JP2001109494A - Voice identification device and voice identification method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To identify whether input voice is reproduced voice or not. SOLUTION: It is identified whether input voice is reproduced voice or not based on the difference of phase information between real voice and reproduced voice obtained by recording and reproducing real voice. Phase information is preferably a phase difference between a basic wave and the higher harmonic or the phase difference between the higher harmonics. The phase difference is compared between input voice and registered voice.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to an apparatus for evaluating an input voice, and more particularly to an apparatus for discriminating whether an input voice is a raw voice just uttered or a reproduced voice after recording.

[0002]

2. Description of the Related Art A variety of applications are conceivable for a voice identification device. Here, as one application example, an individual using voice is used.
The ID device will be described. Generally, a voice personal ID device is a device in which a user's voice is registered in advance, and whether or not an input voice from a microphone is the same as the registered voice is automatically identified.

[0003] Explaining the prior art in detail, first, a voice having a predetermined utterance content is registered for each user. This is called a registered voice or a verification voice. actually,
When a voice is input by the user, for example, spectrum envelope information is extracted from the voice signal, and the extracted information is stored as reference data at the time of matching.

At the time of collation of input speech, spectrum envelope information is extracted from a speech signal inputted by a user by the same analysis as at the time of registration, and a matching process is performed with stored reference data. As a result of the matching processing, if the difference from the reference data is equal to or greater than a certain threshold, the input voice is rejected as belonging to another person, and if the difference is within a certain threshold, the input voice is the same as the voice of the registered speaker. And performs predetermined processing such as unlocking of the electric lock provided on the door.

However, at the time of registration or the like, the voice uttered by the registered speaker himself is recorded from behind or by installing a hidden microphone in the apparatus, and then the recorded voice is reproduced from a speaker or the like. Speech with very similar spectral envelope information can be input. Although it is necessary to finely adjust the installation position or the direction of the microphone or the speaker in the sound input in this way, the possibility of being identified with the registered sound cannot be denied. In addition, this is called a recording plan, and various measures are taken as follows.

Conventionally, as a method for preventing such an action, Japanese Patent Application Laid-Open No. 5-323990 discloses a method in which the system gives an instruction to input different utterance contents every time authentication is performed. Further, Japanese Patent Application Laid-Open No. 9-127974
Japanese Patent Application Laid-Open No. H11-157,087 describes that the system outputs a different acoustic signal each time and superimposes it on the input voice, and uses the signal after removing the audio signal output by the system from the input voice.

[0007]

However, in the above-mentioned conventional method, the utterance content of the input speech is changed or a specific acoustic signal is superimposed and used. There is a problem that the signal must be changed.

The present invention has been made in view of the above-mentioned conventional problems, and an object of the present invention is to realize highly reliable identification of input speech.

Another object of the present invention is to determine with high precision whether an input sound is a live sound or a reproduced sound.

It is still another object of the present invention to utilize the difference between the properties of a live voice and its reproduced voice for voice evaluation.

[0011]

Means for Solving the Problems (1) Description of the Means In order to achieve the above object, the present invention provides a sound input means for inputting a sound, and phase information from the sound input to the sound input means. , And evaluation means for evaluating the input voice by comparing the phase information of the collation voice with the phase information of the input voice.

According to various experiments by the present inventor, a difference in signal waveform, specifically, a difference in phase information between a live sound and a sound obtained by recording and reproducing the reproduced sound (reproduced sound) was recognized. This is presumed to be due to the influence of the phase characteristics of the recording / reproducing system (especially the speaker). The present invention evaluates an input voice using the phenomenon, and desirably identifies whether the input voice is a live voice or a reproduced voice. ADVANTAGE OF THE INVENTION According to this invention, it is not necessary to change the utterance content, and it is not necessary to superimpose another sound, and a simple and reliable input speech evaluation system can be realized.

Preferably, the phase information is a phase difference between a fundamental wave and a harmonic (or a phase difference between harmonics).
However, other than this, various types of information such as a phase ratio, a signal correlation value, and a phase change can be used as the phase information, as long as the information indicates the difference between the phases of the input audio and the reproduced audio. Further, the phase information may be compared by directly comparing the waveforms themselves.

Preferably, the evaluation means includes means for identifying whether or not the input sound is a reproduced sound based on a comparison between the two pieces of phase information. That is, as described above, the difference between the waveforms of the input voice and the reproduced voice is extracted by comparing the phase information, and the recording plan is determined based on the comparison.

