JP2003140699A - Voice decoding device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a voice decoding device which can reduce the influence of a decoding error on a decoding process after recovery from a frame error if a frame error occurs during transmission as for a voice decoding device which receives encoded information generated by linear prediction analysis and outputs a synthesized voice. SOLUTION: The voice decoding device receives encoded information generated by linear prediction analysis and outputs the synthesized voice by using the encoded information and information held in an internal memory wherein voice information generated in the past is fed back and held, and is equipped with an error recovering means which updates the contents held in the internal memory by using the information held in the internal memory at least right after a frame error as to a frame recovered to a state wherein the encoded information can be received normally from the state of the frame error wherein the encoded information can not be received normally.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention receives coded information generated by linear prediction analysis, and uses the coded information and the contents stored in an internal memory for feeding back and storing voice information generated in the past. The present invention relates to a speech decoding apparatus that outputs synthetic speech, and particularly, a code error and loss of coding information during transmission (hereinafter, “code error” and “loss of coding information” are unified into one technical term. The present invention relates to a speech decoding apparatus capable of reducing the spread of decoding error in decoding processing after recovery from a frame error.

A speech coding apparatus and a speech decoding apparatus to which a linear predictive coding method is applied are used in a mobile telephone system and a VoIP (Voice over Internet Protocol) system which performs telephone voice communication on the Internet. The system cost is reduced by reducing the bandwidth of the applicable system, and the spread of the applicable system is promoted.

However, in the mobile phone system, since the wireless communication section is between the mobile phone and the base station, there is a possibility that a code error is mixed in the coded information due to various noises existing in space. Further, in the VoIP system, transmission information may be lost due to various factors. Once a frame error such as a code error or loss of transmission information occurs, a speech decoding apparatus that performs linear prediction reproduces speech information by using past coding information and speech information together with current coding information. Therefore, an error remains in the decoding result for a while even after recovering from the frame error, which causes deterioration of voice communication quality.

Therefore, it is strongly desired to realize a speech decoding apparatus capable of reducing an error in a decoding result after recovering from a frame error.

[0005]

2. Description of the Related Art FIG. 18 shows a configuration of a conventional speech decoding apparatus (No. 1) in the case of receiving encoded information in a normal state without frame error and outputting synthesized speech. There is. In the actual speech decoding system, the adaptive post filter and the like are arranged in the latter stage of the configuration of FIG.
In the present specification, their illustration is omitted.

In FIG. 18, 1 is an adaptive codebook for storing excitation information having periodicity after removing speech correlation, and 2
Is a fixed codebook for storing the excitation information of the noise component after removing the speech correlation, 3 is an adaptive codebook gain multiplier for multiplying the excitation information read from the adaptive codebook 1 by the adaptive codebook gain, 4
Is a fixed codebook gain multiplier for multiplying the excitation information read from the fixed codebook by a fixed codebook gain, and 5 is the sum of the outputs of the adaptive codebook gain multiplier 3 and the fixed codebook gain multiplier 4 to generate excitation information. The linear predictive coefficien 6 and the linear predictive coefficient 6 are abbreviated as “LPC coefficient” in the figure.
It is an abbreviation for the initial letter t, and it is not necessary to add a "coefficient", but it is customarily read as an LPC coefficient. ) And a linear prediction synthesis filter that outputs synthesized speech (in the figure,
It is abbreviated as "LPC synthesis filter". The subsequent description will be similarly described in the drawings. ), 7 is a memory for linear prediction synthesis that stores past synthetic speech information used when the linear prediction synthesis filter 6 generates synthetic speech (abbreviated as "LPC synthesis memory" in the figure. , Is also described in the figure).

The outline of the operation in the case of receiving the encoded information and outputting the synthesized voice in the normal state without the frame error is as follows. That is, a linear prediction coefficient, which is coding information, from the speech coding apparatus, an adaptive codebook index indicating a reading start position of excitation information from the adaptive codebook 1,
The fixed codebook index indicating the position at which the excitation information from the fixed codebook 2 is read, the adaptive codebook gain by which the excitation information read from the adaptive codebook 1 is multiplied, and the fixed codebook index by which the excitation information read from the fixed codebook 2 is multiplied are It will be sent.

The speech decoding apparatus configured as shown in FIG. 18 reads the excitation information from the adaptive codebook 1 with reference to the adaptive codebook index, and multiplies the adaptive codebook gain by the adaptive codebook gain multiplier 3. On the other hand, the excitation information is read from the fixed codebook 2 with reference to the fixed codebook index, and the fixed codebook gain multiplier 4 multiplies the fixed codebook gain.

Next, the outputs of the adaptive codebook gain multiplier 3 and the fixed codebook gain multiplier 4 are added up by an adder 5 to generate excitation information. Then, the stored content of the adaptive codebook is updated with the excitation information generated in preparation for the decoding process in the next frame. Further, in the linear prediction synthesis filter 6, the received linear prediction coefficient, the sound source information output from the adder 5, and the past synthetic speech information read from the linear prediction synthesis memory 7 are referred to generate the present synthetic speech. And output. Then, the contents stored in the memory 7 for linear prediction synthesis are updated with the synthesized speech information at the present time to prepare for the decoding process in the next frame.

FIG. 19 shows a configuration of a conventional speech decoding apparatus (part 2) in the case where encoded information cannot be obtained due to a frame error. The components are shown in Figure 1.
Since it is the same as the configuration of FIG. 8, the description of the components themselves is omitted, but in this case, the adder is not used and the output of the adaptive codebook gain multiplier 3 or the output of the fixed codebook gain multiplier 4 is not used. The difference from the configuration of FIG. 18 is that one of them is used as sound source information.

The outline of the operation for outputting the synthesized speech when the coded information cannot be obtained due to the occurrence of the frame error is as follows. That is, since the coding information cannot be obtained due to a frame error, the previous adaptive codebook index is used as the adaptive codebook index, and the fixed codebook index is randomly generated and used. The fixed codebook gain and the fixed codebook gain are used by attenuating a fixed coefficient by writing to the immediately preceding adaptive codebook gain and fixed codebook gain. Here, for the adaptive codebook gain and the fixed codebook gain, a fixed coefficient can be written to the immediately preceding adaptive codebook gain and the fixed codebook gain to be attenuated to use a large output of synthetic speech including an error in the decoding process. Rather, it is to reduce the influence of the error in the decoding process when the synthesized speech including the error in the decoding process is output in a smaller size when the sound is heard. Also, the previous linear prediction coefficient is used as the linear prediction coefficient.

In this case, either the output of the adaptive codebook gain multiplier 3 or the output of the fixed codebook gain multiplier 4 is set as the excitation information, depending on the strength of the periodicity of the immediately preceding excitation information.
The excitation information of the adaptive codebook is updated by either the output of the adaptive codebook gain multiplier 3 or the output of the fixed codebook gain multiplier 4. Then, the sound source information obtained as described above is supplied to the linear prediction synthesis filter 6, and the linear prediction coefficient immediately before the frame error occurs and the past synthetic speech information stored in the linear prediction synthesis memory 7 are referred to. Then, the synthesized speech is generated and output, and the stored content of the linear predictive synthesis memory 7 is updated with the synthesized speech information corresponding to the outputted synthesized speech.

Since the decoding process is performed as described above, the decoding process can be continued even when the coding information cannot be received due to a frame error. When the coding error is recovered, the received coding information is used. Decoding processing is performed while updating the adaptive codebook 1 and the linear prediction synthesis memory 7.

[0014]

However, in the case where a frame error occurs, while the frame error continues, as shown in FIG.
Since the decoding process as described above is performed by the above configuration, the excitation information stored in the adaptive codebook 1 and the linear prediction synthesis memory 7 are stored.
There is a risk that errors will accumulate in the synthesized voice information stored by.

In this state, even if the correct coded information can be received after recovering from the frame error, the information stored in the adaptive codebook 1 and the linear prediction synthesizing memory 7 immediately after recovering from the frame error has an error. Because of the inclusion, the generated sound source information and the synthesized speech information are affected by the error accumulated during the frame error, and the error does not immediately return to zero.

Therefore, the error in the information stored in the adaptive codebook 1 or the memory 7 for linear prediction synthesis cannot be immediately returned to zero, and this remains as an error in the generated sound source information or synthesized speech information. For this reason, the quality of the synthesized speech output from the speech decoding device is deteriorated, and the quality of the synthesized speech cannot be restored immediately. In view of such a problem, the present invention receives the coding information generated by the linear prediction analysis,
A speech decoding apparatus for outputting a synthesized speech from the encoded information and the contents held in an internal memory that feeds back and stores speech information generated in the past, and when a frame error occurs during transmission, An object of the present invention is to provide a speech decoding apparatus capable of reducing the spread of decoding error in the decoding processing after restoration.

[0017]

A first invention is an internal memory for receiving coded information generated by linear prediction analysis, feeding back the coded information and voice information generated in the past, and holding the information. In a voice decoding device that outputs a synthesized voice from holding information, in a frame that has returned to a state in which normal coding information can be received from a state in which a normal coding information cannot be received, at least immediately before the frame error. The speech decoding apparatus is characterized by comprising an error correction means for updating the contents held in the internal memory using the information held in the internal memory.

According to the first invention, the error correction means is
In the frame returned from the state of frame error where normal coding information cannot be received to the state where normal coding information can be received, at least the holding information of the internal memory immediately before the frame error is used to hold the contents of the internal memory. Since the stored information in the internal memory can be approximated to the correct stored information in the frame recovered from the frame error, the spread of the decoding error in the subsequent decoding processing can be reduced.

