JP5792994B2 - Voice comparison device and voice comparison program - Google Patents

Description

  The present invention relates to an audio comparison apparatus and an audio comparison program that compare two audio signals to determine whether or not they are the same material.

  In broadcasting stations, various monitoring devices are used in order to stably transmit broadcasts. Broadcast stations can monitor broadcast stability by monitoring for abnormalities such as “stop of radio waves”, “no modulation of sound”, “disconnection of line from network station”, and “reverse phase”, for example. Collateral. However, when the same broadcast content (same material) is modulated by various modulation schemes and transmitted, whether or not each broadcast signal is transmitted without error depends on whether the broadcast signal received by the receiver is actually transmitted by humans. It was done by watching.

Patent Document 1 proposes a technique for monitoring whether or not each broadcast signal is transmitted without error with respect to “simultaneous broadcasting” in which the same program is broadcast simultaneously by analog broadcasting and digital broadcasting.
In Patent Literature 1, whether or not the monitoring person notifies the mismatch of audio while listening to one of the analog broadcast audio signal and the digital broadcast audio signal output from the speaker. Is described as a technique for monitoring whether both audio signals are normal (the same content).

JP 2005-340935 A

However, since the technique described in Patent Document 1 compares audio signals on the time axis, the calculation accuracy of the delay time of the other audio signal with respect to one audio signal may drop due to the influence of noise or the like. There is. In addition, sound quality may change when modulated by different modulation methods, and there is a possibility that correct detection results cannot be output.
That is, in the technique described in Patent Document 1, even if audio signals have the same content (the same material), the sound quality changes when the modulation method is different, and it may not be possible to appropriately determine whether the signals are the same material. .

  Therefore, an object of the present invention is to provide an audio comparison device and the like that can appropriately determine the identity of two audio signal materials that are demodulated signals when modulated by different modulation schemes.

The present invention was devised to achieve the above object, and the voice comparison apparatus according to claim 1 is a first voice signal and a second voice that are demodulated signals when modulated by different modulation schemes. A speech comparison device for determining whether or not a signal is the same material, comprising a synchronization means, a frame acquisition means, a frequency conversion means, a spectrum envelope conversion means, a high-pass filter means, and a Mel scale conversion Means, normalization means, distance calculation means, and comparison means.
In such a configuration, the voice comparison device detects a time difference at which the cross-correlation function between the first voice signal and the second voice signal is maximized by the synchronization means, and the first voice signal and the second voice signal are detected according to the time difference. Synchronize with audio signal. Thus, even when a time difference occurs between the time when the first audio signal arrives at the audio comparison device and the time when the second audio signal arrives at the audio comparison device, the synchronization means can synchronize. .
Then, the audio comparison device multiplies each of the first audio signal and the second audio signal by a window function of a predetermined time width by the frame acquisition means, and acquires a frame corresponding to the time width. Thereby, the voice comparison apparatus can divide the waveform of the voice signal into a short time that can confirm the stationarity using the window function.

The audio comparison unit, the frequency converting means, each of the first audio signal and the second audio signal input as a frame corresponding to the predetermined time width by the frame acquisition unit, by a Fourier transform to the frequency domain transform To do. Thereby, the voice comparison apparatus can convert the first voice signal and the second voice signal input as frames by the frame acquisition unit from the time domain to the frequency domain.
Then, the voice comparison device converts each of the first voice signal and the second voice signal that are input after being converted into the frequency domain by the frequency converter by the spectrum envelope converter, into a spectrum envelope. As a result, the speech comparison apparatus can increase the reliability of detection accuracy by comparing spectral envelopes that are effective in determining the characteristics of the speech signal.
Then, the voice comparison device uses a high-pass filter means to convert the spectrum envelope to a spectrum envelope by the spectrum envelope converter and to input a frequency band lower than a predetermined frequency for each of the first voice signal and the second voice signal. By removing this signal, processing for extracting a signal in a frequency band of the predetermined frequency or higher is performed. As a result, the voice comparison apparatus performs processing for emphasizing a high frequency region for the first voice signal and the second voice signal by the high-pass filter means, and converts the frequency characteristic to a frequency characteristic closer to human auditory characteristics. be able to.
Then, the voice comparison device performs processing for extracting a signal in a frequency band of a predetermined frequency or higher by the high-pass filter means by the mel scale conversion means, and inputs each of the first voice signal and the second voice signal that are input. On the other hand, the Mel scale is used to weight the frequency corresponding to the human auditory sensitivity characteristic.
Then, the sound comparison apparatus performs normalization processing on each of the first sound signal and the second sound signal input by weighting in frequency by the mel scale conversion means by the normalizing means. Thereby, the voice comparison device can extract the first voice signal and the second voice signal in a form close to human auditory characteristics by the mel scale conversion means .
Then, the voice comparison device calculates a distance between the first voice signal and the second voice signal normalized by the normalization means by the distance calculation means. Further, the voice comparison apparatus determines whether or not the distance value, which is the calculated value of the distance calculation means, is equal to or greater than a predetermined reference value by the comparison means, and outputs the result. As a result, the voice comparison apparatus determines whether the distance value, which is the difference between the frequency spectra of the first voice signal and the second voice signal, is greater than or equal to a predetermined reference value. It is possible to appropriately determine whether or not the second audio signal is the same material.

Further, the speech comparison apparatus according to claim 2 is the speech comparison apparatus according to claim 1, wherein the distance calculation means calculates the absolute difference between the first speech signal and the second speech signal at a predetermined sample frequency. The square of the value is calculated, and the square root of the sum of the squares of the absolute values of the differences calculated at each sample frequency is calculated as the distance. As a result, the voice comparison device can increase the reliability of detection accuracy by comparing the first voice signal and the second voice signal in the frequency domain and calculating the distance.