Preferably, the phase information is information on at least one of a phase difference between a fundamental wave and a harmonic of a voice and a phase difference between harmonics. The phases are relative, and basically the difference between the two phases (phase difference) is physically significant. Therefore, it is desirable to use the phase difference between the fundamental wave and the harmonic as a comparison object. In this case, accurate evaluation can be performed by using a phase difference between a fundamental wave having a large power and a harmonic having a large power and a low order among harmonics.

Preferably, the phase information extracting means includes a phase difference Am between a fundamental wave of the verification voice and its m-th harmonic and a phase difference of the verification voice as phase information of the verification voice. The phase difference An between the fundamental and its nth harmonic
And a phase difference Bm between a fundamental wave of the input voice and its m-th harmonic, as phase information of the input voice.
And a means for determining a phase difference Bn between a fundamental wave of the input voice and its nth harmonic, wherein the identification means includes: a difference between the phase difference Am and the phase difference Bm; The input voice is identified based on a difference between a phase difference An and a phase difference between the phase difference Bn.

As described above, if a plurality of phase differences are obtained and compared with each other, the input voice can be evaluated more accurately. Note that the above m and n are integers of 2 or more, and m and n are not the same.

Preferably, the phase information extracting means includes: a preliminary analyzing means for estimating a fundamental wave of the input voice; a main analyzing means for performing a frequency analysis of the input voice based on the estimated fundamental wave; Extracting means for extracting the phase information from the frequency analysis result of the analyzing means.

According to the above configuration, the fundamental wave is first estimated, and the next main analysis can be executed based on the period (frequency) of the fundamental wave. Therefore, the processing conditions of the main analysis can be optimized. As a result, the analysis accuracy can be improved.

Preferably, the preliminary analysis means cuts out the input voice by a fixed time window having a fixed-length window width in frame units to perform frequency analysis, and the main analysis means performs the frequency analysis on the estimated fundamental wave. A variable time window having a window width that is variably set based on the input time is set, and the input voice is cut out in frame units by the variable time window to perform frequency analysis.

The variable time window is desirably set to, for example, about three wavelengths in consideration of the average frequency of a male voice and the average frequency of a female voice. However, if the length is too short, sufficient sampling (cutout) of data cannot be performed. On the other hand, if the length is too long, the frequency shift within the window is greatly affected, and the analysis accuracy may be reduced. is there. If the fundamental wave is estimated by the preliminary analysis, an optimal variable time window can be set based on the fundamental wave, that is, the frequency analysis in the present analysis can be appropriately performed. In the preliminary analysis and the main analysis, an FFT operation and the like are executed, but other various methods can be used.

Preferably, the phase information is extracted from a frame having a predetermined stability condition in the audio signal. If audio signals are cut out in an unstable state, phase information may not be properly extracted. Therefore, the phase information is extracted after confirming the stable state.

Further, in order to achieve the above object, the method according to the present invention utilizes the fact that the phase information differs between a live voice and a reproduced voice obtained by recording and reproducing the raw voice, so that the input voice can be reproduced. It is characterized in that it is distinguished between a live sound and a reproduced sound. This configuration is useful in a security device, and various other applications can be considered.

According to another aspect of the present invention, there is provided a voice recognition apparatus comprising: a determination unit configured to determine personal identity between a verification voice and an input voice; Identification means for identifying whether or not the sound is a reproduced sound based on the phase information.

(2) Explanation of Principle FIG. 1 shows respective signal waveforms of the live sound and the reproduced sound. The upper row (A1) to (D1) is the reproduced voice (that is, the recorded voice), and the lower row (A2) to (D2) is the live voice. Here, the reproduced sound is obtained by recording once a sound in which the same speaker utters the same utterance content and then reproducing the sound. Each audio signal has been passed through a low-pass filter for convenience of explanation. Also, FIG.
(D) is obtained by dividing the audio signal over a certain period into four parts and arranging them in order from the top, with the horizontal axis representing the time axis and the vertical axis representing the amplitude.

When comparing the raw voice and the reproduced voice, the repetition period (basic frequency) of the waveform corresponding to the pitch of the voice coincides with each other, while the peak position and the waveform of the basic unit of repetition are the same. It can be seen that the shapes of the live sound and the reproduced sound are slightly different.