A second invention is the speech decoding apparatus of the first invention, wherein the error correction means stores information immediately before a frame error, which is stored in an adaptive codebook constituting the speech decoding apparatus. Source information holding means for holding the source code information and at least the adaptive codebook information held in the source code information holding means to obtain the source code information, and the source code calculation for updating the adaptive codebook in the frame recovered from the frame error. And a means for decoding the speech.

According to the second aspect of the invention, the error recovery means stores the excitation information storage means for storing the information immediately before the frame error stored in the adaptive codebook constituting the speech decoding device, and the excitation source. Storing the adaptive codebook, since it includes excitation information calculation means for calculating the excitation using the information of the adaptive codebook held in the information holding means and updating the adaptive codebook in the frame recovered from the frame error. It is possible to approximate the correct adaptive codebook information in the frame in which the information is recovered from the frame error, and reduce the spread of decoding error in the subsequent decoding processing.

A third invention is the speech decoding apparatus according to the first invention, wherein the error correction means stores the synthesized speech information synthesized by the prediction synthesis filter constituting the speech decoding apparatus. The prediction memory is provided with a memory unit for predictive synthesis that holds the synthesized speech information stored immediately before the frame error, and in the frame recovered from the frame error,
A speech decoding apparatus characterized in that the storage information held by the prediction synthesis memory holding means is given to the prediction synthesis filter to generate synthetic speech information, and the prediction synthesis memory is updated by the generated synthetic speech information. Is.

According to the third aspect of the present invention, the error recovery means stores the synthesized speech information synthesized by the predictive synthesis filter constituting the speech decoding device, and stores it in the predictive synthesis memory immediately before the frame error. A prediction synthesizing memory holding unit for holding synthetic speech information is provided, and in a frame recovered from a frame error, storage information held by the prediction synthesizing memory holding unit is given to the prediction synthesizing filter to generate synthetic speech information, Since the prediction synthesis memory is updated with the generated synthesized speech information, the storage information of the prediction synthesis memory can be approximated to the correct storage information in the frame recovered from the frame error. Ripple can be reduced.

A fourth aspect of the invention is the speech decoding apparatus of the first aspect of the invention, in which the error repairing means stores sound source information holding means for holding synthesized speech information immediately before a frame error and a frame recovered from the frame error. In the above, the method further comprises excitation information calculation means for generating adaptive codebook information from the synthesized speech information held by the excitation information holding means and the linear prediction coefficient of the frame recovered from the frame error, and updating the adaptive codebook. Is a speech decoding device characterized by.

According to the fourth aspect of the invention, the error repairing means holds the sound source information holding means for holding the synthesized speech information immediately before the frame error and the synthesis held by the sound source information holding means for the frame recovered from the frame error. Voice information,
Since the adaptive codebook information is generated from the linear prediction coefficient of the frame recovered from the frame error and the excitation code information is updated, a frame in which the stored information of the adaptive codebook is recovered from the frame error The correct stored information can be approximated, and the spread of decoding error in the subsequent decoding processing can be reduced.

A fifth invention is a speech decoding device of the first invention.
And the error correction means detects the synthesized speech information from the synthesized speech information.
Calculate the frame power and calculate the frame power immediately before the frame error.
Power E ₁Frame power calculation means that holds the
Frame power calculation for the frame returned from
From the means, the frame power E immediately before the frame error₁Receive
However, the synthesized sound in the frame recovered from the frame error
Generates voice information and recovers from the frame error
Frame power E in₂And the adaptive codebook is stored
E information₁/ E₂Sound source information calculator that doubles
A speech decoding apparatus comprising: a stage.

According to the fifth aspect of the invention, the error recovery means calculates the frame power from the synthesized voice information and holds the frame power E ₁ immediately before the frame error, and the error recovery means recovers from the frame error. In the frame, the frame power E ₁ immediately before the frame error is received from the frame power calculating means, synthetic speech information in the frame recovered from the frame error is generated, and the frame power E ₂ in the frame recovered from the frame error is calculated. , And the excitation information calculation means for updating the information stored in the adaptive codebook by E ₁ / E ₂ times, so that the stored information in the adaptive codebook approximates the correct stored information in the frame recovered from the frame error. Therefore, it is possible to reduce the spread of decoding error in the subsequent decoding processing.

[0027]

DETAILED DESCRIPTION OF THE INVENTION The technique of the present invention will be described in detail below with reference to the drawings. In the first embodiment of the present invention, FIG. 1 shows a solid line for exchanging information while a frame error continues, and a broken line for exchanging information after recovering from a frame error.

In FIG. 1, reference numeral 1 is an adaptive codebook for storing excitation information having periodicity after speech correlation is removed, and 2 is a fixed code for storing excitation information of noise component after speech correlation is removed. Book 3, 3 is an adaptive codebook gain multiplier that multiplies the excitation information read from adaptive codebook 1 by the adaptive codebook gain, and 4 is a fixed codebook gain multiplier that multiplies the excitation information read from the fixed codebook by the fixed codebook gain Reference numeral 5 denotes an adder that adds the outputs of the adaptive codebook gain multiplier 3 and the fixed codebook gain multiplier 4 to generate excitation information, and 6 denotes the excitation information output by the adder 5 and the received linear prediction coefficient. A linear prediction synthesis filter for receiving and outputting synthetic speech, 7 is a linear prediction synthesis filter 6
Is a memory for linear predictive synthesis that stores past synthetic speech information used when generating synthetic speech. The above constituent elements are constituent elements of a conventional speech decoding apparatus.

Reference numeral 8 is an excitation information holding means (information is omitted in the figure) for holding the excitation information immediately before the frame error stored in the adaptive codebook 1, and 9 is held in the excitation information holding means 8. Information is stored in the fixed codebook 2 and the stored information in the fixed codebook 2 is used to calculate the excitation information, and the stored information in the adaptive codebook 1 is updated in the frame recovered from the frame error (information is omitted in the figure. The sound source information holding means 8 and the sound source information holding means 9 constitute error correction means. Then, error information indicating whether or not there is a frame error is given to the sound source information holding unit 8 and the sound source information calculating unit 9, and the operations of the sound source information holding unit 8 and the sound source information calculating unit 9 are in the frame error state. It depends on

While the frame error continues, the speech decoding apparatus carries out the decoding process according to the configuration of FIG.
Therefore, the feature of the configuration of FIG. 1 is that the excitation information holding means 8 holds the information held in the adaptive codebook 1 immediately before the frame error, and the information held by the excitation information holding means 8 is fixed. The point is that the excitation information calculation means 9 calculates the excitation information using the storage information of the codebook 2 and updates the storage information of the adaptive codebook 1 in the frame recovered from the frame error.

FIG. 2 is a flowchart showing the error correction operation of the configuration of FIG. 1, which is an operation flowchart of the voice information holding means 8 and the sound source information calculating means 9 of FIG. After that, FIG.
The error correction operation of the configuration of FIG.
First, the operation of the sound source information holding means is as follows. S1. It is monitored whether or not there is a frame error (abbreviated as "error" in the figure. The same applies to the figures thereafter).

S2. If there is no frame error in step S1 (No), the held information in the excitation information holding means is rewritten with the held contents in the adaptive codebook, and the process jumps to step S1. In FIG. 2, what is simply described as “memory” is a memory that is held by the excitation information holding means and that holds the contents of the adaptive codebook. On the other hand, if there is no frame error in step S1 (Yes), the process is temporarily terminated and stands by.

Therefore, the memory held by the excitation information holding means holds the information stored in the adaptive codebook immediately before the frame error occurs. Next, the operation of the sound source information calculation means is as follows. S11. It is monitoring whether or not the frame error has recovered. If the frame error has not been recovered (No), the process is once terminated and waits.

S12. In step S11, if the frame error is recovered (Yes), the stored information of the adaptive codebook immediately before the frame error occurs is acquired from the excitation information holding unit. S13. The error frame number counter i is set to 1. S14. The pitch before and after the frame error is linearly interpolated to calculate the pitch P _i during the frame error.

This may be calculated as follows. That is, the pitch of the frame immediately before the frame error is P ₁ ,
When the pitch of the frame immediately after the frame error is P ₂ , i
Is the count value of the error frame number counter and N is the error frame number, P _i is given by the equation (1). P _i = P ₁ + (P ₂ −P ₁ ) · i / N (1) S15. It is determined whether the pitch difference between the frames before and after the frame error is smaller than a predetermined threshold value.

If the pitch difference between the frames before and after the frame error is smaller than a predetermined threshold value, it means that the periodicity is strong, and if the pitch difference between the frames before and after the frame error is larger than the predetermined threshold value, the periodicity is weak and the noise is large. Means that. Therefore, the subsequent processing is branched depending on the strength of the periodicity. S16. In step S15, when the pitch difference between the frames before and after the frame error is smaller than the predetermined threshold value (Yes), the sound source information is calculated using the held information of the sound source information holding unit and the pitch P _i . It should be noted that although abbreviated as “using adaptive codebook” in the figure, a simplified representation is used because the excitation information holding means holds the information stored in the adaptive codebook immediately before the frame error. ing. ) S17. On the other hand, in step S15, when the pitch difference between the frames before and after the frame error is larger than a predetermined threshold (N
In o), the sound source information is calculated using the stored information in the fixed codebook.

S18. The adaptive codebook is updated with the excitation information calculated in either step S16 or step S17, corresponding to the determination result in step S15. S19. The error frame number counter i is incremented. S20. It is determined whether or not the count value of the error frame number counter i is larger than the error frame number N.