According to a third aspect of the present invention , in the voice comparison device according to the first or second aspect , the Mel scale function used in the Mel scale conversion means is Mel if the frequency of the voice signal is f. (f) = 2595log10 (1 + f / 700) Thereby, the voice comparison apparatus can extract the first voice signal and the second voice signal in a form close to human auditory characteristics.

According to a fourth aspect of the present invention, there is provided a program for determining whether or not a first audio signal and a second audio signal which are demodulated signals when modulated by different modulation methods are the same material. Are made to function as synchronization means, frame acquisition means, frequency conversion means, spectrum envelope conversion means, high-pass filter means, mel scale conversion means, normalization means, distance calculation means, and comparison means.
In this configuration, the audio comparison program, depending on the Kite stage, the cross-correlation function between the first audio signal and the second audio signal by detecting a time difference that maximizes the first audio signal in accordance with the time difference And the second audio signal are synchronized.
The audio comparison program, the frame acquisition hand stage Therefore, by multiplying a window function of a predetermined time width to each of the first audio signal and second audio signal synchronized by the synchronization means, the time Get the frame corresponding to the width.
The audio comparison program, depending on the frequency conversion hand stage, each of the first audio signal and the second audio signal input as a frame corresponding to the predetermined time width by the frame acquisition unit, a frequency by Fourier transform Convert to area.
Then, the voice comparison program converts each of the first voice signal and the second voice signal that are input after being converted into the frequency domain by the frequency converter by the spectrum envelope converter, into a spectrum envelope.
Then, the voice comparison program has a frequency band lower than a predetermined frequency for each of the first voice signal and the second voice signal input after being converted into a spectral envelope by the spectral envelope converting means by the high-pass filter means. By removing this signal, processing for extracting a signal in a frequency band of the predetermined frequency or higher is performed.
Then, the voice comparison program uses the mel scale conversion means to perform processing for extracting a signal in a frequency band equal to or higher than a predetermined frequency by the high-pass filter means, and inputs each of the first voice signal and the second voice signal. On the other hand, the Mel scale is used to weight the frequency corresponding to the human auditory sensitivity characteristic.

The audio comparison program, depending on the normalized hand stage, the normalization processing with respect to each of the first audio signal and the second audio signal input by the weighting on the frequency by mel scale conversion means Do.
The audio comparison program, depending on distance calculation hand stage calculates the distance between the first audio signal and second audio signal normalized by the normalization means.
Further, the audio comparison program, depending on the ratio較手stage, the value of the distance is a calculated value of the distance calculation means, to determine whether more than a predetermined reference value, and outputs the result.
Thus, the audio comparison program determines whether the first audio signal and the second audio signal depend on whether the distance value in the frequency domain between the first audio signal and the second audio signal is equal to or greater than a predetermined reference value. It is possible to make the computer appropriately determine whether or not the signal is the same material.

  ADVANTAGE OF THE INVENTION According to this invention, the audio | voice comparison apparatus etc. which can determine appropriately the identity of the raw material of two audio | voice signals which are the demodulation signals at the time of modulating with a different modulation system can be provided.

1 is a schematic configuration diagram of a voice comparison device, an interface device, and a display device according to a first embodiment of the present invention. It is a block block diagram of an audio | voice comparison apparatus, an interface apparatus, and a display apparatus. It is a block block diagram of the synchronization means with which an audio | voice comparison apparatus is provided. It is explanatory drawing about the process which determines the relative delay time of the 2nd audio | voice signal with respect to a 1st audio | voice signal by a synchronizing means, (a) shows the waveform of a 1st audio | voice signal, (b) is the 1st. 2 shows the waveform of the audio signal 2. It is a flowchart which shows the flow of a series of processes of an audio | voice comparison apparatus. It is a block block diagram of the audio | voice comparison apparatus, interface apparatus, and display apparatus which concern on 2nd Embodiment of this invention. It is a flowchart which shows the flow of a series of processes of an audio | voice comparison apparatus. It is a block block diagram of the audio | voice comparison apparatus, interface apparatus, and display apparatus which concern on 3rd Embodiment of this invention. It is a flowchart which shows the flow of a series of processes of an audio | voice comparison apparatus. The figure which shows the experimental data which compared the case where the audio | voice comparison apparatus which concerns on 1st Embodiment of this invention is used, and the case where the audio | voice comparison apparatus which concerns on 3rd Embodiment of this invention is used for 12 SEG broadcast and 1 SEG broadcast. It is. The figure which shows the experimental data which compared the AM broadcast and FM broadcast in the case where the audio | voice comparison apparatus which concerns on 1st Embodiment of this invention is used, and the case where the audio | voice comparison apparatus which concerns on 3rd Embodiment of this invention is used. It is.

  Embodiments of the present invention will be described in detail with reference to the drawings as appropriate. In each figure, common portions are denoted by the same reference numerals, and redundant description is omitted.

<< First Embodiment >>
The audio comparison device 1 according to the present embodiment determines whether or not the first audio signal and the second audio signal, which are demodulated signals when modulated by different modulation schemes, are the same material (same content). Device. The first audio signal and the second audio signal are modulated, for example, by a predetermined broadcast program at a broadcast station (not shown) according to different modulation schemes (for example, AM modulation, FM modulation), and transmitted from the broadcast station as radio waves. Each transmitted signal is received by a receiver (not shown) and further demodulated audio signals.

Before describing each configuration of the voice comparison device 1, devices connected to the voice comparison device 1 will be described.
As shown in FIG. 1, the interface device 2 is connected to the input side of the voice comparison device 1. The interface apparatus 2 is an apparatus that performs A / D conversion on a first audio signal and a second audio signal, which are demodulated signals when modulated by different modulation schemes, according to a predetermined sampling frequency. The interface device 2 includes an A / D converter 201a and an A / D converter 201b. The A / D converter 201 a converts the first audio signal input as an analog signal into a digital signal at a predetermined sampling frequency and outputs the digital signal to the audio comparison device 1. Similarly, the A / D converter 201b performs A / D conversion on the second audio signal input as an analog signal, and outputs it to the audio comparison device 1 as a digital signal.