As shown in FIGS. 2 and 3, in general, a vowel portion of an audio signal includes a fundamental wave which provides a fundamental frequency (voice pitch) and twice, three times,. . . Is composed of superimposed harmonics having the following frequencies. The shape is determined by the relative positional relationship between the individual sine waves called the phase.
For example, as is clear from the comparison between FIG. 2 and FIG. 3, when considering the synthesis of three frequency signals, if the phase of any one of the signals is different even at the same frequency, the synthesized waveform greatly changes.

The difference between the waveform shapes of the reproduced voice and the raw voice is a result of the relative positional relationship of the phases being destroyed during the recording / reproducing process.
It has been empirically known that the influence of is large.

Therefore, the present invention utilizes the fact that the signal waveform (specifically, the phase) changes between the reproduced voice and the raw voice even when the same speaker utters the same voice content. , The reproduced voice and the live voice are identified with high accuracy.

[0030]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A case where the principle according to the present invention is applied to a personal ID device using voice will be described below. Of course, the present invention can be applied to devices other than the personal ID device.

FIG. 4 is a block diagram showing the overall configuration of the personal ID device according to this embodiment. The input unit 1 is a means for inputting audio to the device, and includes a microphone, an A / D converter, an amplifier, and the like. The feature amount extraction unit 2 performs a fast Fourier transform (FFT) or a linear prediction (LPC; Linear Prediction Coding) analysis on the digitized audio signal input by the input unit 1 to obtain a frequency domain feature. This is to extract the amount. The matching unit 3
The currently entered voice and the voice spoken during registration are
This is a means for comparing using a known method such as DP matching. The storage unit 4 is a memory that stores the extracted feature amounts. The feature amount extracted from the voice uttered by the user at the time of registration is stored here and used as reference data for comparison with the voice input at the time of collation. The output unit 5 is, for example, a circuit that outputs an unlocking signal to the electric lock when the collation unit 3 determines that the voice is a live voice of the same speaker. On the other hand, if it is determined that the speaker is not the same speaker or that the voice is reproduced, a rejection signal is presented to the user with a buzzer sound or a monitor screen as necessary. In addition, the rejected voice is stored in the storage unit 4 as necessary.
Note that the feature amount extraction unit 2 and the collation unit 3 can be configured by hardware, but can also be substantially configured by software.

Next, using FIG. 5 and FIG.
The flow of processing at the time of registration and collation of a device will be described.

FIG. 5 is a flowchart showing the flow of processing at the time of registration. When the voice is input,
From the signal sequence digitized by the input unit 1, a section (speech section) containing an audio signal is cut out using information such as the amplitude of a waveform and the presence or absence of a fundamental frequency (S).
1). Next, the input speech is subjected to frame analysis to extract a parameter representing the spectrum envelope information (S2). Here, the spectrum envelope information is an outline of the distribution of each frequency component included in the audio signal at a certain moment, and can be obtained by calculating an FFT or LPC cepstrum for each analysis frame. The extracted parameters are stored in the storage unit 4 (S3) and used as reference data at the time of collation. The same applies to the conventional device.

Next, a method of extracting phase information will be described.

In order to obtain the phase more accurately, the FFT
It is important how many fundamental period waveforms are included in the analysis window used in the analysis. Empirically, it has been found that the analysis can be performed with high accuracy when about three are included. Therefore, at the time of registration, two-stage analysis of “preliminary analysis (S100)” and “main analysis (S101)” is performed as described below. When the identification accuracy is not so required, only the latter main analysis may be executed.

First, in the preliminary analysis, a fixed-length window width is used.
(Eg, about 40 ms) to estimate the fundamental period of the audio signal (S4). According to the fixed-length analysis window, although the accuracy is lowered, it is possible to cover a wide range from a low voice of a male (about 70 Hz) to a high voice of a woman (about 500 Hz). The analysis window is continuously scanned along the time axis, and the fundamental period is estimated at each position. In this preliminary analysis, an FFT operation is desirably performed, but an autocorrelation operation may be executed in addition.

Next, in this analysis, the analysis is performed again using an analysis window having a size three times the fundamental period obtained as a result of the preliminary analysis. That is, the above-described spectrum envelope information is obtained by preliminary analysis, that is, analysis using a fixed-length window width without considering the fundamental period. In this analysis, however, about three fundamental periods obtained by the preliminary analysis are included. A window width is newly set and a frequency analysis is performed, thereby extracting phase information (S5).

