JP2002082696A - Inspecting method and inspecting device for voice coding and decoding equipment - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain a voice inspecting method capable of performing the voice inspection of a long time in real time as to the output voices of voice coding and decoding equipment which is made to be the object of the inspection. SOLUTION: In this voice inspecting method, the input voices and the output voices of the voice coding and decoding equipment which produces the output voices by coding and decoding the input voices are sampled respectively and estimated voices which are obtained by estimating voices of the output side of the equipment with the operation between the sampled input voices and the filter coefficient of a digital filter whose filter coefficient is controlled by an adaptive algorithm are produced and differences among the estimated voices and the sampled output voices are calculated and the differences are outputted as an error signal and also the error signal is inputted to the adaptive algorithm to update the filter coefficient of the digital filter by following up the characteristic of the voice coding and decoding equipment.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an inspection method and an inspection apparatus for an audio encoding / decoding device for encoding, decoding, and outputting an input audio.

[0002]

2. Description of the Related Art As a conventional method for testing speech encoding / decoding equipment, Japanese Patent Laid-Open No. 7-84596 discloses a method for evaluating the quality of encoded speech. This quality evaluation method is as shown in the flowchart of FIG. 7. In SB1, the sound obtained by encoding the original audio data by the encoding / decoding device to be measured is combined into one frame every 20 ms, for example. The audio data and the decoded and output audio data are subjected to fast Fourier transform. In SB2, the real part and the imaginary part of the short-time audio spectrum are separated by the power spectrum calculation process, and the short-time is calculated by sum of squares. A time power spectrum is output, and the short-time power spectrum is converted from the frequency axis to the Bark frequency.

In SB3, a multiplication (hereinafter referred to as convolution) of the short-time power spectrum and a filter coefficient of a critical band filter shown in FIG. 8 stored in advance in a filter coefficient storage unit is performed. A plurality of critical band power spectra are obtained by convolving the set and the short-time power spectrum. At SB4, a pre-emphasis process and a calculation process of the Bark spectrum are performed. At SB5, a masking amount calculation process is performed for each frame to obtain a Bark spectrum at the current frame. At SB6, a distortion calculation process is performed. Is performed.

According to this method, it is shown that the quality of a coded speech can be estimated with high accuracy, and the effect of reducing the amount of calculation can be obtained.

[0005]

In the above-described conventional method for evaluating the quality of a speech encoding / decoding device, an input original speech (hereinafter referred to as an input speech) and an input speech are encoded and decoded. This is a method of performing a fast Fourier transform on a voice (hereinafter, referred to as an output voice) and extracting and comparing and evaluating the feature amounts of an input voice and an output voice in each frequency domain. Compare the amount of features in the frequency domain when there is a problem that a long time voice test cannot be performed in real time due to the time required for conversion, and when the time required for encoding and decoding of the device under test varies. However, this method has a problem in that it is not possible to cope with a time variation required for encoding / decoding and to perform an accurate inspection.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problem, and provides a voice inspection method and a voice inspection apparatus capable of performing real-time and long-term voice inspection of an output voice of a target encoding / decoding device. It is intended to do so.

[0007]

According to a first aspect of the present invention, there is provided an audio encoding / decoding device inspection method for encoding / decoding an input audio to generate an output audio. An input voice and an output voice are sampled, and an estimated voice is generated by estimating a voice on the output side by calculating a sampled input voice and a filter coefficient of a digital filter whose filter coefficient is controlled by an adaptive algorithm. The difference between the estimated speech and the sampled output speech is calculated and output as an error signal.
This is a method of updating the filter coefficient of the digital filter following the characteristic fluctuation of the decoding device.

According to a second aspect of the present invention, a speech encoding / decoding device inspection apparatus samples an input speech of a speech encoding / decoding device that encodes / decodes an input speech to create an output speech. An input audio detection unit, an output audio detection unit that samples output audio, and a digital filter in which a filter coefficient is controlled so as to follow a characteristic change of the audio encoding / decoding device by the sampled input audio and an adaptive algorithm. An estimated voice generating unit that generates an estimated voice by estimating the output-side voice based on the filter coefficient; and calculating a difference between the estimated voice and the output voice and outputting the calculated error as an error signal, and an adaptive algorithm of the estimated voice generating unit. It is composed of an input error signal generator and an error evaluator for evaluating the error signal.

According to a third aspect of the present invention, in the inspection apparatus for an audio encoding / decoding device, the error evaluator having the configuration according to the second aspect is provided with a voice abnormality detection level, and the error signal is set to the audio abnormality detection level. When the value exceeds the limit, an abnormal signal is output.

[0010] According to a fourth aspect of the present invention, in the inspection apparatus for audio encoding / decoding equipment, the error evaluator having the configuration according to the second aspect is provided with a speech abnormality detection level, and the error signal indicates the speech abnormality detection level. The configuration is such that the time exceeded is counted, and the magnitude of the error signal is evaluated and output.

According to a fifth aspect of the present invention, in the inspection apparatus for an audio encoding / decoding device, the audio abnormality detection level provided in the error estimating unit according to the third or fourth aspect is such that the level of the input audio is large. It is configured to be set according to the degree.

According to a sixth aspect of the present invention, there is provided a method for inspecting a speech encoding / decoding device, comprising the steps of: obtaining a filter coefficient by an adaptive algorithm of a digital filter when an error signal exceeds a predetermined level; This is a method of stopping the control for a certain time.

According to a seventh aspect of the present invention, in the inspection apparatus for a speech encoding / decoding device, the estimated speech creation unit having the configuration according to any one of the second to fifth aspects is configured such that when the error signal exceeds a predetermined level,
The control of the filter coefficient by the adaptive algorithm of the digital filter is stopped for a certain period of time.

According to an eighth aspect of the present invention, there is provided a method for inspecting an audio encoding / decoding device according to claim 1 or 6, wherein the time required for encoding / decoding of the audio encoding / decoding device. In this method, the estimated speech is created by varying the input speech and the time interval used for the operation of the digital filter in accordance with the variation of the input speech.

According to a ninth aspect of the present invention, there is provided an inspection apparatus for an audio encoding / decoding device. The estimated speech is created by varying the time interval used for the operation of the input speech and the digital filter in accordance with the time variation required for encoding / decoding.

[0016]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiment 1 FIG. 1 is a block diagram showing a configuration of an inspection apparatus for a speech encoding / decoding device according to Embodiment 1, and FIG. 2 is a flowchart of an inspection method by the inspection apparatus having the configuration shown in FIG. In the figure, reference numeral 1 denotes an audio encoding / decoding device to be inspected. Reference numeral 2 denotes an input voice detection unit that samples input voice, and reference numeral 3 denotes an output voice detection unit that samples output voice.

Reference numeral 4 denotes an estimated voice generation unit for generating an estimated voice by estimating the output voice from the sampled input voice. The adaptive digital filter 4a receives the input voice and the difference between the estimated voice and the output voice. An adaptive algorithm 4 for controlling the filter coefficients of the adaptive digital filter so as to follow the characteristic fluctuation of the speech encoding / decoding device 4
b. Reference numeral 5 denotes an error signal creation unit that calculates a difference between the estimated speech and the output speech, outputs the difference to the error evaluation unit, and inputs the difference to the adaptive algorithm 4b. Reference numeral 6 denotes an error evaluator for evaluating the error signal from the error signal generator 5.

Next, the inspection method according to the first embodiment will be described with reference to the flowchart of FIG. In step S1,
Input voice x (n) in input voice detection unit 2, output voice detection unit 3
Sample the output sound d (n). n
Is the sampling time. In step S2, an estimated sound y (n) obtained by performing a filter process in (Equation 1) in the estimated sound creation unit 4 and estimating the output-side sound from the input sound
Is calculated.

[0019]

(Equation 1)

Wi is the filter coefficient of the i-th digital filter, I is the number of taps of the filter, and id is the time delay.

Next, in step S3, the error signal generator 5
, The difference e (n) between the output speech d (n) and the estimated speech y (n)
d (n) −y (n) is calculated and output to the error evaluator 6 as the error signal e (n), and is also input to the adaptive algorithm 4 b of the estimated speech generator 4. In step S4, the error evaluator 6 outputs the error signal e of the speech encoding / decoding device.
The error signal e is evaluated by monitoring the waveform.

In step 5, the error signal e (n) is input from the error signal generator 5 and the difference e (n) between the output voice d (n) and the estimated voice y (n) is calculated by the adaptive algorithm 4b. The square mean value is calculated by (Equation 2), and the process returns to step S1 so that the difference e (n) is minimized, and the process is repeated to update the filter coefficient of the digital filter.

[0023]

(Equation 2)

Next, a description will be given of the state of each audio waveform when the audio encoder / decoder performs the audio inspection according to the flowchart of FIG. 2 in the configuration of FIG. FIG. 3 shows the state of each voice and error signal when a voice test is performed on a voice encoding / decoding device. (A) is an input voice x, (b) is an output voice d,
(C) is the estimated voice y, and (d) is the error signal e. Actually, the output audio d has a longer time for encoding / decoding in the encoding / decoding device than the input audio x, but FIG. 3 shows the start points of the time axes of the input audio x and the output audio y together. are doing.

In the example of the voice test shown in FIG.
During eight seconds, the error signal e is 0, indicating that it is normal. Error signal e during the time axis of 0.8 to 1.17 seconds
Appears, indicating that there is an abnormality. The actual output sound d is in a state where the sound is interrupted. In the audio inspection, the quality of the audio encoding / decoding device can be continuously inspected in real time by monitoring the amplitude of the error signal e.

The encoding / decoding device of the encoding / decoding device to be inspected
Even when the time required for decoding varies, the time variation is converted to the number of sample steps, and if it is in the range of id to id + I-1, the time variation can be handled by the calculations of (Equation 1) and (Equation 2). The output sound d can be estimated and an accurate inspection can be performed. That is, the values of id and I are set so that the time variation required for encoding / decoding of the inspection target device is in the range of id to id + I-1.

As described above, the input voice x and the output voice d are detected, and the output voice d is estimated from the input voice x using an adaptive digital filter.
Is compared in the time domain to obtain an audio encoding / decoding inspection apparatus capable of performing audio inspection continuously in real time for a long time.

Further, even when the time required for encoding / decoding of the encoding / decoding device to be inspected fluctuates, if the time variation required for encoding / decoding ranges from id to id + I-1, The estimated voice y corresponding to the time variation is estimated, and an accurate inspection can be performed.

Embodiment 2 In the configuration of FIG. 1 of the first embodiment, the error signal e is displayed as a waveform. However, in the configuration of the second embodiment, the audio abnormality detection level is set in the error evaluator 6 of the first embodiment. The configuration is as follows.

The error evaluator 6 outputs an abnormal signal when the error signal e generated by the error signal generator 5 exceeds the audio abnormality detection level, thereby providing a sound inspection corresponding to the required audio abnormality level. Can be implemented efficiently.

Embodiment 3 The second embodiment has a configuration in which the error evaluator 6 of the first embodiment is provided with a speech abnormality detection level, and outputs an abnormality signal to the error evaluator 6 when the error signal e exceeds the speech abnormality detection level. However, in the third embodiment, the configuration is such that the duration of the portion that is further exceeded is detected. FIG. 4 shows an error evaluator 6 according to the second embodiment.
Shows an example of the error signal e when the audio abnormality detection level is set in FIG.

FIG. 4 shows a case where the audio abnormality detection level is set to the level 0.5 of the error signal e, and the error evaluator 6 constantly monitors the amplitude of the error signal e and sets the error signal e. The number of sample steps exceeding the detected audio abnormality detection level, that is, the number of error signals e outside the audio abnormality detection level shown in FIG.
The duration of the error signal is evaluated by this count number,
An inspection can be performed in consideration of the time of the voice abnormality.

Embodiment 4 The fourth embodiment is configured such that the audio abnormality detection level set in the error evaluation unit 6 of the configuration of the second or third embodiment can be set stepwise according to the level of the input audio.

[0034] With this configuration, it is possible to carry out the inspection in accordance with the required evaluation level regardless of the fluctuation of the input audio level of the encoding / decoding equipment whose audio abnormality inspection level is inspected. An inspection device applicable to equipment can be configured.

Embodiment 5 FIG. In the configuration of the test apparatus of the speech encoding / decoding device in FIG. 1 according to the first embodiment, if the update of the filter coefficient by the adaptive algorithm is continued, the error signal converges, and the speech in the second or third embodiment There is a possibility that the error signal e outside the audio abnormality detection level when the abnormality detection level is exceeded becomes small, making it difficult to perform an accurate audio inspection. In the fifth embodiment, in addition to the configuration of the second or third embodiment in which the audio abnormality detection level is provided in the configuration of FIG. 1, a certain amount of time after the error signal e exceeds the audio abnormality detection level for the first time. It is configured to stop the control of the filter coefficient by the time and the adaptive algorithm.