The first audio signal is transmitted by modulating a predetermined broadcast signal corresponding to one segment broadcast (hereinafter referred to as 1 SEG broadcast) at a broadcast station, and the signal is further demodulated by a demodulator (not shown). It is a demodulated audio signal. The second audio signal is transmitted by modulating a predetermined broadcast signal corresponding to a 12-segment broadcast (hereinafter referred to as 12SEG) at a broadcasting station, and the signal is demodulated by a demodulator (not shown). Audio signal.
By the way, in terrestrial digital broadcasting, one channel is divided into 13 frequency bands (segments), and the modulation using one of the frequency bands is “1 SEG broadcasting”, and the frequency of 12 segments The case where modulation is performed using a band is “12 SEG broadcasting”.
Further, as described above, the first audio signal and the second audio signal are not limited to audio signals obtained by demodulating the 1 SEG broadcast or the 12 SEG broadcast. That is, as long as it is a demodulated signal when modulated by a different modulation method, the voice comparison apparatus 1 can support various modulation methods and demodulation methods. For example, the first audio signal may be an audio signal obtained by demodulating AM (Amplitude Modulation) broadcast, and the second audio signal may be an audio signal obtained by demodulating FM (Frequency Modulation) broadcast.

As shown in FIG. 1, the first audio signal and the second audio signal, which are demodulated signals, are input to the interface device 2 in real time, A / D converted, respectively, and output to the audio comparison device 1.
In addition, one of the first audio signal and the second audio signal arrives at the interface device 2 with a predetermined time delay with respect to the other. As a cause of the delay, there is a delay associated with encoding when transmitting a broadcast signal from a broadcasting station and decoding processed by a receiver (not shown). For example, one audio signal arrives at the interface device 2 with a delay of about 1.7 sec from the other audio signal.

  The display device 3 is connected to the output side of the voice comparison device 1. The display device 3 notifies the monitoring person of normality or abnormality by displaying a determination result when the audio comparison device 1 compares the first audio signal and the second audio signal. In other words, when the voice comparison device 1 determines that the first voice signal and the second voice signal are the same material (same content), the display device 3 displays a display corresponding to “no abnormality”. On the other hand, if the audio comparison device 1 determines that the first audio signal and the second audio signal are different materials (not the same content), the display device 3 is “abnormal”. Display corresponding.

Next, the voice comparison device 1 according to the present embodiment will be described. As shown in FIG. 2, the voice comparison apparatus 1 includes a synchronization unit 101, frame acquisition units 102a and 102b, frequency conversion units 103a and 103b, normalization units 104a and 104b, a distance calculation unit 105, and a comparison unit. 106.
The synchronization unit 101 cuts out a waveform for a predetermined time from each of the first audio signal and the second audio signal, obtains a cross-correlation function for each of the cut-out waveforms, and performs a relative operation when the cross-correlation function is maximized. To determine the delay time. Here, the relative delay time is represented by the difference between the time when the first sound signal reaches the sound comparison device 1 and the time when the second sound signal reaches the sound comparison device 1. For example, assuming that the delay time when the second audio signal arrives after a predetermined time t with respect to the first audio signal is “t”, the opposite case (the second audio signal is the second audio signal). In the case where the audio signal arrives earlier by a predetermined time) is represented by “−t”.
Since the value of the cross-correlation function takes a larger value as the similarity between the two functions is higher, the relative delay time is obtained as a time difference between the waveforms when the value of the cross-correlation function is maximized. Details of the derivation of the relative delay time will be described later.

As shown in FIG. 3, the synchronization means 101 includes storage units 1011a and 1011b, audio signal cutout units 1012a and 1012b, a cross-correlation function calculation unit 1013, a cross-correlation function storage unit 1014, and a cross-correlation function comparison unit 1015. And an audio signal reading unit 1016.
The storage unit 1011a functions as a buffer that temporarily stores the first audio signal converted into a digital signal by the A / D converter 201a of the interface device 2. Similarly, the storage unit 1011b functions as a buffer that temporarily stores the second audio signal converted into a digital signal by the A / D converter 201b of the interface device 2.

The audio signal cutout unit 1012a reads out the first audio signal converted into a digital signal by the A / D converter 201a from the storage unit 1011a, cuts out a waveform for a predetermined time, and outputs it to the cross correlation function calculation unit 1013. For example, as illustrated in FIG. 4A, the audio signal cutout unit 1012a cuts out a waveform having a time width from t1 to t2 from the first audio signal.
In addition, said time width (t2-t1) is a predetermined value. A waveform having a time width from t1 to t2 can be obtained by multiplying the first audio signal by a function having a value of 1 at t1 to t2 and a value of 0 at other times.
Similarly, the audio signal cutout unit 1012b reads out the second audio signal converted into a digital signal by the A / D converter 201b from the storage unit 1011b, cuts out and acquires a waveform for a predetermined time, and calculates a cross-correlation function. Output to the unit 1013. As shown in FIG. 4B, the audio signal cutout unit 1012b first cuts out a waveform having the same time (t1 to t2 in the above example) as the waveform time acquired by the audio signal cutout unit 1012a. And output to the cross-correlation function calculation unit 1013.

Furthermore, the audio signal cutout unit 1012b acquires the waveform of the second audio signal at a time (t1 + t0 to t2 + t0) advanced by t0 from the time (t1 to t2) of the waveform acquired by the audio signal cutout unit 1012a. Output to correlation function calculator 1013. Note that t0 is a predetermined time.
Similarly, the audio signal cutout unit 1012b is (t1 + 2t0 to t2 + 2t0: see FIG. 4), (t1-t0 to t2-t0: not shown), (t1-2t0 to t2-2t0: not shown). Then, the second audio signal at the time shifted by t0 in the positive or negative direction from the time (t1 to t2) is cut out and output to the cross-correlation function calculating unit 1013. It is assumed that the number of data obtained by cutting out from the second audio signal for a predetermined time (in the above case, 5) is set in advance.