Here, the test frame will be described.
In general, a vowel part of speech is more stationary than a consonant part, so that speech parameters such as spectrum information and a fundamental frequency can be stably extracted. However, in order to extract the phase information with higher accuracy, it is necessary to exclude a portion of the vowel portion in which the fundamental frequency further changes and a portion where the amplitude level of the harmonic is small from the analysis target. Therefore, only when the following two conditions relating to the amplitude and the phase are satisfied, the analysis frame is set as a frame (test frame) used for extracting the phase difference information.

[Conditions related to amplitude] Fundamental wave, second harmonic,
The ratio between the maximum value and the minimum value of the amplitude level (A _K , A _2K , A _3K ) of the third harmonic is within a predetermined range (for example, 20 dB) (or the minimum value is higher than the predetermined value) thing).

Here, FIGS. 7A and 8A show F
The power at each frequency after the FT analysis is shown,
The horizontal axis corresponds to the number of FFT points (frequency), and the vertical axis represents the logarithmic amplitude value. 7 (B) and FIG. 8 (B)
The phase distribution obtained by performing the unwrapping process on the FFT analysis result is shown. The horizontal axis corresponds to the number of FFT points, and the vertical axis represents the phase. 7 shows a case where the fundamental frequency is stable, and FIG. 8 shows a case where the fundamental frequency is transitioning. The result of the FFT analysis is obtained as a complex number, and the vector angle on the complex plane corresponds to the phase. Originally, the phase is discontinuous between -π and + π, but here, the phase is converted into a continuous numerical value in consideration of the periodicity of the phase after removing the linear phase component. . That is, FIGS. 7B and 8B
Has been subjected to a known unwrapping process.

In FIG. 7, the fundamental wave A _K (around 55 points [about 320 Hz]), the second harmonic A _2K (around 110 points [about 640 Hz]), and the third harmonic A _3K (170 Near the point [about 980Hz]). The subscript K indicates the number of FFT points corresponding to the fundamental frequency (about 320 Hz in FIG. 8). Although a continuous spectrum is shown in FIGS. 7A and 8A, it actually exists as a plurality of line spectra and is observed as a continuous spectrum because the FFT analysis window has a finite length. Have been.

According to the first condition, if any one of these levels is low, the phase difference extraction error becomes large, and is excluded from the test frame.

[Conditions Regarding Phase] All the w _j (j = 1, 2, 3) given by the following equations must be within a certain threshold value (each j is a fundamental wave, a second harmonic, 3 harmonics, respectively). Equivalent to the second harmonic).

[0044]

(Equation 1) Here, Pj represents the phase at the l-th point of the FFT, and K represents the number of FFT points corresponding to the fundamental frequency.
P _2K and P _3K represent the phases of the second harmonic and the third harmonic, respectively. M is an integer determined by the analysis frame length N,

(Equation 2) Given by Here, [x] represents the largest integer not exceeding x. L is the FFT window width, that is, the number of points.

In the example of FIG. 7 described above, the fundamental wave, the second harmonic,
The third harmonic is stable and suitable for phase difference extraction. On the other hand, the example of FIG. 8 is not stable for the fundamental wave, the second harmonic, and the third harmonic, and is generally not suitable for phase difference extraction.

The above equation (1) is obtained by comparing FIG. 7 (B) and FIG.
In (B), it is a measure showing the stability of the phase in the vicinity of each of the fundamental wave, the second harmonic, and the third harmonic (M points before and after). This value is extremely small for an ideal signal, that is, a waveform obtained by synthesizing a sine wave containing no noise component. Therefore, when the value of w _j is large, it can be determined that the fundamental frequency is unstable, that is, the fundamental frequency is changing.

Specifically, in FIG. 7B, the phase in the vicinity of the fundamental frequency (around 55 points) and the second harmonic (around 110 points) and the frequency corresponding to the third harmonic (around 170 points) are shown. It can be seen that the value of is quite stable. On the other hand, in FIG. 8B, the value of the phase is not stable except for the fundamental frequency (around 50 points), and it can be seen that the fundamental frequency has shifted.

In the main analysis (S101) of FIG. 5, a stationary frame that satisfies the above conditions and from which phase information can be stably extracted is labeled as a test frame (S6), and phase difference information is calculated. (S7). The calculated phase difference information is stored as reference data at the time of comparison (S8).

FIG. 6 is a flow chart showing the flow of the process at the time of collation. The matching process is roughly classified into spectral matching for identifying the personality and phonologicalness of the voice using the frequency characteristics of the voice, and phase matching for determining whether the voice is a recorded / reproduced voice using the waveform characteristics. It consists of two steps.