When the count value of the error frame number counter i is smaller than the error frame number N (No), the process jumps to step S15 and the subsequent processes are continued. On the other hand, when the count value of the error frame number counter i is larger than the error frame number N (Yes), the processing is ended. As described above, the excitation information holding means holds the storage information of the adaptive codebook immediately before the frame error, and when recovering from the frame error, the excitation information calculating means stores the storage information of the adaptive codebook immediately before the frame error and the frame error. Since the excitation information is calculated based on the pitch of the frame, or the excitation information is calculated from the storage information in the fixed codebook, the storage information in the adaptive codebook after recovery from the frame error will approximate the excitation information immediately before the frame error. , The decoding error in the subsequent decoding process can be reduced, and the spread of the decoding error in the decoding process after the recovery from the frame error can be reduced.

FIG. 3 shows a second embodiment of the present invention. In FIG. 3, reference numeral 1 is an adaptive codebook that stores excitation information having periodicity after removing speech correlation, and 2 is a fixed codebook that stores excitation information of noise components after removing speech correlation. Is an adaptive codebook gain multiplier that multiplies the excitation information read from the adaptive codebook 1 by the adaptive codebook gain, 4 is a fixed codebook gain multiplier that multiplies the excitation information read from the fixed codebook by the fixed codebook gain, and 5 is An adder that adds the outputs of the adaptive codebook gain multiplier 3 and the fixed codebook gain multiplier 4 to generate excitation information, and 6 receives the excitation information output from the adder 5 and the received linear prediction coefficient to synthesize the excitation information. A linear predictive synthesis filter for outputting speech, 7 is a memory for linear predictive synthesis that stores past synthetic speech information used when the linear predictive synthesis filter 6 generates synthetic speech, and the above components are conventional speech decoding units. Configurator It is.

Reference numeral 8 denotes a sound source information holding means for holding the sound source information immediately before the frame error, which is stored in the adaptive code book 1, and 9a receives the information held in the sound source information holding means 8 and receives the adaptive code book. 1 is a sound source information calculating means for updating the stored information, and is a sound source information holding means 8 and a sound source information calculating means 9
The error correction means is constituted by a. Then, error information indicating whether or not there is a frame error is given to the sound source information holding unit 8 and the sound source information calculating unit 9a, and the operations of the sound source information holding unit 8 and the sound source information calculating unit 9a are in the frame error state. It depends on

While the frame error continues, the speech decoding apparatus carries out the decoding process according to the configuration of FIG.
Therefore, the feature of the configuration of FIG. 3 is that the information held in the adaptive codebook 1 immediately before the frame error is held in the sound source information holding unit 8 and the information held in the sound source information holding unit 8 is set as the sound source. The point is that the information stored in the adaptive codebook 1 is updated in the frame received by the information calculation means 9a and recovered from the frame error.

FIG. 4 is a flowchart showing the error correcting operation of the configuration of FIG. 3, which is an operation flowchart of the sound source information holding means 8 and the sound source information calculating means 9a of FIG. Hereinafter, the error correction operation of the configuration of FIG. 3 will be described with reference to the reference numerals of FIG. First, the operation of the sound source information holding means is as follows. S1. It is monitoring whether there was a frame error.

S2. If there is no frame error in step S1 (No), the holding information of the excitation information holding means is rewritten with the holding information of the adaptive codebook, and the process jumps to step S1. On the other hand, if there is no frame error in step S1 (Yes), the process is temporarily terminated and stands by.

Therefore, the memory held by the excitation information holding means holds the information stored in the adaptive codebook immediately before the frame error occurs. Next, the operation of the sound source information calculation means is as follows. S31. It is monitoring whether or not the frame error has recovered. If the frame error has not been recovered (No), the process is once terminated and waits.

S32. In step S31, when the frame error is recovered (Yes), the adaptive codebook storage information immediately before the frame error occurs is acquired from the excitation information holding unit. S33. The adaptive codebook is updated with the information stored in the adaptive codebook immediately before the frame error acquired from the excitation information holding means, and the process ends.

As described above, in the case of the configuration of FIG. 3, the excitation information holding means 8 holds the information held in the adaptive codebook 1 immediately before the frame error, and the excitation information holding means 8 is held.
The sound source information calculation means 9a receives the information held by
Since the stored information in the adaptive codebook 1 is updated in the frame recovered from the frame error, the stored information in the adaptive codebook after the frame error recovery becomes the excitation information immediately before the frame error, and the decoding error in the subsequent decoding process can be reduced. Therefore, it is possible to reduce the spread of decoding error in the decoding process after recovering from the frame error.

In the error recovery means of the configuration shown in FIG. 1, the excitation information holding means holds the storage information of the adaptive codebook immediately before the frame error, and when the error recovery means recovers from the frame error, the excitation information calculation means immediately before the frame error. Of the adaptive codebook and the pitch of the frame in the frame error, or the excitation information is calculated from the stored information of the fixed codebook to update the adaptive codebook. The error recovery means having the above-mentioned configuration stores in the excitation information storage means 8 the information stored in the adaptive codebook 1 immediately before the frame error,
The information stored in the excitation information storage unit 8 is received by the excitation information calculation unit 9, and the storage information of the adaptive codebook 1 is updated in the frame recovered from the frame error.
This is a simplified configuration and operation of.

Therefore, the configuration of FIG. 1 and the configuration of FIG. 3 are at least excitation information storage means for storing the information immediately before the frame error, which is stored in the adaptive codebook constituting the speech decoding apparatus, and at least the excitation information storage means. It is said that error correction means is configured with excitation information calculation means for obtaining excitation information using the information of the adaptive codebook held in the excitation information holding means, and updating the adaptive codebook in the frame recovered from the frame error. be able to.

FIG. 5 shows a third embodiment of the present invention. In FIG. 5, reference numeral 1 denotes an adaptive codebook that stores excitation information having periodicity after speech correlation is removed, and 2 is a fixed codebook that stores noise component excitation information after speech correlation is removed, 3 Is an adaptive codebook gain multiplier that multiplies the excitation information read from the adaptive codebook 1 by the adaptive codebook gain, 4 is a fixed codebook gain multiplier that multiplies the excitation information read from the fixed codebook by the fixed codebook gain, and 5 is An adder that adds the outputs of the adaptive codebook gain multiplier 3 and the fixed codebook gain multiplier 4 to generate excitation information, and 6 receives the excitation information output from the adder 5 and the received linear prediction coefficient to synthesize the excitation information. A linear predictive synthesis filter for outputting speech, 7 is a memory for linear predictive synthesis that stores past synthetic speech information used when the linear predictive synthesis filter 6 generates synthetic speech, and the above components are conventional speech decoding units. Configurator It is.

Reference numeral 10 denotes a linear prediction synthesis memory holding means (in the figure, abbreviated as "synthesis memory holding means" for holding the information stored in the linear prediction synthesis memory 7 immediately before the frame error. In the same manner, the error recovery means is constituted by the linear prediction synthesis memory holding means 10 and the linear prediction synthesis filter 6. Then, error information indicating whether or not there is a frame error is given to the linear prediction synthesis memory holding means 10, and the operation of the linear prediction synthesis memory holding means 10 differs depending on whether or not there is a frame error state.

While the frame error continues, the speech decoding apparatus carries out the decoding process according to the configuration of FIG.
Therefore, the feature of the configuration of FIG. 5 is that the linear prediction synthesis filter that constitutes the speech decoding device is provided with a prediction synthesis memory holding unit that holds the synthesized speech information synthesized immediately before the frame error, and in the frame recovered from the frame error. The point is that the information held by the memory for holding linear prediction synthesis is given to the linear prediction synthesis filter to generate synthetic speech information, and the stored information in the memory for linear prediction synthesis is updated by the generated synthetic speech information.

FIG. 6 is a flow chart showing the error recovery operation of the configuration of FIG. 5, which is an operation flow chart of the memory unit 10 for linear prediction synthesis and the linear prediction synthesis filter of FIG. Hereinafter, the error correction operation of the configuration of FIG. 5 will be described with reference to the symbols of FIG. First, the operation of the memory holding means for linear prediction synthesis is as follows. S41. It is monitoring whether there was a frame error.

S42. If there is no frame error in step S41 (No), the information held in the linear prediction synthesis memory holding means is rewritten with the information held in the linear prediction synthesis memory, and the process jumps to step S41. On the other hand, in step S41, if there is no frame error (Yes), the process is once terminated and stands by.

Therefore, the memory held by the linear prediction synthesis memory holding means holds the storage information of the linear prediction synthesis memory immediately before the frame error occurs. Then, the operation of the linear prediction synthesis filter is as follows. S51. It is monitoring whether or not the frame error has recovered. If the frame error has not been recovered (No), the process is once terminated and waits.

S52. In step S51, if the frame error is recovered (Yes), the holding information is acquired from the linear prediction synthesis memory holding unit. S53. Synthesize voice. S54. The memory for linear prediction synthesis is updated with the synthesized voice information, and the process ends.

As described above, in the case of the configuration of FIG. 5, the linear prediction synthesis memory holding means 10 holds the information of the linear prediction synthesis memory immediately before the frame error, and the linear prediction is performed when the frame error is recovered. The linear prediction synthesis filter generates synthetic speech using the information held in the synthesis memory holding means, and the linear prediction synthesis memory is updated with the generated synthetic speech information, so that the linear prediction synthesis memory stores it immediately after a frame error. The synthesized speech information can approximate the synthesized speech immediately before the frame error, reduce the decoding error in the subsequent decoding processing, and reduce the spread of the decoding error in the decoding processing after returning from the frame error. be able to.