The cross-correlation function calculation unit 1013 has a plurality of (five in the above example) that are the audio signals of the predetermined times t1 to t2 acquired by the audio signal cutout unit 1012a and the time width (t2-t1) acquired from 1012b. Are sequentially calculated and output to the cross-correlation function storage unit 1014.
For example, the cross-correlation function calculates a cross-correlation function between the first audio signal at t1 to t2 (see FIG. 4) and the second audio signal at t1 + t0 to t2 + t0 (see FIG. 4), and stores the cross-correlation function. To the unit 1014. The cross-correlation function storage unit 1014 stores a plurality of cross-correlation functions calculated by the cross-correlation function calculation unit 1013.
The cross-correlation function comparison unit 1015 reads out the cross-correlation function values stored in the cross-correlation function storage unit 1014 and sequentially compares them, and the cross-correlation function value calculated by the cross-correlation function calculation unit 1013 is maximized. The relative delay time n * t0 (n is an integer) is output to the audio signal reading unit 1016.

For example, in FIG. 4, the waveform of the second audio signal (see FIG. 4B) is advanced by time t0 from the waveform of the first audio signal (see FIG. 4A). In this case, among the plurality of cross-correlation functions calculated by the cross-correlation function calculation unit 1013, a mutual relationship between the first audio signal in the range of t1 to t2 and the second audio signal in the range of t1 + t0 to t2 + t0. The correlation function is maximized.
In this case, the cross-correlation function comparison unit 1015 outputs “−t0”, which is the relative delay time of the second audio signal to the first audio signal, to the audio signal reading unit 1016. Note that the minus sign indicates that the second audio signal is advanced by time t0 (delayed by -t0) with respect to the first audio signal.
The audio signal reading unit 1016 performs a reading process so that each audio signal is synchronized according to the relative delay time output from the cross-correlation function comparison unit 1015. In the above example, the audio signal reading unit 1016 reads the second audio signal from the storage units 1011a and 1011b by delaying the second audio signal by the time t0 with respect to the timing input to the synchronization unit 101, and acquires the frame. Output to means 102a, 102b.
Through the processing as described above, the synchronization unit 101 can synchronize the first audio signal and the second audio signal.

Returning to FIG. 2 again, the description of the voice comparison device 1 will be continued. In order to avoid redundant description, the frame acquisition units 102a and 102b, the frequency conversion units 103a and 103b, and the normalization units 104a and 104b will be described only when processing the first audio signal, and the second audio signal will be described. Is the same as the processing of the first audio signal, and a description thereof will be omitted.
The frame acquisition unit 102a cuts out a frame from the first audio signal output from the synchronization unit 101 using a predetermined window function. Here, the “frame” refers to a portion of the audio signal cut out by the window function.
Although the statistical properties of speech change with time, it is thought that speech can be assumed to be stationary within a sufficiently short period. Therefore, in the analysis of the voice signal, the window function is used to divide the waveform of the voice signal into a short time so that its continuity can be confirmed. As the window function, for example, a rectangular window, a Barrett window, a Hanning window, a Hamming window, a Blackman window, or the like can be used. In the present embodiment, analysis is performed while sequentially shifting a time window having a length of 10 to 1000 msec at intervals of 10 to 1000 msec.
The first audio signal that has been subjected to synchronization processing by the synchronization means 101 is cut into frames by the frame acquisition means 102a and output to the frequency conversion means 103a.

The frequency transform unit 103a transforms the first audio signal cut into frames by the frame acquisition unit 102a from the time domain to the frequency domain by executing a fast Fourier transform (FFT). That is, the frequency conversion means 103a converts the first audio signal input as a time domain signal into a frequency spectrum by performing discrete Fourier transform on the first audio signal at high speed.
Therefore, the first audio signal cut into the frame by the frame acquisition unit 102a and input is converted from the time domain to the frequency domain audio signal by the frequency conversion unit 103a and output to the normalization processing unit 104a. .

The normalizing unit 104a performs normalization processing on the first audio signal input after being converted into the frequency domain by the frequency converting unit 103a. Specifically, the normalizing means 104a performs normalization by dividing the function S (f) representing the spectrum by the sum of values (intensities) at the respective sample frequencies. That is, the normalizing means 104a executes normalization processing expressed by the following formulas (1) and (2).
S (f) is a function representing a spectrum, f is a frequency, f _min is a minimum frequency, f _max is a maximum frequency, R is a sum of values (intensities) at each sample frequency, and N (f) is normalized. Spectrum.

When the volume of the audio signal is increased, the waveforms of the first audio signal and the second audio signal are also increased, so that an error in comparing frequency components is increased. As in the present embodiment, the normalization unit 104a is provided in the subsequent stage of the frequency conversion unit 103a, so that the audio signal can be normalized by normalizing the waveform of each audio signal even when the reference audio level is shifted at the time of input. It is possible to eliminate variations in signal size and prevent erroneous determination.
The first audio signal input after being converted into the frequency domain by the frequency conversion unit 103 a is normalized by the normalization unit 104 a and output to the distance calculation unit 105.

The distance calculating unit 105 calculates a distance (norm) between the first audio signal normalized and input by the normalizing unit 104a and the second audio signal normalized and input by the normalizing unit 104b. That is, the distance calculation unit 105 calculates the absolute value of the intensity difference for each sample frequency for the first audio signal input from the normalization unit 104a and the second audio signal input from the normalization unit 104b. Further, a distance (norm) is calculated from the absolute value of the intensity difference at each sample frequency calculated, and is output to the comparison means 106.
When the spectrum of the first audio signal is S (f) and the spectrum of the second audio signal is T (f), the following (Equation 3) and (Equation 4) are obtained.

Accordingly, the norm ‖ S (f) −T (f) ‖ that is the distance between the first audio signal and the second audio signal is expressed by the following equation (5).

The comparison unit 106 determines whether or not the distance (norm ‖ S (f) −T (f) 算出) calculated by the distance calculation unit 105 is equal to or greater than a predetermined value, and displays the result. Output to the device 3.
If the distance calculated by the distance calculation means 105 is equal to or greater than a predetermined value, the first audio signal and the second audio signal are considered to have different contents. Accordingly, in this case, the comparison means 106 outputs a signal corresponding to “abnormal” to the display device 3. In addition, when the distance calculated by the distance calculation unit 105 is smaller than a predetermined value, the first audio signal and the second audio signal are considered to have the same content. Therefore, in this case, the comparison means 106 outputs a signal corresponding to “no abnormality” to the display device 3.
Note that the comparison unit 107 calculates the total value of the distances between the spectral envelopes calculated within a predetermined time (for example, 10 seconds), and whether the total value is equal to or greater than the predetermined value. It may be determined whether or not.

FIG. 5 is a flowchart showing a flow of a series of processes of the voice comparison device 1 according to the present embodiment. In order to avoid redundant description, only steps S20 to S40 will be described for the first audio signal.
In step S <b> 10 of FIG. 5, the voice comparison apparatus 1 performs synchronization processing by the synchronization unit 101 so that the input first voice signal and second voice signal are synchronized. That is, the synchronization unit 101 obtains a relative delay time when the cross-correlation function between the first audio signal and the second audio signal is maximized, and outputs one after delaying to the other according to the relative delay time. Thus, the synchronization process is performed.
In step S20, the voice comparison apparatus 1 uses the frame acquisition unit 102a to cut out a frame from the first voice signal input from the synchronization unit 101 using a predetermined window function. In step S30, the speech comparison apparatus 1 performs a Fourier transform at high speed on the first speech signal input from the frame acquisition unit 102a by the frequency conversion unit 103a, so that the first input as a time domain signal. 1 audio signal is converted into a frequency spectrum. In step S40, the speech comparison apparatus 1 performs normalization processing on the first speech signal input from the frequency conversion unit 103a by the normalization unit 104a.

  In step S50, the speech comparison apparatus 1 determines the distance (norm) between the first speech signal input from the normalization unit 104a and the second speech signal input from the normalization unit 104b by the distance calculation unit 105. calculate. In step S60, the voice comparison apparatus 1 determines whether the distance (norm) calculated by the distance calculation unit 105 is equal to or greater than a predetermined value by the comparison unit 106. If the distance (norm) calculated by the distance calculation unit 105 is greater than or equal to a predetermined value in step S60 (step S60 → Yes), the voice comparison device 1 causes the display unit to display the display unit. 3 outputs a signal corresponding to “abnormal” (step S70). If the distance (norm) calculated by the distance calculation unit 105 is smaller than a predetermined value in step S60 (step S60 → No), the voice comparison device 1 displays the distance by the comparison unit 106. A signal corresponding to “no abnormality” is output to the unit 3 (step S80).

According to the speech comparison apparatus 1 according to the present embodiment, the frame acquisition means 102a and 102b cut out the waveform of each speech signal using a predetermined window function in such a short time that its continuity can be confirmed, and further the frequency The conversion means 103a and 103b perform processing for conversion to the frequency domain.
Therefore, even when an error occurs in the relative delay time calculated by the synchronization unit 101, the fluctuation result of the frequency characteristic of the audio signal cut out in a short time that can confirm the continuity is small. Has little effect.
When it is known that the relative delay time of the second audio signal with respect to the first audio signal does not vary, the process of calculating the relative delay time by the synchronization unit 101 needs to be sequentially performed for each frame. It is not necessary to perform the comparison process by the voice comparison device 1 only once.
Alternatively, the relative delay time may be calculated every predetermined time.

When the sound quality monitoring apparatus 1 according to the present embodiment modulates the same sound quality material with different modulation methods, even if a difference in sound quality due to the difference in the modulation method appears in the demodulated signal, each sound signal It is possible to appropriately detect whether the frequency is normal or abnormal by comparing the distance (norm) with a predetermined value.
In addition, the sound quality monitoring apparatus 1 according to the present embodiment uses the normalization means 104a and 104b to normalize the first audio signal input from the frequency conversion means 103a and the second audio signal input from the frequency conversion means 103b. Normalization processing is performed. As a result, even when the reference audio level is shifted at the time of input, by normalizing the waveform of each audio signal, variations in the size of the audio signal can be eliminated and erroneous determination can be prevented.
Therefore, according to the sound quality monitoring apparatus 1 according to the present embodiment, it is possible to appropriately determine the identity of the material of the first audio signal and the second audio signal, which are demodulated signals when modulated by different modulation methods. it can

<< Second Embodiment >>
The voice comparison device 1A according to the present embodiment is different from the first embodiment in that it includes spectrum envelope conversion means 107a and 107b, but the other points are the same as those in the first embodiment. Therefore, the difference will be mainly described, and the description of the points common to the first embodiment will be omitted. Further, in the following description, the process for the first audio signal will be mainly described, and the process for the second audio signal is the same as the process for the first audio signal, and thus the description thereof is omitted.

As shown in FIG. 6, in the speech comparison apparatus 1A according to the present embodiment, a spectrum envelope conversion unit 107a is inserted between the frequency conversion unit 103a and the normalization unit 104a. Note that the synchronization unit 101 shown in FIG. The processing by the frame acquisition units 102a and 102b and the frequency conversion units 103a and 103b is the same as the processing described in the first embodiment (see FIG. 2).
The spectrum envelope conversion unit 107a converts the first sound signal input after being converted into the frequency domain by the frequency conversion unit 103a into a spectrum envelope. As a method for obtaining the spectral envelope, linear predictive coding (LPC) is effective. Linear prediction analysis is a method for obtaining each coefficient (LPC coefficient vector) in a linear combination in order to predict a current output that is one step ahead using a past output and a current input.
Specifically, the spectral envelope conversion means 107a uses linear prediction analysis to represent each sampling value of the speech signal as a linear combination of several past sampling values, and based on the mean square value of the predicted minimum error. Find individual weighting factors. Incidentally, the predicted error is obtained by the difference between the actual sampling value and the predicted value.
That is, the spectrum envelope conversion means 107a estimates the parameters (formant, etc.) of the first speech signal using linear prediction analysis, converts them into a spectrum envelope, and outputs them to the normalization means 104a.

Then, the normalizing unit 104a performs normalization processing on the first spectral envelope corresponding to the first speech signal input from the spectral envelope converting unit 107a. Specifically, as described above, the normalizing unit 104a performs normalization by dividing the function S (f) representing the spectrum by the sum of values (intensities) at each sample frequency of the spectrum envelope.
The distance calculation means 105 then normalizes the first sound signal (first spectrum envelope) input by the normalization means 104a and the second sound signal (normalization input by the normalization means 104b). The distance (norm) from the second spectrum envelope) is calculated.
Further, the comparison unit 106 determines whether or not the distance (norm ‖ S (f) −T (f) ‖) calculated by the distance calculation unit 105 is equal to or greater than a predetermined value. Is output to the display device 3.

FIG. 7 is a flowchart showing a flow of a series of processes of the voice comparison device 1A according to the present embodiment. Steps S10 to S30 shown in FIG. 7 are the same as steps S10 to S30 described in FIG.
In step S31 of FIG. 7, the speech comparison apparatus 1A converts the first speech signal input from the frequency conversion unit 104a into a spectrum envelope by the spectrum envelope conversion unit 105a. Then, in step S40, the speech comparison apparatus 1A performs normalization processing on the first speech signal (first spectrum envelope) input from the spectrum envelope conversion unit 105a by the normalization unit 104a. In step S50, the speech comparison apparatus 1A causes the distance calculation unit 105 to input the first audio signal (first spectrum envelope) input from the normalization unit 104a and the second audio signal input from the normalization unit 104b. The distance (norm) from (second spectrum envelope) is calculated. Steps S60 to S80 are the same as steps S60 to S80 described with reference to FIG.

In the speech comparison apparatus 1A according to the present embodiment, the spectrum envelope conversion units 107a and 107b output the spectrum envelope of each speech signal using linear prediction analysis, and further, normalization processing is performed by the normalization units 104a and 104b. Then, the distance calculation means 105 calculates the distance (norm) between the two spectrum envelopes. When only frequency conversion is performed on each audio signal, there is a possibility that information that is not necessary for determining the characteristics of the audio signal may be included, but the audio comparison apparatus 1A according to the present embodiment As described above, the reliability of detection accuracy can be improved by comparing the spectral envelopes that are effective in determining the characteristics of the audio signal.
That is, in the sound quality monitoring apparatus 1A according to the present embodiment, when the same sound quality material is modulated and demodulated with different modulation methods, a difference in sound quality due to the difference in the modulation method appears in each audio signal. However, by performing a comparison focusing on the spectral envelope that is the “sound characteristic” of the sound source, it is possible to appropriately determine the identity of the material for each audio signal.

«Third embodiment»
The voice comparison device 1B according to the present embodiment is different from the second embodiment in that the high-pass filter means 108a and 108b and the mel scale conversion means 109a and 109b are provided, but other points are the same as those of the second embodiment. Same as the case. Therefore, the difference will be mainly described, and the description of points that are common to the second embodiment will be omitted. Also, the processing for the first audio signal will be mainly described, and the processing for the second audio signal is the same as the processing for the first audio signal, and thus the description thereof is omitted.

As shown in FIG. 8, in the speech comparison apparatus 1B according to the present embodiment, a high-pass filter unit 108a and a mel scale conversion unit 109a are inserted between the spectrum envelope conversion unit 107a and the normalization unit 104a. Yes.
The processing by the synchronization unit 101, the frame acquisition units 102a and 102b, and the frequency conversion units 103a and 103b shown in FIG. 8 is the same as the processing described in the first embodiment (see FIG. 2). Further, the processing by the spectrum envelope conversion means 107a and 107b shown in FIG. 8 is the same as the processing described in the second embodiment (see FIG. 6).

The high-pass filter means 108a removes a signal in a frequency band lower than a predetermined frequency from the first speech signal (first spectrum envelope) input after being converted into a spectrum envelope by the spectrum envelope conversion means 107a. Thus, processing for extracting a signal in a frequency band equal to or higher than the predetermined frequency is performed.
It is known that the volume of sound felt by humans varies with frequency even at the same sound pressure level, and the volume of sound felt by humans is represented by an “equal loudness curve (Fletcher Manson curve)”. The speech comparison apparatus 1B according to the present embodiment approximates the characteristics of the above equal loudness curve by the high-pass filter means 108a that performs processing for emphasizing a high frequency region. As the high-pass filter means 108a, for example, an FIR filter (finite impulse response filter) can be used.

The high-pass filter means 108a has a characteristic of taking out an audio signal in a high frequency band equal to or higher than a predetermined reference frequency and removing a low frequency component. The reference frequency is a predetermined frequency.
The characteristic of the FIR filter that is effective as the high-pass filter means 108a is represented by the transfer function H (z) when the impulse response is Z-transformed, and is expressed by the following equation (6). The FIR filter has features such as stability without requiring feedback and maintaining linear phase information.

  As for the first sound signal input after being converted into a spectrum envelope by the spectrum envelope conversion means 107a, a high frequency band sound signal is taken out by the high-pass filter means 108a and output to the mel scale conversion means 109a. .

The mel scale conversion unit 109a performs a process of extracting a signal in a frequency band equal to or higher than a predetermined frequency by the high-pass filter unit 108a, and uses the mel scale function to perform human auditory sensitivity on the input first audio signal. Weighting on the frequency corresponding to the characteristic is performed. Here, the Mel scale function is expressed by the following equation (7). Note that f [Hz] represents the frequency of the audio signal.

The Mel scale is a scale that represents the pitch of sounds felt by humans, and in particular, is a speech recognition technique with improved frequency resolution in the low frequency range. Therefore, the first audio signal can be extracted by the mel scale conversion means 109a in a form close to human auditory characteristics. For example, an MFCC method (Mel Frequency Cepstrum Coefficient) can be used as a technique for performing mel scale conversion. The MFCC method is executed by performing an inverse Fourier transform on the output of a band-pass filter (not shown) having a triangular frequency characteristic and extracting a low quefrency portion.
The first audio signal that has been extracted from the high frequency band audio signal by the high-pass filter means 108a and converted into a mel scale by the mel scale conversion means 109a is output to the normalization means 104a.

Then, the normalizing means 104a performs normalization processing on the first audio signal (first spectrum envelope) input from the mel scale conversion means 109a. Specifically, as described above, the normalizing unit 104a performs normalization by dividing the function S (f) representing the spectrum by the sum of values (intensities) at each sample frequency of the spectrum envelope.
The distance calculation means 105 then normalizes the first sound signal (first spectrum envelope) input by the normalization means 104a and the second sound signal (normalization input by the normalization means 104b). The distance (norm) from the second spectrum envelope) is calculated.
Further, the comparison unit 106 determines whether or not the distance (norm ‖ S (f) −T (f) ‖) calculated by the distance calculation unit 105 is equal to or greater than a predetermined value. Is output to the display device 3.

FIG. 9 is a flowchart showing a flow of a series of processes of the voice comparison device 1B according to the present embodiment. Steps S10 to S31 shown in FIG. 9 are the same as steps S10 to S31 described in FIG.
In step S32 of FIG. 9, the speech comparison apparatus 1B performs predetermined processing on the first speech signal (first spectrum envelope) input by the high-pass filter unit 108a and converted into the spectrum envelope by the spectrum envelope conversion unit 107a. By removing signals in a frequency band lower than the predetermined frequency, signals in a frequency band above the predetermined frequency are extracted. Next, in step S33, the speech comparison apparatus 1B uses the mel scale function for the first speech signal (first spectral envelope) input from the high-pass filter means 108a by the mel scale conversion means 109a. Performs frequency weighting corresponding to human auditory sensitivity characteristics. In step S40, the speech comparison apparatus 1B performs normalization processing on the first speech signal (spectrum envelope) input from the mel scale conversion unit 109a by the normalization unit 104a. In step S50, the voice comparison apparatus 1B causes the distance calculation unit 105 to input the first voice signal (first spectrum envelope) input from the normalization unit 104a and the second voice signal input from the normalization unit 104b. The distance (norm) from (second spectrum envelope) is calculated. Steps S60 to S80 are the same as steps S60 to S80 described with reference to FIG.

The voice comparison device 1B according to the present embodiment converts the frequency characteristics close to human auditory characteristics by emphasizing the high frequency region by the high-pass filter means 108a and 108b. Furthermore, the voice comparison device 1B can convert the frequency characteristic closer to the human auditory characteristic by performing weighting on the frequency corresponding to the human auditory sensitivity characteristic by the mel scale conversion means 109a and 109b.
That is, when the sound quality monitoring apparatus 1B according to the present embodiment modulates and demodulates the same sound quality material with different modulation schemes, a difference in sound quality due to the difference in the modulation schemes appears in each audio signal. However, it is possible to appropriately determine the identity of the material for each audio signal by taking a human auditory characteristic and making a comparison focusing on the “sound characteristics” of the sound source.

[Audio comparison program]
The voice comparison devices 1, 1 </ b> A, and 1 </ b> B described above can be realized by operating a general computer with a program that causes each of the above-described units and units to function.

  In the above, the voice comparison devices 1, 1A, 1B according to the present invention have been specifically described by the embodiments for carrying out the invention. However, the gist of the present invention is not limited to these descriptions, and It should be interpreted broadly based on the scope description. Needless to say, various changes and modifications based on these descriptions are also included in the spirit of the present invention.

In the voice comparison devices 1, 1A, 1B, as shown in FIG. 4, the range is shifted in the positive or negative direction by t0 from the range of time t1 to t2 with respect to the first audio signal at the time of t1 to t2. In the above description, the second audio signal in each range is acquired, but the present invention is not limited to this.
For example, the first audio signal in each range is acquired by shifting the range in the positive or negative direction by t0 from the range of time t1 to t2 with reference to the second audio signal at time t1 to t2. Also good. Further, for example, if the relative delay time of the second audio signal with respect to the first audio signal is previously registered (the relative delay time in this case is assumed to be t3), the relative delay time t3 is the center. The relative delay time may be calculated in each case by sequentially shifting the range of cutting out the waveform in the positive or negative direction.

  Further, when the relative delay time of the second audio signal with respect to the first audio signal is known in advance (the relative delay time in this case is t4), the audio signal is sequentially shifted by a predetermined time. The process of cutting out and calculating a plurality of cross-correlation functions can be omitted. In this case, the audio signal reading unit 1016 (see FIG. 3) may synchronize the first audio signal and the second audio signal according to the relative delay time t4 and output them to the frame acquisition units 102a and 102b, respectively. . Incidentally, in this case, a known relative delay time t4 is input in advance via an input unit (not shown) and stored in a storage unit (not shown), and the audio signal reading unit 1016 receives the relative delay time. It is only necessary to read t4 from the storage unit and synchronize the first audio signal and the second audio signal.

  Further, in the voice comparison devices 1, 1A, 1B according to the present invention, the synchronization unit 101 (see FIG. 3) includes the storage units 1011a, 1011b and the cross-correlation function storage unit 1014. Not limited to. That is, the storage units 1011a and 1011b and the cross-correlation function storage unit 1014 may be configured to exist outside the voice comparison devices 1, 1A and 1B.