As described above, by stopping the control of the filter coefficient by the adaptive algorithm for a certain period of time, it is possible to obtain an inspection apparatus capable of preventing a convergence of the error signal e and performing a voice test in which a voice abnormality is emphasized.

Embodiment 6 FIG. If the fluctuation of the time required for encoding / decoding of the encoding / decoding device to be inspected exceeds the number of filter taps I in terms of the number of sample steps, it becomes difficult to perform an accurate inspection. In this case, the problem can be solved by increasing the number of filter taps I. However, there is a problem that the amount of calculation for updating the filter coefficient of the filtering process increases, and the inspection cannot be performed in real time. Embodiment 6
In order to solve this problem, the time interval of a digital filter used when estimating the estimated speech y from the input speech x of the estimated speech creation unit 4 is varied according to the variation in the time required for encoding and decoding. Configuration.

Hereinafter, a specific method will be described. FIG.
5 is a flowchart of a speech inspection method in a case where a time interval used for calculating an input speech x and a filter coefficient of a digital filter is varied according to a variation in time required for encoding / decoding of a speech encoding / decoding device. The voice inspection device has the same configuration as that of FIG. In step S11, the input voice x (n) is output by the input voice detection unit 2 to the output voice d (n).
Are sampled in the output sound detection unit 3 respectively. n is the sampling time. Step S12
, The estimated voice creating unit 4 performs a filtering process using the following (Equation 3), (Equation 4), and (Equation 5) to obtain three estimated voices yJ, yJ + 1, and yJ-1.

[0039]

(Equation 3)

J is a variable indicating the amount of change in the filter operation time. The digital filter and the input voice used for the operation according to the change in the time required for the encoding / decoding processing from the initial filter operation time (J = 0). The time interval of x is varied. yJ is the estimated voice at the current filter operation time, and yJ + 1 is one sample step from the current operation time.
The estimated voice when advanced by one, yJ-1 is an estimated voice obtained by delaying the sample step by one from the current calculation time.

Next, at step 13, three error signals e, ie, eJ = d (n) -yJ eJ + 1 = d (n) -yJ + 1 eJ-1
= D (n) -yJ-1 and, in step 14, the one having the smallest absolute value among eJ, eJ + 1, and eJ-1 is taken as the true error signal e (n), and the filter operation time J To update.

In step 15, the magnitude of the error signal e of the error evaluator 6 is evaluated and a voice test is performed. In step 16, the calculated error signal e (n) is input to the adaptive algorithm 4b, and the output speech d (n) and the estimated speech y (n)
The mean square value of the difference e (n) is calculated by (Equation 2), and the process returns to step S11 to repeat the process so as to update the filter coefficient of the digital filter so that the difference e (n) is minimized. . The evaluation of the error signal e in the fifth embodiment is performed in the same manner as in the first to fourth embodiments.

Next, a description will be given of a digital filter used in the calculation of the fifth embodiment and a process of changing the time interval of the input voice. The initial value of the filter operation time J is 0,
The output voice d is assumed to be changed by a time delay id + D with respect to the input voice x, and the filter coefficients for estimating the output voice d from the input voice x are ideally WD = 1 and Wi = 0 (i ≠ D). It is a target. FIG. 6 shows the input voice x, the output voice d, and the filter coefficient in this case. In the figure, a filter coefficient for obtaining yJ and a calculation section of the input voice x,
Is a calculation section of the filter coefficient and input voice for obtaining yJ + 1, and is a calculation section of the filter coefficient and input voice x for obtaining yJ-1.

In the case of 0 ≦ D <I, an ideal filter coefficient can be realized in the operation section, so that there is no need to delay the operation section. When the time delay fluctuates and D ≧ I,
In the calculation section, ideal filter coefficients cannot be realized,
The filter coefficient can be realized only in the calculation section. Therefore, the error signal eJ + 1 becomes minimum, the filter operation time J is incremented by 1, and the operation section moves to. Similarly, when the time delay fluctuates and D <0, an ideal filter coefficient can be realized only in the calculation section, and thus the error signal eJ
−1 becomes the minimum, the filter operation time J is decremented, and the operation section moves to. As described above, the time interval between the digital filter used for the calculation and the input voice x moves according to the fluctuation of the time delay between the input voice x and the output voice y.

As described above, according to the sixth embodiment, even when the fluctuation range of the time required for the encoding / decoding device to be inspected is large, the digital filter used for the calculation and the time of the input voice x in accordance with the fluctuation. By varying the section, it is possible to inspect the quality of the voice in real time.

[0046]

According to the first aspect of the present invention, the speech encoding
The decoding method of the decoding device is to generate an estimated sound in which an output-side sound is estimated by calculating a sampled input sound and a filter coefficient of a digital filter whose filter coefficient is controlled by an adaptive algorithm, and generate the estimated sound and the output. This is a method of calculating the difference from speech, outputting it as an error signal, inputting it to an adaptive algorithm, and updating the filter coefficient of a digital filter following the characteristic fluctuation of speech encoding / decoding equipment. The decoding device can perform a long-term voice test in real time, and can perform a voice test in real time in response to the time variation required for encoding and decoding of the device to be tested.

According to a second aspect of the present invention, there is provided an inspection apparatus for an audio encoding / decoding device, comprising: an input audio detecting section for sampling an input audio; an output audio detecting section for sampling an output audio; An estimated speech creation unit that creates an estimated speech based on the filter coefficients of a digital filter whose filter coefficients are controlled so as to follow a characteristic change of the speech encoding / decoding device by an adaptive algorithm, and an estimated speech on the output side. A difference between the generated estimated voice and the output voice is calculated and output as an error signal, and an error signal generating unit that inputs the adaptive algorithm of the estimated voice generating unit and an error evaluating unit that evaluates the error signal are configured. As a result, the encoding / decoding device can perform a long-term voice test in real time, It is possible to perform voice test in real time corresponds to the time fluctuation.

According to a third aspect of the present invention, an inspection apparatus for an audio encoding / decoding device sets an audio abnormality detection level in the error evaluation section having the configuration of the second aspect, and the error signal exceeds the audio abnormality detection level. In this case, an abnormal signal is output when the abnormal condition occurs, so that an audio test can be efficiently performed in accordance with a required abnormal audio level.

According to a fourth aspect of the present invention, there is provided an inspection apparatus for an audio encoding / decoding device, wherein an error evaluation level is set in the error evaluator having the configuration of the second aspect, and the error signal indicates the audio abnormality detection level. Since the configuration is such that the time elapsed is counted and the magnitude of the error signal is evaluated, it is possible to perform an inspection in consideration of the duration of the error signal of the encoding / decoding device.

According to a fifth aspect of the present invention, in the inspection apparatus for an audio encoding / decoding apparatus, the audio abnormality detection level provided in the error evaluator having the configuration of the third or fourth aspect is such that the level of the input audio is large. Since it is configured so as to be set in accordance with the level, it is possible to carry out the inspection in accordance with the required evaluation level regardless of the fluctuation of the input audio level of the encoding / decoding device, and the inspection device applicable to a wide range of inspection target devices. Becomes

According to a sixth aspect of the present invention, in the method for inspecting a speech encoding / decoding device, when the error signal exceeds a predetermined level in the method of the first aspect, the filter coefficient of the digital filter is determined by the adaptive algorithm of the digital filter. This is a method in which the control is stopped for a certain period of time, and the inspection can be performed in which the convergence of the error signal is prevented and the abnormal voice is emphasized.

According to a seventh aspect of the present invention, in the inspection apparatus for an audio encoding / decoding device, the estimated audio creating section having the configuration according to any one of the second to fifth aspects is configured such that when the error signal exceeds a predetermined level, Since the control of the filter coefficient by the adaptive algorithm of the digital filter is stopped for a certain period of time, it is possible to prevent the convergence of the error signal and perform a test in which the speech abnormality is emphasized.

According to an eighth aspect of the present invention, there is provided a method for inspecting an audio encoding / decoding device according to the first or sixth aspect, wherein the time required for encoding / decoding of the audio encoding / decoding device is determined. This is a method of calculating the difference between the input voice and the output voice by changing the time interval used for the operation of the digital filter in accordance with the fluctuation. The time fluctuation of the coding / decoding of the voice coding / decoding device is large. Even in this case, a long-term voice test can be performed in real time.

According to a ninth aspect of the present invention, there is provided an inspection apparatus for an audio encoding / decoding device, wherein the estimated speech creating section having the configuration according to any one of the second to fifth and seventh aspects comprises the step of: The configuration is such that the estimated speech is estimated by varying the time interval used for the operation of the input speech and the digital filter according to the time variation required for encoding / decoding.
Even when the fluctuation of the decoding time is large, a long-time voice test can be performed in real time.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of an inspection device of a speech encoding / decoding device according to a first embodiment.