FIG. 7 shows a fourth embodiment of the present invention. In FIG. 7, 1 is an adaptive codebook for storing excitation information having periodicity after removing speech correlation, 2 is a fixed codebook for storing excitation information of noise components after removing speech correlation, 3 Is an adaptive codebook gain multiplier that multiplies the excitation information read from the adaptive codebook 1 by the adaptive codebook gain, 4 is a fixed codebook gain multiplier that multiplies the excitation information read from the fixed codebook by the fixed codebook gain, and 5 is An adder that adds the outputs of the adaptive codebook gain multiplier 3 and the fixed codebook gain multiplier 4 to generate excitation information, and 6 receives the excitation information output from the adder 5 and the received linear prediction coefficient to synthesize the excitation information. A linear predictive synthesis filter for outputting speech, 7 is a memory for linear predictive synthesis that stores past synthetic speech information used when the linear predictive synthesis filter 6 generates synthetic speech, and the above components are conventional speech decoding units. Configurator It is.

Reference numeral 8 denotes a sound source information holding means for holding the sound source information immediately before the frame error, which is stored in the adaptive code book 1, and 9a receives the information held in the sound source information holding means 8 and receives the adaptive code book. The sound source information calculating means 10 for updating the stored information of 1 is a linear predictive synthesizing memory holding means for holding the stored information immediately before the frame error of the linear predictive synthesizing memory 7, and the sound source information holding means 8 and the sound source information calculating means 9a. The error holding means is constituted by the linear prediction synthesis memory holding means 10. Then, the sound source information holding means 8, the sound source information calculating means 9a, and the linear predictive synthesis memory holding means 10 are provided with error information indicating whether or not there is a frame error, and the sound source information holding means 8 and the sound source information calculating means. 9a and the operation of the linear prediction synthesis memory holding means 10 differ depending on whether or not a frame error state exists.

While the frame error continues, the speech decoding apparatus carries out the decoding process according to the configuration of FIG.
By the way, comparing FIGS. 3 and 5 with FIG. 7, it can be seen that the error repairing means in the configuration of FIG. 7 is made up of the error repairing means of the configuration of FIG. 3 and the error repairing means of the configuration of FIG. . Therefore, further description of the configuration of FIG. 7 is omitted.

FIG. 8 shows a fifth embodiment of the present invention. In FIG. 8, reference numeral 1 is an adaptive codebook for storing excitation information having periodicity after speech correlation is removed, 2 is a fixed codebook for storing noise source excitation information after speech correlation is removed, 3 Is an adaptive codebook gain multiplier that multiplies the excitation information read from the adaptive codebook 1 by the adaptive codebook gain, 4 is a fixed codebook gain multiplier that multiplies the excitation information read from the fixed codebook by the fixed codebook gain, and 5 is An adder that adds the outputs of the adaptive codebook gain multiplier 3 and the fixed codebook gain multiplier 4 to generate excitation information, and 6 receives the excitation information output from the adder 5 and the received linear prediction coefficient to synthesize the excitation information. A linear predictive synthesis filter for outputting speech, 7 is a memory for linear predictive synthesis that stores past synthetic speech information used when the linear predictive synthesis filter 6 generates synthetic speech, and the above components are conventional speech decoding units. Configurator It is.

Reference numeral 8a denotes a sound source information holding means for holding the synthesized speech information immediately before the frame error, and 9b denotes the synthesized speech information held by the sound source information holding means 8a and the frame restored from the frame error in the frame restored from the frame error. Excitation information storage means 8 generates excitation information from the linear prediction coefficient and updates the adaptive codebook.
A and the sound source information calculation means 9b constitute error correction means. Then, the sound source information holding unit 8a and the sound source information calculating unit 9b are provided with error information indicating whether or not there is a frame error, and the operations of the sound source information holding unit 8a and the sound source information calculating unit 9b are in the frame error state. It depends on

While the frame error continues, the speech decoding apparatus carries out the decoding process according to the configuration of FIG.
Therefore, the feature of the configuration of FIG. 8 is that the sound source information holding unit 8a holds the synthesized speech information immediately before the frame error, and the sound source information calculation unit 9b receives the information held by the sound source information holding unit 8a. The point is that information is generated and the stored information of the adaptive codebook 1 is updated in the frame recovered from the frame error.

FIG. 9 is a flow chart showing an error correction operation of the configuration of FIG. 8, which is an operation flow chart of the sound source information holding means 8a and the sound source information calculating means 9b. After that, FIG.
The error recovery operation of the configuration of FIG. 8 will be described with reference to symbols.
First, the operation of the sound source information holding means is as follows. S61. It is monitoring whether there was a frame error.

S62. If there is no frame error in step S61 (No), the held information of the sound source information holding means is rewritten with the information of the synthesized voice, and the step S61 is performed.
Jump to 61. On the other hand, in step S61, if there is no frame error (Yes), the process is temporarily terminated and the process stands by.

Therefore, the memory held by the sound source information holding means holds the synthesized speech information immediately before the frame error occurs. Next, the operation of the sound source information calculation means is as follows. S71. It is monitoring whether or not the frame error has recovered. If the frame error has not been recovered (No), the process is once terminated and waits.

S72. In step S71, if the frame error is recovered (Yes), the holding information is acquired from the sound source information holding unit. S73. The linear prediction coefficient of the frame recovered from the frame error is input to the linear prediction analysis filter forming the sound source information calculation means. S74. The linear prediction analysis filter calculates the sound source information based on the information acquired from the sound source information holding unit and the linear prediction coefficient of the frame recovered from the frame error.

The linear prediction analysis filter is the same as that installed on the side of the speech coder. S75. The adaptive codebook is updated with the sound source information calculated by the linear prediction analysis filter. As described above, in the case of the configuration of FIG. 9, the sound source information holding means 8
The synthetic speech information immediately before the frame error is held in a,
In the frame recovered from the frame error, the excitation information calculation means 9b calculates the excitation information and updates the adaptive codebook, so that the excitation information stored in the adaptive codebook 1 immediately after the frame error is the excitation information immediately before the frame error. It can be approximated, the decoding error in the subsequent decoding process can be reduced, and the spread of the decoding error in the decoding process after recovering from the frame error can be reduced.

FIG. 10 shows a sixth embodiment of the present invention. In FIG. 10, 1 is an adaptive codebook that stores excitation information having periodicity after removing speech correlation, 2 is a fixed codebook that stores excitation information of noise components after removing speech correlation, 3 Is an adaptive codebook gain multiplier that multiplies the excitation information read from the adaptive codebook 1 by the adaptive codebook gain, 4 is a fixed codebook gain multiplier that multiplies the excitation information read from the fixed codebook by the fixed codebook gain, and 5 is An adder that adds the outputs of the adaptive codebook gain multiplier 3 and the fixed codebook gain multiplier 4 to generate excitation information, and 6 receives the excitation information output from the adder 5 and the received linear prediction coefficient to synthesize the excitation information. A linear predictive synthesis filter for outputting speech, 7 is a memory for linear predictive synthesis that stores past synthetic speech information used when the linear predictive synthesis filter 6 generates synthetic speech, and the above components are conventional speech decoding units. Configuration Is the elementary.

Reference numeral 11 is a frame power calculating means for calculating the frame power from the synthesized speech information and holding the frame power E ₁ immediately before the frame error, and 9c is a frame from the frame power calculating means 11 for the frame recovered from the frame error. Information stored in the adaptive codebook by receiving the frame power E ₁ immediately before the error, generating synthetic speech information in the frame recovered from the frame error, and calculating the frame power E ₂ in the frame recovered from the frame error. Is a sound source information calculating unit that updates the time by E ₁ / E ₂ times, and the frame power calculating unit 11 and the sound source information calculating unit 9c constitute an error correcting unit. Then, error information indicating whether or not there is a frame error is given to the frame power calculation means 11 and the sound source information calculation means 9c, and the operations of the frame power calculation means 11 and the sound source information calculation means 9c are in the frame error state. It depends on

While the frame error continues, the speech decoding apparatus carries out the decoding process according to the configuration of FIG.
Therefore, the feature of the configuration of FIG. 10 is that the frame power calculation means calculates the frame power from the synthesized speech information, holds the frame power E ₁ immediately before the frame error, and the sound source information calculation means 9c recovers from the frame error. At
The frame power E ₁ immediately before the frame error is received from the frame power calculation means 11, synthetic speech information in the frame recovered from the frame error is generated, the frame power E ₂ in the frame recovered from the frame error is calculated, and the adaptive code is calculated. The point is to update the information stored in the book by E ₁ / E ₂ times.

FIG. 11 is a flowchart showing the error correction operation of the configuration of FIG. 10, which is an operation flowchart of the flowchart power calculation means and the sound source information calculation means. Or later,
The error recovery operation of the configuration of FIG. 10 will be described with reference to the reference numerals of FIG. First, the operation of the flowchart power calculation means is as follows. S81. It is monitoring whether there was a frame error.

S82. In step S81, if there is no frame error (No), the frame power E ₁ is calculated for the synthetic speech. Jump to. S83. With the frame E ₁ calculated in step S82, the memory held by the frame power calculation means is updated, and the process jumps to step S81.

On the other hand, in step S81, if there is no frame error (Yes), the process is once terminated and stands by. Therefore, the frame power E ₁ immediately before the frame error occurs is held in the memory of the frame power calculation means. Next, the operation of the sound source information calculation means is as follows.

S91. It is monitoring whether or not the frame error has recovered. If the frame error has not been recovered (No), the process is once terminated and waits. S92. In step S91, if the frame error is recovered (Yes), the frame power E ₁ is acquired from the frame power calculation means.