At the time of spectrum matching, when a speech is input, the above parameters calculated in analysis frame units are associated with each analysis frame of the registered speech and the input speech by performing DP matching. Is calculated (S3). The obtained distance is compared with a predetermined threshold value (S4). If the distance is equal to or greater than the threshold value, both voices are determined to be uttered by different speakers and rejected (S4).
5). This is the same as before.

If the distance is equal to or less than the threshold value, it is further focused on the phase information, and it is determined whether or not the input sound is a reproduced sound (phase matching).

When a plurality of input speech frames correspond to the test frames of the registered speech due to delay of the utterance or the like, the same number of frames as the number of test frames are selected from the corresponding input speech frames. Conversely, when a plurality of test frames correspond to the same input voice frame due to a rapid voice or the like, only the frame located near the center of the continuous test frames is used.

Here, it may be determined again whether or not the corresponding input speech frame satisfies the condition as a test frame. It is also possible to check whether or not a frame exists, and if so, it is possible to use the frame in place of the corresponding frame.

Next, a method for obtaining the phase difference between the fundamental wave (period T) and the second harmonic using the test frame obtained above will be described (S6). Based on the analysis result of FFT or the like, the phase of the fundamental wave and the phase of the second harmonic at the center of the analysis window are obtained. It is necessary that the position in the analysis window be the same in the fundamental wave and the second harmonic, and it is preferable to use the center position because the error increases at both ends of the analysis window.

FIG. 9 is a diagram for explaining the definition of the phase difference. The upper waveform is the fundamental wave, and the lower waveform is the phase δ
The second harmonic, which is delayed by only the second harmonic, is shown.

The phase (θ ₁ ) of the fundamental wave is

(Equation 3) Given by Here, d ₁ is the number of points from the center of the analysis window to the peak position of the waveform. Similarly, the phase of the second harmonic (θ ₂ ) is

(Equation 4) Given by The phase difference u is calculated by calculating the amount corresponding to δ in FIG.
It is defined as follows as a value normalized by the period of the second harmonic.

[0057]

(Equation 5) The phase difference (v) between the fundamental wave and the third harmonic can be similarly obtained.

[0058]

(Equation 6) A difference between each value of (u, v) is calculated between the test frame of the registered voice and the corresponding frame of the input voice, and whether the input voice is a recorded voice or a live voice is calculated based on the difference. Is determined. The following _Di is used as the evaluation value in the test frame i (S7).

[0059]

(Equation 7) The above-mentioned _Di is calculated for each of all the test frames. The determination as to whether or not the sound is a reproduced sound is made using a condition such that the value of the sum is equal to or less than a predetermined threshold value, or all the evaluation values are equal to or less than a predetermined threshold value (S8).
In the above formula, the difference (first difference) regarding the phase difference between the fundamental wave and the second harmonic between the input voice and the registered voice,
In addition, the value obtained by adding the difference (second difference) regarding the phase difference between the fundamental wave and the third harmonic is used as the evaluation value, but one of the first difference and the second difference may be used as the evaluation value. However, depending on the reproduction system, there is a possibility that one of the differences does not become so large, so it is desirable to consider a plurality of differences as in the above calculation formula.

As a result of the matching, if the difference from the reference data is equal to or more than the threshold, the rejection is made (S9). If the difference is less than the threshold, it is accepted as the same voice as the registered speaker, and, for example, the electric key of the door is unlocked. A signal is output (S10). Thus, only the voice determined to be the same in the two-stage matching is accepted. Note that various methods can be used as the determination method in S9. For example, a rejection determination may be made when the number of times the matching result is equal to or greater than the threshold is equal to or greater than a predetermined number. May be subjected to various kinds of statistical processing and evaluated.

FIG. 10 shows an example of an evaluation value obtained by analyzing the phase information of a male speaker's voice. The horizontal axis represents the analysis frame number, and the vertical axis represents the evaluation value. In the figure, the results of comparing the raw voices are all gathered in the area where the evaluation value is 0.20 or less, whereas the results of comparing the raw voice and the recording / playback voice are all gathered in the area where the evaluation value is 0.60 or more. It can be seen that the recorded / reproduced sound and the live sound can be clearly separated.

FIG. 11 shows an example of an evaluation value obtained by analyzing the phase information of a female speaker's voice. Although the difference between the evaluation values of the recorded / reproduced voice and the live voice is slightly smaller than that of the male voice, it can be seen that the voice can be separated reliably as with the male voice.