FIG. 2 is a flowchart of an inspection method by the inspection device having the configuration of FIG.

FIG. 3 is a diagram showing the status of each voice and error signal when a voice test is performed on a voice encoding / decoding device according to the flowchart of FIG. 2;

FIG. 4 is a diagram illustrating a state of an error signal when a speech abnormality detection level is provided in an error evaluation unit according to the second embodiment.

FIG. 5 is a flowchart of an audio inspection method according to a fifth embodiment when a time section of a digital filter is delayed according to encoding / decoding time.

FIG. 6 is a diagram illustrating a state of a speech waveform when a speech test is performed by delaying a time section of a digital filter in accordance with a change in time required for encoding and decoding.

FIG. 7 is a flowchart of a conventional quality evaluation method for an encoding / decoding device.

FIG. 8 is a diagram showing filter coefficients of a critical band filter used for filtering a critical band power spectrum.

[Explanation of symbols]

1 audio encoding / decoding device, 2 input audio detector, 3
Output voice detector, 4a adaptive digital filter,
4b Adaptive algorithm, 4 Estimated speech generator, 5 Error signal generator, 6 Error evaluator.

 ────────────────────────────────────────────────── ─── Continued on the front page (72) Inventor Yuji Narita 2-3-2 Marunouchi, Chiyoda-ku, Tokyo F-term (reference) 5D045 DA20 in Mitsubishi Electric Corporation

Claims (9)

[Claims] An input speech and an output speech of a speech encoding / decoding device for encoding and decoding an input speech to produce an output speech are respectively sampled, and a filter coefficient is controlled by the sampled input speech and an adaptive algorithm. An estimated sound is generated by estimating the sound on the output side by calculation with the filter coefficient of the digital filter, and a difference between the estimated sound and the sampled output sound is calculated and output as an error signal. A method for testing a speech encoding / decoding device, comprising inputting an algorithm to update a filter coefficient of a digital filter following a characteristic change of the speech encoding / decoding device. 2. An input audio detector for sampling an input audio of an audio encoding / decoding device for encoding / decoding an input audio to generate an output audio, an output audio detector for sampling an output audio, An estimated speech is created by estimating the speech on the output side using the sampled input speech and a filter coefficient of a digital filter whose filter coefficient is controlled so as to follow the characteristic fluctuation of the speech encoding / decoding device by an adaptive algorithm. An estimated voice creation unit, an error signal creation unit that calculates a difference between the estimated speech and the output speech and outputs the difference signal as an error signal, and inputs the error signal to an adaptive algorithm of the estimated speech creation unit; An inspection apparatus for a speech encoding / decoding device, comprising: an evaluation unit. 3. The error evaluation section is configured to set a speech abnormality detection level, and output an abnormality signal when the error signal exceeds the speech abnormality detection level. Inspection device for the audio encoding / decoding device described in the above. 4. An error evaluator is provided with a speech abnormality detection level, and is configured to count a time when the error signal exceeds the speech abnormality detection level, evaluate the magnitude of the error signal, and output the result. 3. The inspection apparatus for speech encoding / decoding equipment according to claim 2, wherein: 5. The speech code according to claim 3, wherein the speech abnormality detection level provided in the error evaluation unit is set according to the volume of the input speech. Inspection equipment for encryption / decryption equipment. 6. The speech encoding / decoding apparatus according to claim 1, wherein when the error signal exceeds a predetermined level, control of the filter coefficient by the adaptive algorithm of the digital filter is stopped for a predetermined time. Inspection methods. 7. The estimated speech creation section is configured to stop control of a filter coefficient by an adaptive algorithm of a digital filter for a predetermined time when an error signal exceeds a predetermined level. An inspection device for an audio encoding / decoding device according to any one of claims 2 to 5. 8. An estimated speech is created by varying a time interval used for calculating an input speech and a filter coefficient of a digital filter according to a variation in a time required for encoding / decoding of the speech encoding / decoding device. The method for inspecting a speech encoding / decoding device according to claim 1 or 6, wherein: 9. The estimated voice generating unit changes the time interval used for calculating the input voice and the filter coefficient of the digital filter according to the fluctuation of the time required for the encoding / decoding of the encoding / decoding device. The speech encoding / decoding device inspection apparatus according to any one of claims 2 to 5, wherein the apparatus is configured to generate an estimated speech.