S93. The synthesized speech at the time of recovery from a frame error is calculated. S94. The frame power E ₂ at the time of recovering from the frame error is calculated from the synthesized voice calculated in step S93. S95. The information held in the adaptive codebook is updated to E ₁ / E ₂ times. Here, the information held in the adaptive codebook is updated to E ₁ / E ₂ times because the adaptive codebook gain and the fixed codebook gain are attenuated during the frame error to generate the synthesized speech, and thus the frame error is generated. Immediately after returning from the above, the frame power of the synthesized voice is smaller than that when no frame error occurred, and its attenuation can be approximated by E ₁ / E ₂ . Therefore, if the voice information stored in the adaptive codebook is updated by E ₁ / E ₂ times when the frame error is recovered, the continuity of the frame power immediately before the frame error and immediately after the frame error is improved. The decoding error in the decoding process can be reduced, and the spread of the decoding error in the decoding process after the recovery from the frame error can be reduced.

FIG. 12 shows the seventh embodiment of the present invention. In FIG. 12, 1 is an adaptive codebook that stores excitation information having periodicity after removing speech correlation, 2 is a fixed codebook that stores excitation information of noise components after removing speech correlation, 3 Is an adaptive codebook gain multiplier that multiplies the excitation information read from the adaptive codebook 1 by the adaptive codebook gain, 4 is a fixed codebook gain multiplier that multiplies the excitation information read from the fixed codebook by the fixed codebook gain, and 5 is An adder that adds the outputs of the adaptive codebook gain multiplier 3 and the fixed codebook gain multiplier 4 to generate excitation information, and 6 receives the excitation information output from the adder 5 and the received linear prediction coefficient to synthesize the excitation information. A linear predictive synthesis filter for outputting speech, 7 is a memory for linear predictive synthesis that stores past synthetic speech information used when the linear predictive synthesis filter 6 generates synthetic speech, and the above components are conventional speech decoding units. Configuration Is the elementary.

Reference numeral 12 denotes a voice detecting means for detecting whether it is a voice frame or a non-voice frame from the synthesized voice information, and 8b corresponds to the detection result of the voice detecting means 12, and the adaptive codebook immediately before the frame error. A sound source information holding means for holding information, 9d corresponds to the detection result of the voice detecting means 12, and in the frame recovered from the frame error, the adaptive code book is held by the adaptive code book information held by the sound source information holding means 8b. It is a sound source information calculation means for updating.

While the frame error continues, the speech decoding apparatus carries out the decoding process according to the configuration of FIG.
Therefore, the feature of the configuration of FIG. 12 is that the voice detection means 12 detects whether the frame is a voice frame or a non-voice frame in the frame recovered from the frame error, and corresponds to the detection result of the voice detection means 12. Then, the excitation information holding unit 8b holds the adaptive codebook information of the frame immediately before the frame error, and the excitation information calculation unit 9d updates the adaptive codebook with the adaptive codebook information held by the excitation information holding unit 8b. There is a point to do.

FIG. 13 is a flowchart showing the error correcting operation of the configuration of FIG. 12, which is an operation flowchart of the voice detecting means, the sound source information holding means and the sound source information calculating means.
Hereinafter, the error recovery operation of the configuration of FIG. 12 will be described with reference to the symbols of FIG. First, the voice detection means determines whether the voice frame is a non-voice frame when no frame error has occurred. Since the operation before the frame error is the same as that after the recovery from the frame error, the operation after the recovery from the frame error is similar. Only the operation is shown in the flowchart.

S101. It is monitoring whether or not the frame error has recovered. S102. In step S101, if the frame error has not recovered (No), the frame power E of the synthesized speech is calculated. S103. Frame power E calculated in step S102
Is greater than a predetermined threshold value.

S104. When it is determined in step S103 that the frame power E is larger than a predetermined threshold value (Yes
In s), a determination that the frame recovered from the frame error is a voice frame is output, supplied to the sound source information holding means and the sound source information calculating means, and the processing is ended. S105. When it is determined in step S103 that the frame power E is smaller than the predetermined threshold value (No), the average linear prediction coefficient LPC _A and the average pitch P _A are calculated, and the stored average linear prediction coefficient is calculated. Update the average pitch.

S106. A determination that the frame recovered from the frame error is a non-voice frame is output and supplied to the sound source information holding means and the sound source information calculating means, and the processing is ended. S107. When the frame error is recovered in step S101 (Yes), the difference between the linear prediction coefficient LPC and the average linear prediction coefficient LPC _A is smaller than a predetermined threshold value, or the pitch P and the average pitch P are set. _It is determined whether the difference of _A is smaller than a predetermined threshold value.

If the difference between the linear prediction coefficient LPC and the average linear prediction coefficient LPC _A is smaller than a predetermined threshold value, or if the difference between the pitch P and the average pitch P _A is smaller than the predetermined threshold value (Yes). , Jump to step S106.
On the other hand, the linear prediction coefficient LPC and the average linear prediction coefficient LPC
_When the difference between _A and the pitch P is larger than a predetermined threshold and the difference between the pitch P and the average pitch P _A is larger than a predetermined threshold (N
In step o), the process jumps to step S104.

Here, in step S105, the average linear
Prediction coefficient LPC_AAnd the average pitch P _ACalculate and store
Updating the mean linear prediction coefficient and mean pitch
Is a preparation for the determination in step S107.
It Even before a frame error occurs, the voice detection means
The operations of steps S102 to S107 are performed.
The frame immediately before the frame error is a voice frame.
It is stored by judging whether it is a non-voice frame.

Next, the operation of the sound source information holding means is as follows. S111. It is monitoring whether a frame error has occurred. S112. If no frame error occurs in step S111 (No), the stored information of the adaptive codebook is held and the process jumps to step S111. On the other hand, if a frame error occurs in step S111 (Yes), the process proceeds to step S113.

That is, the excitation information storage means holds the storage information of the adaptive codebook immediately before the frame error occurs. S113. It is determined whether the determination made by the voice detecting means immediately before the frame error is the voice determination. S114. In the case of voice determination in step S113 (Yes), the adaptive codebook information of the voice frame is replaced with the adaptive codebook information held in step S112, and the process ends.

S115. If the non-voice determination is made in step S113 (No), the adaptive codebook information of the non-voice frame is replaced with the adaptive codebook information held in step S112, and the process ends. Further, the operation of the sound source information calculation means is as follows. S121. It is determined whether or not the voice detection means outputs the voice determination after recovering from the frame error.

S122. When it is determined in step S121 that the voice determination is performed (Yes), the adaptive codebook information of the voice frame is acquired from the sound source information holding unit. S123. When it is determined in step S121 that the non-voice determination is made (No), the adaptive codebook information of the non-voice frame is acquired from the sound source information holding unit. S124. The adaptive codebook is updated with the adaptive codebook information acquired in either step S122 or step S123, and the process ends.

As described above, in the configuration of FIG. 12, the voice detecting means 12 detects whether the frame is a voice frame or a non-voice frame in the frame immediately before the frame error and the frame recovered from the frame error, The excitation information holding unit 8b holds the adaptive codebook information of the frame immediately before the frame error, and the adaptive codebook information held corresponding to the detection result of the voice detection unit 12 is the adaptive codebook information of the voice frame or the non-voice. Since the adaptive codebook information of the frame is updated by the excitation information calculation means 9d as the adaptive codebook information of the frame, the adaptive codebook information is updated by the adaptive codebook information held by the excitation information holding means 8b. Adaptive codebook information immediately before the error,
Continuity can be maintained including voice frames and non-voice frames. Therefore, it is possible to reduce the decoding error in the subsequent decoding process and reduce the spread of the decoding error in the decoding process after recovering from the frame error.

The configuration of FIG. 12 is obtained by adding a voice detecting means to the configuration of FIG. 3, and considering that the configuration of FIG. 1 and the configuration of FIG. 3 are essentially the same, the configuration of FIG. It is understood that the voice detection means is added to the configuration of FIG. FIG. 14 shows an eighth embodiment of the present invention. In FIG. 14, 1 is an adaptive codebook that stores excitation information having periodicity after removing speech correlation, 2 is a fixed codebook that stores excitation information of noise components after removing speech correlation, 3 Is an adaptive codebook gain multiplier that multiplies the excitation information read from the adaptive codebook 1 by the adaptive codebook gain, 4 is a fixed codebook gain multiplier that multiplies the excitation information read from the fixed codebook by the fixed codebook gain, and 5 is An adder that adds the outputs of the adaptive codebook gain multiplier 3 and the fixed codebook gain multiplier 4 to generate excitation information, and 6 receives the excitation information output from the adder 5 and the received linear prediction coefficient to synthesize the excitation information. A linear predictive synthesis filter for outputting speech, 7 is a memory for linear predictive synthesis that stores past synthetic speech information used when the linear predictive synthesis filter 6 generates synthetic speech, and the above components are conventional speech decoding units. Configuration Is the elementary.

Reference numeral 12 denotes a voice detecting means for detecting whether it is a voice frame or a non-voice frame from the synthesized voice information, and 8c corresponds to the detection result of the voice detecting means 12, and the synthesized voice information immediately before the frame error. The sound source information holding unit 9e holds the synthesized voice information held by the sound source information holding unit 8c and the frame restored from the frame error in the frame recovered from the frame error corresponding to the detection result of the voice detection unit 12. The excitation information calculation means for generating adaptive codebook information from the linear prediction coefficient and updating the adaptive codebook, and the voice detection means 12, the excitation information holding means 8c, and the excitation information calculation means 9e constitute error correction means. . Note that, while the frame error continues, the voice decoding device performs the decoding process according to the configuration of FIG. Therefore, the feature of the configuration of FIG. 14 is that in the frame recovered from the frame error, the voice detection means 12 detects whether the frame is a voice frame or a non-voice frame,
Corresponding to the detection result of the voice detection means 12, the sound source information holding means 8c holds the synthesized speech information immediately before the frame error, and the sound source information calculation means 9e holds the synthesized speech information held by the sound source information holding means 8c. And adaptive codebook information is generated from the linear prediction coefficient and the adaptive codebook is updated.

FIG. 15 is a flowchart showing the error correction operation of the configuration of FIG. 14, which is an operation flowchart of the voice detecting means, the sound source information holding means, and the sound source information calculating means.
Hereinafter, the error recovery operation of the configuration of FIG. 14 will be described with reference to the reference numerals of FIG. First, the voice detection means determines whether the voice frame is a non-voice frame when no frame error has occurred. Since the operation before the frame error is the same as that after the recovery from the frame error, the operation after the recovery from the frame error is similar. Only the operation is shown in the flowchart.

S101. It is monitoring whether or not the frame error has recovered. S102. In step S101, if the frame error has not recovered (No), the frame power E of the synthesized speech is calculated. S103. Frame power E calculated in step S102
Is greater than a predetermined threshold value.

S104. When it is determined in step S103 that the frame power E is larger than a predetermined threshold value (Yes
In s), a determination that the frame recovered from the frame error is a voice frame is output, supplied to the sound source information holding means and the sound source information calculating means, and the processing is ended. S105. When it is determined in step S103 that the frame power E is smaller than the predetermined threshold value (No), the average linear prediction coefficient LPC _A and the average pitch P _A are calculated, and the stored average linear prediction coefficient is calculated. Update the average pitch.

S106. A determination that the frame recovered from the frame error is a non-voice frame is output and supplied to the sound source information holding means and the sound source information calculating means, and the processing is ended. S107. When the frame error is recovered in step S101 (Yes), the difference between the linear prediction coefficient LPC and the average linear prediction coefficient LPC _A is smaller than a predetermined threshold value, or the pitch P and the average pitch P are set. _It is determined whether the difference of _A is smaller than a predetermined threshold value.

When the difference between the linear prediction coefficient LPC and the average linear prediction coefficient LPC _A is smaller than a predetermined threshold value or when the difference between the pitch P and the average pitch P _A is smaller than a predetermined threshold value (Yes). , Jump to step S106.
On the other hand, the linear prediction coefficient LPC and the average linear prediction coefficient LPC
_When the difference between _A and the pitch P is larger than a predetermined threshold and the difference between the pitch P and the average pitch P _A is larger than a predetermined threshold (N
In step o), the process jumps to step S104.

Here, in step S105, the average linear
Prediction coefficient LPC_AAnd the average pitch P _ACalculate and store
Updating the mean linear prediction coefficient and mean pitch
Is a preparation for the determination in step S107.
It Even before a frame error occurs, the voice detection means
The operations of steps S102 to S107 are performed.
The frame immediately before the frame error is a voice frame.
It is stored by judging whether it is a non-voice frame.

Next, the operation of the sound source information holding means 8c is as follows. S131. It is monitoring whether a frame error has occurred. S132. If no frame error occurs in step S131 (No), the synthesized voice information is held and the process jumps to step S131. On the other hand, if a frame error occurs in step S131 (Yes), the process proceeds to step S133.

That is, the excitation information storage means holds the storage information of the adaptive codebook immediately before the frame error occurs. S133. It is determined whether the determination made by the voice detecting means immediately before the frame error is the voice determination. S134. In step S133, in the case of the voice determination (Yes), the synthetic voice information of the voice frame is replaced with the retained synthetic voice information, and the process ends.

S135. In step S133, if the non-voice determination is made (No), the voice information of the non-voice frame is replaced with the held synthetic voice information, and the process ends. Further, the operation of the sound source information calculation means is as follows. S141. It is determined whether or not the voice detection means outputs the voice determination after recovering from the frame error.

S142. When it is determined in step S141 that the voice determination is performed (Yes), the synthesized voice information of the voice frame is acquired from the sound source information holding unit. S143. When it is determined in step S141 that the non-voice frame is determined (No), the synthesized voice information of the non-voice frame is acquired from the sound source information holding unit. S144. The synthesized speech information and the linear prediction coefficient acquired in either step S142 or step S143 are supplied to the linear prediction analysis filter to calculate the sound source information.

S145. The adaptive codebook is updated with the sound source information calculated in step S144, and the process ends. As described above, in the configuration of FIG. 14, the voice detecting means 12 is
In the frame immediately before the frame error and the frame recovered from the frame error, it is detected whether the frame is a voice frame or a non-voice frame, and the sound source information holding unit 8c holds the synthesized voice information of the frame immediately before the frame error, The adaptive codebook information held in correspondence with the detection result of the voice detection means 12 is used as the synthesized voice information of the voice frame or the synthesized voice information of the non-voice frame, and the sound source information calculation means 9e holds the sound source information holding means 8c. Since the adaptive codebook is updated by calculating the excitation information by the synthesized speech information and the linear prediction coefficient, the adaptive codebook information in the frame recovered from the frame error is the adaptive codebook information immediately before the frame error and the speech frame. It is possible to maintain continuity including non-voice frames. Therefore, it is possible to reduce the decoding error in the subsequent decoding process and reduce the spread of the decoding error in the decoding process after recovering from the frame error.

It can be seen that the configuration of FIG. 14 is obtained by adding voice detection means to the configuration of FIG. FIG.
Is a ninth embodiment of the present invention. In FIG. 16, 1 is an adaptive codebook that stores excitation information having periodicity after removing speech correlation, 2 is a fixed codebook that stores excitation information of noise components after removing speech correlation, 3 Is an adaptive codebook gain multiplier that multiplies the excitation information read from the adaptive codebook 1 by the adaptive codebook gain, 4 is a fixed codebook gain multiplier that multiplies the excitation information read from the fixed codebook by the fixed codebook gain, and 5 is An adder for adding the outputs of the adaptive codebook gain multiplier 3 and the fixed codebook gain multiplier 4 to generate excitation information,
Reference numeral 6 denotes a linear predictive synthesis filter that receives the sound source information output from the adder 5 and the received linear prediction coefficient and outputs synthetic speech. Reference numeral 7 denotes a past synthesis used when the linear predictive synthesis filter 6 generates synthetic speech. This is a memory for linear prediction synthesis that stores speech information, and the above constituent elements are constituent elements of a conventional speech decoding apparatus.

Reference numeral 12 denotes a voice detecting means for detecting whether it is a voice frame or a non-voice frame from the synthesized voice information, and 11a calculates frame power from the synthesized voice information and corresponds to the detection result of the voice detecting means. Frame power calculating means for holding the frame power E ₁ immediately before the frame error, and 9f for the frame recovered from the frame error, corresponding to the detection result of the voice detecting means 12 from the frame power calculating means 11a. When E ₁ is received, synthetic speech information in the frame recovered from the frame error is generated, frame power E ₂ in the frame recovered from the frame error is calculated, and the information stored in the adaptive codebook is E ₁ / E in the sound source information calculating means for updating _doubled, voice detection means 12, the frame power calculation unit 11a and the sound source information calculating Constituting the error repair means by stages 9f.

While the frame error continues, the speech decoding apparatus carries out the decoding process according to the configuration of FIG.
Therefore, the feature of the configuration of FIG. 16 is that the voice detection unit 12 detects whether the frame is a voice frame or a non-voice frame in the frame recovered from the frame error, and corresponds to the detection result of the voice detection unit 12. Then, the frame power calculating means 11a calculates the frame power from the synthesized voice information, holds the frame power E ₁ immediately before the frame error corresponding to the detection result of the voice detecting means, and the sound source information calculating means 9f calculates the frame power from the frame error. In the returned frame,
Corresponding to the detection result of the voice detection means 12, the frame power E ₁ immediately before the frame error is calculated from the frame power calculation means 11a.
Is received, the synthetic speech information in the frame recovered from the frame error is generated, the frame power E ₂ in the frame recovered from the frame error is calculated, and the information stored in the adaptive codebook is multiplied by E ₁ / E _2. There is a point to update.

FIG. 17 is a flow chart showing the error recovery operation of the configuration of FIG. 16, which is the operation of the voice detection means, frame power calculation means and sound source information calculation means. After that, FIG.
The error correction operation of the configuration of FIG. 16 will be described with reference to the reference numeral 7. First, the voice detection means determines whether the voice frame is a non-voice frame when no frame error has occurred. Since the operation before the frame error is the same as that after the recovery from the frame error, the operation after the recovery from the frame error is similar. Only the operation is shown in the flowchart.

S101. It is monitoring whether or not the frame error has recovered. S102. In step S101, if the frame error has not recovered (No), the frame power E of the synthesized speech is calculated. S103. Frame power E calculated in step S102
Is greater than a predetermined threshold value.

S104. When it is determined in step S103 that the frame power E is larger than a predetermined threshold value (Yes
In s), a determination that the frame recovered from the frame error is a voice frame is output, supplied to the sound source information holding means and the sound source information calculating means, and the processing is ended. S105. When it is determined in step S103 that the frame power E is smaller than the predetermined threshold value (No), the average linear prediction coefficient LPC _A and the average pitch P _A are calculated, and the stored average linear prediction coefficient is calculated. Update the average pitch.

S106. A determination that the frame recovered from the frame error is a non-voice frame is output and supplied to the sound source information holding means and the sound source information calculating means, and the processing is ended. S107. When the frame error is recovered in step S101 (Yes), the difference between the linear prediction coefficient LPC and the average linear prediction coefficient LPC _A is smaller than a predetermined threshold value, or the pitch P and the average pitch P are set. _It is determined whether the difference of _A is smaller than a predetermined threshold value.