In this embodiment, the fundamental wave and its second harmonic, 3
Although the phase difference between the harmonics is used, the present invention is not limited to this. The phase difference between the second harmonic and the third harmonic may be used, or more than the fourth harmonic may be used.

As for the characteristic amount used for spectrum matching, it is possible to use a characteristic amount generally used for a personal ID device by voice, such as a change pattern of a fundamental frequency, in addition to the spectral envelope information. In addition, in addition to the above-described method of associating analysis frames by DP matching, a vowel portion of a voice is extracted using an HMM (Hidden Markov Model) or the like. It is also possible to use higher-order frames more suitable for the conditions for spectrum matching and phase matching.

[0065]

As described above in detail, according to the present invention, highly reliable identification of input speech can be realized. Further, according to the present invention, it is possible to determine with high accuracy whether the input sound is a live sound or a reproduced sound. Further, according to the present invention, the difference between the properties of the live voice and the reproduced voice can be used for voice evaluation.

[Brief description of the drawings]

FIG. 1 is a waveform diagram showing signal waveforms of a reproduced voice and a raw voice after recording.

FIG. 2 is an explanatory diagram showing the synthesis of three signals.

FIG. 3 is an explanatory diagram showing the synthesis of three signals.

FIG. 4 is a block diagram illustrating a basic configuration of a voice identification device.

FIG. 5 is a flowchart showing a flow of processing at the time of voice registration.

FIG. 6 is a flowchart showing the flow of processing at the time of voice collation.

FIG. 7 is a diagram showing an FFT analysis result and an unwrap processing result.

FIG. 8 is a diagram showing an FFT analysis result and an unwrap processing result.

FIG. 9 is a diagram for explaining the definition of a phase difference.

FIG. 10 is a diagram showing an analysis result of phase information.

FIG. 11 is a diagram showing an analysis result of phase information.

[Explanation of symbols]

1 input unit, 2 feature amount extraction unit, 3 collation unit, 4 storage unit, 5 output unit.

Claims (9)

[Claims] 1. A voice input unit for inputting voice, a phase information extracting unit for extracting phase information from the input voice, and an input by comparing phase information of the collation voice with phase information of the input voice. A voice identification device, comprising: evaluation means for evaluating voice. 2. The apparatus according to claim 1, wherein the phase information is information on at least one of a phase difference between a fundamental wave and a harmonic of a voice and a phase difference between the harmonics. Voice recognition device. 3. The apparatus according to claim 1, wherein said evaluation means includes identification means for identifying whether or not said input sound is reproduced sound based on a comparison between said two pieces of phase information. Voice identification device. 4. The apparatus according to claim 3, wherein the phase information extracting means includes a phase difference Am between a fundamental wave of the verification voice and an m-th harmonic thereof as phase information of the verification voice. And a phase difference A between the fundamental wave of the matching voice and its nth harmonic.
a phase difference Bm between a fundamental wave of the input voice and its m-th harmonic, and a phase difference Bm between the fundamental wave of the input voice and its n-th harmonic. Means for determining a phase difference Bn, wherein the identification means includes a step for determining a phase difference between the phase difference Am and the phase difference Bm and a phase difference between the phase difference An and the phase difference Bn. Wherein the input voice is identified based on a difference between the input voice and the input voice. 5. The apparatus according to claim 1, wherein the phase information extracting means includes: a preliminary analysis means for estimating a fundamental wave of the input voice; and a frequency analysis of the input voice based on the estimated fundamental wave. A voice identification device comprising: a main analysis unit that performs the analysis; and an extraction unit that extracts the phase information from a frequency analysis result of the main analysis unit. 6. The apparatus according to claim 5, wherein the preliminary analysis unit performs frequency analysis by cutting out the input voice by a fixed time window having a fixed length window width in units of frames. A voice identification device, comprising: setting a variable time window having a window width variably set based on the estimated fundamental wave, and performing frequency analysis by cutting out the input voice by a frame unit using the variable time window. . 7. The apparatus according to claim 6, wherein the phase information is extracted from a frame satisfying a predetermined stability condition in the audio signal. 8. Using the fact that the phase information differs between a live voice and a reproduced voice that has been recorded and played back, it is possible to identify whether the input voice is a live voice or a reproduced voice. Characteristic voice identification method. 9. A determining means for determining the personal identity of the matching voice and the input voice, and an identifying means for determining whether or not the input voice for which the identity has been determined is a reproduced voice based on phase information thereof. A voice identification device comprising: