decoding device of the present invention.

FIG. 5 is a flowchart showing a procedure of noise determination processing of the second embodiment of the speech encoding / decoding apparatus of the present invention.

FIG. 6 is a flowchart showing the procedure of an encoding process of the second embodiment of the audio encoding / decoding device of the present invention.

FIG. 7 is a block diagram showing a configuration example of a third embodiment of a speech encoding / decoding device of the present invention.

FIG. 8 is a block diagram showing a configuration example of a conventional speech encoding / decoding device.

[Explanation of symbols]

 2 A / D converter 3 Encoding operation unit 4 Decoding operation unit 5 D / A converter 10 Encoding side registration unit 11, 12, 13 ... 1n Encoding side codebook 20 Decoding side registration unit 21, 22, 23 ... 2n Decoding side codebook 30 Basic codebook 31 Environmental noise codebook 32 Synthesis circuit 40 Basic codebook 100 Noise judgment section 101 Switching circuit 201 Switching circuit

Claims (7)

[Claims] 1. An A / D conversion unit for converting an input audio signal from an analog audio signal to a digital audio signal, a registration unit in which a plurality of sound source patterns are registered in advance, and a digital signal obtained by the A / D conversion unit. A speech coding apparatus comprising: a coding calculation unit that performs speech coding of a speech signal using a sound source pattern registered in the registration unit, wherein the registration unit is different based on a group of speech data. A plurality of sound source pattern groups obtained in a noisy environment are registered in advance, and a noise judging section for judging the noise environment of the input sound signal by analyzing the digital sound signal obtained by the A / D converting section. And a switching unit that selects a sound source pattern group corresponding to the noise environment determined by the noise determination unit from the registration unit and connects the selected excitation pattern group to the encoding operation unit. The unit is configured to perform voice encoding of an input voice signal using a sound source pattern group connected by the switching means. 2. A registration unit in which a plurality of sound source patterns are registered in advance, and data of the sound source pattern registered in the registration unit is selected according to a given voice code, and the data of the selected sound source pattern is digitalized. A voice decoding device comprising a decoding operation unit for generating an audio signal and a D / A conversion unit for converting the digital audio signal generated by the decoding operation unit into an analog audio signal, which is included in the audio code. According to the data, there is provided switching means for selecting a sound source pattern group from the registration unit and connecting it to the decoding operation unit, wherein the decoding operation unit uses the sound source pattern group connected by the switching unit for speech coding. A speech decoding apparatus characterized in that a digital speech signal is generated from the speech decoding apparatus. 3. An A / D conversion unit for converting an input audio signal from an analog audio signal to a digital audio signal, a registration unit in which a plurality of sound source patterns are registered in advance, and a digital signal obtained by the A / D conversion unit. A speech coding apparatus comprising: a coding operation unit that performs speech coding of a speech signal using a sound source pattern registered in the registration unit, wherein the registration unit includes an environment obtained under different noise environments. A plurality of ambient noise source patterns based on noise data and a plurality of basic source patterns based on voice data obtained in a relatively low noise environment are registered in advance, and further obtained by the A / D conversion unit. A noise judgment unit for judging the noise environment of the input sound signal by analyzing the digital sound signal, and an environmental noise source pattern corresponding to the noise environment of the input sound signal judged by the noise judgment unit. The registering unit includes a switching unit to select from, and a synthesizing unit to synthesize the signal of the environmental noise source pattern and the signal of each basic source pattern selected by the switching unit, the encoding operation unit, the synthesizing unit A speech coding apparatus characterized in that an input speech signal is speech-coded by using the output of the speech coding apparatus. 4. The speech coding apparatus according to claim 1, wherein the noise determination unit is configured to obtain a plurality of noise environments as a determination result. 5. An A / D conversion unit for converting an input audio signal from an analog audio signal to a digital audio signal, an encoding side registration unit in which a plurality of sound source patterns are registered in advance, and the A / D conversion unit An encoding operation unit for performing voice encoding of the obtained digital voice signal using a sound source pattern registered in the encoding side registration unit, and a decoding in which the same content as the registration content of the encoding side registration unit is registered. Side registering unit, and according to the voice code obtained by the encoding calculation unit, the data of the sound source pattern registered in the encoding side registering unit is selected, and the data of the selected sound source pattern is generated as a digital voice signal. In a voice encoding / decoding device comprising a decoding operation unit for converting the digital audio signal generated by the decoding operation unit into an analog audio signal, The encoding side registration unit and the decoding side registration unit are registered in advance with a plurality of sound source pattern groups obtained under different noise environments based on a group of voice data, and further, the A / D conversion unit A noise determination unit that determines the noise environment of the input speech signal by analyzing the digital speech signal obtained by, and a sound source pattern group corresponding to the noise environment determined by the noise determination unit is selected from the encoding side registration unit. Then, according to the encoding side switching means connected to the encoding operation section and the data included in the voice code obtained by the encoding operation section, a sound source pattern group is selected from the decoding side registration section and the decoding is performed. And a decoding side switching means connected to the encoding side arithmetic operation section, wherein the encoding arithmetic side section uses the sound source pattern group connected by the encoding side switching means. Voice encoding, and the decoding arithmetic unit is adapted to generate a digital voice signal from a voice code by using a sound source pattern group connected by the decoding side switching means. -Decoding device. 6. An A / D conversion unit for converting an input audio signal from an analog audio signal to a digital audio signal, an encoding side registration unit in which a plurality of sound source patterns are registered in advance, and the A / D conversion unit An encoding operation unit for performing voice encoding of the obtained digital voice signal using a sound source pattern registered in the encoding side registration unit, and a decoding in which the same content as the registration content of the encoding side registration unit is registered. Side registering unit, and selects the sound source pattern data registered in the decoding side registering unit according to the voice code obtained by the encoding arithmetic unit, and generates the selected sound source pattern data as a digital voice signal. In a voice encoding / decoding device comprising a decoding operation unit for converting the digital audio signal generated by the decoding operation unit into an analog audio signal, The encoding side registration unit has a plurality of environmental noise sound source patterns based on the environmental noise data obtained under different noise environments and a plurality of basic sound source patterns based on the speech data obtained under a relatively low noise environment. Pre-registered, the decoding side registration unit has the same contents as the registered contents of the basic excitation pattern of the encoding side registration unit, and further, the digital content obtained by the A / D conversion unit. A noise determination unit that analyzes a voice signal to determine the noise environment of the input voice signal, and an environmental noise source pattern corresponding to the noise environment of the input voice signal determined by the noise determination unit is selected from the encoding side registration unit. And a synthesizing unit for synthesizing the signal of the environmental noise source pattern and the signal of each basic source pattern selected by the switching unit. The unit voice-encodes the input voice signal using the output of the synthesizing unit, and the decoding operation unit digitally converts the voice code using only the basic excitation pattern registered in the decoding-side registration unit. A voice encoding / decoding device characterized by being adapted to generate a voice signal. 7. The speech encoding / decoding device according to claim 5, wherein the noise determining unit is configured to obtain a plurality of noise environments as determination results.