When the difference between the linear prediction coefficient LPC and the average linear prediction coefficient LPC _A is smaller than a predetermined threshold value, or when the difference between the pitch P and the average pitch P _A is smaller than the predetermined threshold value (Yes). , Jump to step S106.
On the other hand, the linear prediction coefficient LPC and the average linear prediction coefficient LPC
_When the difference between _A and the pitch P is larger than a predetermined threshold and the difference between the pitch P and the average pitch P _A is larger than a predetermined threshold (N
In step o), the process jumps to step S104.

Here, in step S105, the average linear
Prediction coefficient LPC_AAnd the average pitch P _ACalculate and store
Updating the mean linear prediction coefficient and mean pitch
Is a preparation for the determination in step S107.
It Even before a frame error occurs, the voice detection means
The operations of steps S102 to S107 are performed.
The frame immediately before the frame error is a voice frame.
It is determined whether it is a non-voice frame or not and is held.

Next, the operation of the frame power calculation means is as follows. S151. It is monitoring whether a frame error has occurred. S152. If no frame error occurs in step S151 (No), the synthesized voice information WY frame power is calculated, and the process jumps to step S151.

On the other hand, if a frame error occurs in step S151 (Yes), the process proceeds to step S153. That is, the frame power calculation means holds the frame power suitability immediately before the frame error occurs. S153. It is determined whether the determination made by the voice detecting means immediately before the frame error is the voice determination.

S154. In step S153, if the audio determination is made (Yes), the calculated frame power is held as the frame power of the audio frame, and the process ends. S155. In step S153, if the non-voice determination is made (No), the held frame power is held as the frame power of the non-voice frame, and the process ends.

Furthermore, the operation of the sound source information calculation means is as follows. S161. It is determined whether or not the voice detection means outputs the voice in the frame recovered from the frame error is the voice determination. S162. When it is determined in step S161 that the audio determination is performed (Yes), the frame power E ₁ of the audio frame is acquired from the frame power calculation means.

S163. When it is determined in step S161 that the non-voice determination is made (No), the frame power E ₂ of the non-voice frame is acquired from the frame power calculation means. S164. Synthetic speech is generated by the frame recovered from the frame error, and the frame power E ₂ is generated from the generated synthetic speech.
To calculate.

S165. The ratio between the frame power E ₁ obtained in either step S162 or step S163 and the frame power E ₂ calculated in step S164 is calculated, the adaptive codebook is updated to E ₁ / E ₂ times, and the process ends. To do. As described above, in the configuration of FIG. 16, the voice detecting means 12 is
In the frame immediately before the frame error and the frame recovered from the frame error, it is detected whether the frame is a voice frame or a non-voice frame, and the frame power calculation means 11a calculates the frame power immediately before the error and the voice detection means 12 The sound source information calculating means 9f obtains the frame power E ₁ calculated by the frame power calculating means 11a, and also calculates the frame power E ₂ in the return frame from the frame error, which is adapted to be held. Since the codebook is updated to E ₁ / E ₂ times, continuity can be provided before and after the frame error, including the information stored in the adaptive codebook and whether it is a voice frame or a non-voice frame. Therefore, it is possible to reduce the decoding error in the subsequent decoding process and reduce the spread of the decoding error in the decoding process after recovering from the frame error.

It can be seen that the configuration of FIG. 16 is obtained by adding voice detection means to the configuration of FIG. Finally, from the above, the basic technique of the present invention is as shown in FIGS.
It can be seen that it is the one shown in FIGS. 5, 8 and 10.
Summarizing these, the basic technique of the present invention is [1] receiving encoded information generated by linear prediction analysis,
A speech decoding apparatus that outputs a synthesized speech from the encoded information and the content stored in an internal memory that feeds back and stores speech information generated in the past, is stored in an adaptive codebook that constitutes the speech decoding apparatus. The excitation code is calculated using the excitation information holding means that holds the information immediately before the frame error and the information of the adaptive codebook that is held in the excitation information holding means, and the adaptive code is calculated in the frame recovered from the frame error. A speech decoding apparatus provided with an error correction means including a sound source calculation means for updating a book.

[2] Receive the coded information generated by the linear prediction analysis, and output the synthesized voice from the coded information and the content stored in the internal memory that feeds back and stores the voice information generated in the past. In a speech decoding apparatus, a predictive synthesis memory for storing synthetic speech information synthesized by a predictive synthesis filter forming the speech decoding apparatus holds a predictive synthesis memory for retaining synthetic speech information stored immediately before a frame error. And a frame recovered from a frame error, the storage information held by the prediction synthesis memory holding means is given to the prediction analysis filter to generate synthetic speech information, and the prediction synthesis memory is stored in the predicted synthesis memory according to the generated synthetic speech information. A speech decoding apparatus provided with an error correction means comprising a sound source information calculation means for updating.

[3] Receive the encoded information generated by the linear prediction analysis, and output the synthesized voice from the encoded information and the content stored in the internal memory that feeds back and stores the voice information generated in the past. In the speech decoding device, a sound source information holding unit that holds the synthesized sound information immediately before the frame error, a synthesized sound information held by the sound source information holding unit, and a frame that is restored from the frame error A speech decoding apparatus comprising: an error recovery means, which comprises adaptive codebook information generated from a linear prediction coefficient, and excitation information calculation means for updating the adaptive codebook.

[4] Receive the coded information generated by the linear prediction analysis, and output the synthesized voice from the coded information and the content stored in the internal memory that feeds back and holds the voice information generated in the past. In the speech decoding device, frame power calculation means for calculating frame power from the synthesized speech information and holding frame power E ₁ immediately before the frame error, and for a frame recovered from the frame error, the frame power calculation means for calculating the frame error It receives the immediately preceding frame power E ₁ , generates synthetic speech information in the frame recovered from the frame error, calculates the frame power E ₂ in the frame recovered from the frame error, and calculates the information stored in the adaptive codebook. A speech decoding apparatus provided with an error correction means comprising a sound source information calculation means for updating E ₁ / E ₂ times. It can be said that

That is, either the adaptive codebook is updated or the linear prediction synthesis memory is updated in the frame recovered from the frame error, and further, the adaptive codebook or the linear prediction synthesis memory is updated. Either the coded information immediately before the frame error is used for, or the coded information before and after the frame error is used. Therefore, the most essential thing of the present invention is that "the information stored in the internal memory that receives the coded information generated by the linear prediction analysis and feeds back the coded information and the voice information generated in the past is held. In a speech decoding apparatus for outputting synthetic speech from and, in a frame returned from a frame error state where normal coding information cannot be received to a state where normal coding information can be received, at least the internal memory immediately before the frame error The speech decoding apparatus is characterized by comprising an error repairing means for updating the contents held in the internal memory by using the holding information of.

Further, the speech decoding apparatus described above updates the content held in the adaptive codebook or the predictive synthesis memory, which is the internal memory, in the frame recovered from the frame error by the holding information immediately before the frame error. When viewed as the entire speech decoding system, for example, a post filter arranged in a subsequent stage of the prediction synthesis memory also has an internal memory for holding past information. Then, it is possible to update the contents held in the internal memory of the post filter with the holding information immediately before the frame error and perform the subsequent decoding processing. Therefore, the speech decoding apparatus of the present invention should naturally include the above technique.

(Supplementary Note 1) The encoded information generated by the linear prediction analysis is received, and the synthesized speech is output from the encoded information and the information held in the internal memory for feeding back and holding the speech information generated in the past. In the audio decoding device, the information held in the internal memory is used at least immediately before the frame error in the frame returned from the state of the frame error where the normal encoded information cannot be received to the state where the normal encoded information can be received. A speech decoding apparatus comprising error correction means for updating the contents held in the internal memory.

(Supplementary Note 2) In the speech decoding apparatus according to Supplementary Note 1, the error repairing means holds information immediately before a frame error stored in an adaptive codebook constituting the speech decoding apparatus. And a sound source calculation means for calculating a sound source using the information of the adaptive codebook held in the sound source information holding means and updating the adaptive codebook in a frame recovered from a frame error. A speech decoding device characterized by the above.

(Supplementary Note 3) In the speech decoding apparatus according to Supplementary Note 1, the error correcting means includes a predictive synthesis memory for storing synthetic speech information synthesized by the predictive synthesis filter forming the speech decoding apparatus. , A prediction synthesis memory holding means for holding the synthesized speech information stored immediately before the frame error, and a storage information held by the prediction synthesis memory holding means for the frame restored from the frame error to the prediction analysis filter. And a sound source information calculating unit for updating the predictive synthesis memory with the generated synthesized speech information.

(Supplementary Note 4) In the speech decoding apparatus according to Supplementary Note 1, the error repairing means includes a sound source information retaining means for retaining synthesized speech information immediately before a frame error and a frame recovered from the frame error. The present invention further comprises excitation information calculation means for generating adaptive codebook information from the synthesized speech information held by the excitation information holding means and the linear prediction coefficient of the frame recovered from the frame error and updating the adaptive codebook. And a voice decoding device.

(Supplementary Note 5) In the speech decoding apparatus according to Supplementary Note 1, the error correcting means calculates a frame power from the synthesized speech information and holds the frame power E ₁ immediately before the frame error. A frame recovered from a frame error by receiving the frame power E ₁ immediately before the frame error from the frame power calculation unit in the frame recovered from the frame error and generating synthetic speech information in the frame recovered from the frame error. And a sound source information calculating unit for calculating the frame power E ₂ in Eq. ₁ and updating the information stored in the adaptive codebook by E ₁ / E ₂ times.

(Supplementary Note 6) The speech decoding apparatus according to supplementary note 1, wherein the error recovery means comprises the error recovery means described in supplementary note 2 and the error recovery means described in supplementary note 3. And a voice decoding device.

(Supplementary Note 7) In the speech decoding apparatus according to Supplementary Note 1, the error correction means may use the synthesized speech information to convert
Voice detection means for detecting whether the frame immediately before the frame error and the frame recovered from the frame error are voice frames or non-voice frames, and adaptation just before the frame error corresponding to the detection result of the voice detection means. An excitation code information holding means for holding codebook information and an adaptive codebook according to the detection result of the voice detection means, in the frame recovered from the frame error, by the adaptive codebook information held by the excitation code information holding means. And a sound source information calculating means for updating the speech decoding apparatus.

(Supplementary Note 8) In the speech decoding apparatus according to Supplementary Note 1, the error recovery means may use the synthesized speech information,
A voice detection unit for detecting whether the frame immediately before the frame error and the frame recovered from the frame error are a voice frame or a non-voice frame, and a synthesized voice immediately before the frame error corresponding to the detection result of the voice detection unit. Sound source information holding means for holding information, and in the frame recovered from the frame error corresponding to the detection result of the sound detecting means, the synthesized speech information held by the sound source information holding means and the frame recovered from the frame error. A speech decoding apparatus, comprising: excitation code information calculating means for generating adaptive codebook information from a linear prediction coefficient and updating the adaptive codebook.

(Supplementary Note 9) In the speech decoding apparatus according to Supplementary Note 1, the error correcting means is configured to convert the synthesized speech information from the synthesized speech information.
A voice detection unit that detects whether the frame immediately before the frame error and the frame recovered from the frame error is a voice frame or a non-voice frame, and frame power from the synthesized voice information corresponding to the detection result of the voice detection unit. And a frame power calculating unit that holds the frame power E ₁ immediately before the frame error and a frame error recovered from the frame error corresponding to the detection result of the voice detecting unit. It receives the immediately preceding frame power E ₁ , generates synthetic speech information in the frame recovered from the frame error, calculates the frame power E ₂ in the frame recovered from the frame error, and calculates the information stored in the adaptive codebook. A speech decoding apparatus comprising: a sound source information calculating unit that updates E ₁ / E ₂ times.

[0133]

According to the first aspect of the invention, at least in a frame in which the error recovery means returns from a frame error state where normal encoding information cannot be received to a state where normal encoding information can be received, Since the held information in the internal memory is updated using the held information in the internal memory immediately before the frame error, the held information in the internal memory can approximate the correct held information in the frame recovered from the frame error, It is possible to reduce the spread of decoding error in the subsequent decoding processing.

According to the second aspect of the invention, the error repairing means stores excitation information holding means for holding information immediately before the frame error, which is stored in the adaptive codebook constituting the speech decoding apparatus, The adaptive codebook is provided by including an excitation information calculation means for calculating an excitation using the information of the adaptive codebook held in the excitation information holding means and updating the adaptive codebook in a frame recovered from a frame error. It is possible to approximate the correct adaptive codebook information in the frame in which the storage information of 1 is recovered from the frame error, and it is possible to reduce the spread of decoding error in the subsequent decoding processing.

According to the third invention, the error repairing means stores the synthesized speech information synthesized by the predictive synthesis filter forming the speech decoding device, and stores it in the predictive synthesis memory immediately before the frame error. A predictive synthesis memory holding means for holding the synthesized speech information, and in a frame recovered from a frame error, the stored information held by the predictive synthesis memory holding means is given to the predictive synthesis filter to generate synthesized speech information. Then, since the prediction synthesis memory is updated with the generated synthesized speech information, the storage information of the prediction synthesis memory can approximate the correct storage information in the frame recovered from the frame error. The spread of error can be reduced.

According to the fourth aspect of the invention, the error recovery means holds the sound source information holding means for holding the synthesized speech information immediately before the frame error and the sound source information holding means for holding the frame recovered from the frame error. Since the adaptive codebook information is generated from the synthesized speech information and the linear prediction coefficient of the frame recovered from the frame error, and the adaptive codebook is updated, the storage information of the adaptive codebook is stored. The correct stored information in the frame recovered from the frame error can be approximated, and the spread of decoding error in the subsequent decoding processing can be reduced.

Further, according to the fifth invention, the error repairing means calculates the frame power from the synthesized speech information and holds the frame power E ₁ immediately before the frame error, and the frame error calculating means. In the restored frame, the frame power E ₁ immediately before the frame error is received from the frame power calculation means, synthetic speech information in the frame recovered from the frame error is generated, and the frame power E ₂ in the frame recovered from the frame error is calculated. And the excitation information calculation means for updating the information stored in the adaptive codebook by E ₁ / E ₂ times, the correct stored information in the frame in which the stored information in the adaptive codebook is recovered from the frame error is obtained. This can be approximated, and the spread of decoding error in the subsequent decoding processing can be reduced.

[Brief description of drawings]

FIG. 1 is a first embodiment of the present invention.

FIG. 2 is a flowchart showing an error recovery operation of the configuration of FIG.

FIG. 3 is a second embodiment of the present invention.

FIG. 4 is a flowchart showing an error correction operation of the configuration of FIG.

FIG. 5 is a third embodiment of the present invention.

FIG. 6 is a flowchart showing an error recovery operation of the configuration of FIG.

FIG. 7 is a fourth embodiment of the present invention.

FIG. 8 is a fifth embodiment of the present invention.

9 is a flowchart showing an error correction operation of the configuration of FIG.

FIG. 10 shows a sixth embodiment of the present invention.

11 is a flowchart showing an error recovery operation of the configuration of FIG.

FIG. 12 is a seventh embodiment of the present invention.

FIG. 13 is a flowchart showing an error recovery operation of the configuration of FIG.

FIG. 14 is an eighth embodiment of the present invention.

FIG. 15 is a flowchart showing an error correction operation of the configuration of FIG.

FIG. 16 is a ninth embodiment of the present invention.

FIG. 17 is a flowchart showing an error recovery operation of the configuration of FIG.

FIG. 18 shows a configuration of a conventional speech decoding device (1).

FIG. 19 shows a configuration of a conventional speech decoding device (2).

[Explanation of symbols]

1 Adaptive codebook 2 fixed codebook 3 Adaptive codebook gain multiplier 4 Fixed codebook gain multiplier 5 adder 6 Linear prediction synthesis filter Memory for linear prediction synthesis 8 Sound source information holding means 8a Sound source information holding means 8b Sound source information holding means 8c Sound source information holding means 9 Sound source information calculation means 9a Sound source information calculation means 9b Sound source information calculation means 9c Sound source information calculation means 9d Sound source information calculation means 9e Sound source information calculation means 9f Sound source information calculation means 10 Memory holding means for linear prediction synthesis 11 Frame power calculation means 11a Frame power calculation means 12 Voice detection means

   ─────────────────────────────────────────────────── ─── Continued front page    (72) Inventor Shinji Naka             3-22-8, Hakata Station, Hakata-ku, Fukuoka City, Fukuoka Prefecture             Issue Fujitsu Kyushu Digital Technology Co., Ltd.             Inside the company F-term (reference) 5D045 DA20                 5K041 AA02 BB08 CC01 EE12 EE22                       GG01 GG13 JJ24

Claims (5)

[Claims] 1. A voice that receives coded information generated by linear prediction analysis, and outputs a synthesized voice from the coded information and information held in an internal memory that feeds back and holds voice information generated in the past. In the decoding device, at least in the frame returned from the state of the frame error where normal encoding information cannot be received to the state where normal encoding information can be received, by using the information held in the internal memory immediately before the frame error, An audio decoding device comprising an error recovery means for updating the contents held in an internal memory. 2. The speech decoding apparatus according to claim 1, wherein the error repairing means holds information immediately before a frame error stored in an adaptive codebook constituting the speech decoding apparatus. Excitation source information holding means and excitation source calculation means for obtaining excitation information using the information of the adaptive codebook held in the excitation information holding means and updating the adaptive codebook in a frame recovered from a frame error A voice decoding device characterized by. 3. The speech decoding apparatus according to claim 1, wherein the error correction means includes a predictive synthesis memory for storing synthetic speech information synthesized by a predictive synthesis filter forming the speech decoding apparatus. A prediction synthesis memory holding unit for holding the synthesized speech information stored immediately before the frame error is provided, and in the frame recovered from the frame error, the storage information held by the prediction synthesis memory holding unit is given to the prediction synthesis filter. A speech decoding apparatus, wherein synthetic speech information is generated in accordance with the generated synthetic speech information, and the prediction synthesis memory is updated with the generated synthetic speech information. 4. The speech decoding apparatus according to claim 1, wherein the error correction means includes a sound source information holding means for holding synthetic speech information immediately before a frame error, and a frame recovered from the frame error. The present invention is characterized by further comprising excitation information calculation means for generating adaptive codebook information from the synthesized speech information held by the excitation information holding means and the linear prediction coefficient of the frame recovered from the frame error, and updating the adaptive codebook. Audio decoding device. 5. The speech decoding apparatus according to claim 1, wherein the error correction means calculates a frame power from the synthesized speech information and holds a frame power E ₁ immediately before a frame error. And the frame power E immediately before the frame error from the frame power calculation means in the frame recovered from the frame error.
_{When 1} is received, synthetic speech information in the frame recovered from the frame error is generated, frame power E ₂ in the frame recovered from the frame error is calculated, and the information stored in the adaptive codebook is E ₁ / E ₂ A speech decoding apparatus comprising: a sound source information calculation unit that doubles the time.