  Also, program information is received from a broadcasting station (not shown) outside the audio comparison devices 1, 1A, 1B, and the first audio signal and the second audio signal correspond to the same program. It is good also as providing the program information receiver (not shown) which outputs whether it is with respect to the audio | voice comparison apparatus 1, 1A, 1B. In this case, a signal indicating whether the first broadcast corresponding to the first audio signal and the second broadcast corresponding to the second audio signal are the same from the program information receiving apparatus is an audio signal. It is input to the comparison means 106 (see FIGS. 2, 6, and 8) of the comparison apparatus 1.

When a signal indicating that the first broadcast and the second broadcast are the same program is input from the program information receiving apparatus, the comparison unit 106 performs the same process as described above. That is, the comparison unit 106 outputs a signal corresponding to “normal” or “abnormal” to the display device 3 depending on whether the value of the distance (norm) calculated by the distance calculation unit 105 is greater than a predetermined value. . On the other hand, when a signal indicating that the first broadcast and the second broadcast are different programs is input from the program information receiving apparatus, the comparison means 107 calculates the distance calculated by the distance calculation means 105. Regardless of the value of (norm), a signal corresponding to “normal” is always output to the display device 3.
From the program information receiving apparatus, for example, the first audio signal is an audio signal corresponding to a “marathon relay” broadcast, and the second audio signal is an audio signal corresponding to a “talk program” broadcast. This is because it is appropriate to determine “normal” regardless of the distance value between the first audio signal and the second audio signal when it is known in advance from the input program information.

As shown in FIG. 8, the voice comparison device 1B has a configuration in which high-pass filter means 108a and 108b and mel scale conversion means 109a and 109b are added to the voice comparison device 1A according to the second embodiment. However, a configuration in which only the high-pass filter means 108a and 108b are added among the above-described configurations may be adopted. In this case, the normalizing unit 104a performs normalization processing on the first speech signal (spectrum envelope) input with the high frequency region emphasized by the high-pass filter unit 108a, and sends the normalized signal to the distance calculating unit 105. Output. The same applies to the second audio signal.
Further, only the mel scale conversion means 109a and 109b may be added to the voice comparison device 1A according to the second embodiment. In this case, the mel scale conversion unit 109a weights the first sound signal (spectrum envelope) converted into the spectrum envelope by the spectrum envelope conversion unit 107a on the frequency corresponding to the human auditory sensitivity characteristic. And output to the normalizing means 104a.

FIG. 10 shows a comparison between 12 SEG broadcasting and 1 SEG broadcasting when the speech comparison apparatus 1 according to the first embodiment of the present invention is used and when the speech comparison apparatus 1B according to the third embodiment of the present invention is used. It is a figure which shows the experimental data which were performed.
The broken line in the graph of FIG. 10 is a case where processing is performed using the speech comparison apparatus 1 according to the first embodiment of the present invention, and synchronization processing, frame acquisition processing, frequency conversion processing, Normalization processing is performed, and the difference (distance) between the strengths of the two audio signals is calculated every second. In the figure, for the sake of simplicity, “frequency spectrum” is described.
On the other hand, the solid line in the graph of FIG. 10 shows a case where processing is performed using the speech comparison apparatus 1B according to the third embodiment of the present invention. In addition to the processing of the speech comparison apparatus 1 according to the first embodiment, A spectrum envelope conversion process, a high-pass filter process, and a mel scale conversion process are performed to calculate a difference (distance) between two spectrum envelopes every second. In the figure, for the sake of simplicity, “spectrum envelope + high pass filter + MEL scale” is described.

In the experimental data shown in FIG. 10, the amount of sample points differs between the case of “frequency spectrum” indicated by a broken line and the case of “spectrum envelope + high-pass filter + MEL scale” indicated by a solid line. Normalized based on quantity. In the experimental data shown in FIG. 10, the shift width of the window function used in the frame acquisition means 102a and 102b (see FIGS. 2 and 8) is 100 msec, and the length of the window function is 66 msec. The sampling frequency was 48000 Hz and the quantization bit was 16 bits. Further, in the experimental data shown in FIG. 10, the difference (distance) between the audio signals when the 12 SEG broadcast and the 1 SEG broadcast have the same broadcast content is 0 dB.
Also, assuming that a failure has occurred in the transmission of 1 SEG broadcast, a section B1 in which the 1 SEG broadcast is silent and a section D1 in which the 1 SEG broadcast is white noise (white noise) are provided to indicate whether an abnormality has been detected. It was. White noise is noise in which all waves overlap with the same energy.

  In the section indicated by A1 and C1 shown in FIG. 10, the voice comparison apparatuses 1 and 1B receive the 12SEG broadcast as the first voice signal, and the 1SEG broadcast having the same content as the first voice signal as the second voice signal. Received. In the section indicated by B1, the voice comparison apparatuses 1 and 1B receive the 12SEG broadcast as the first voice signal, and the second voice signal is silent. In the section indicated by C1, similarly to the range of A1, the voice comparison apparatuses 1 and 1B receive 12SEG broadcast as the first voice signal, and 1SEG having the same content as the first voice signal as the second voice signal. I received the broadcast. In the section indicated by D1, the voice comparison apparatuses 1 and 1B receive 12SEG broadcast as the first voice signal, and the second voice signal is set as stationary white noise.

  As shown in FIG. 10, in both the “spectrum envelope + high-pass filter + MEL scale” indicated by the solid line and the “frequency spectrum” indicated by the broken line, the value of the difference in waveform in the B1 (1 SEG silence) interval Increases to indicate that an abnormality has been detected. In addition, in both “spectrum envelope + high-pass filter + MEL scale” indicated by a solid line and “frequency spectrum” indicated by a broken line, the value of the waveform difference increases in the D1 (1 SEG white noise) section, which is abnormal. It is understood that has been detected. Further, in the case of “spectrum envelope + high-pass filter + MEL scale” than in the case of “frequency spectrum”, the values in the normal section (the sections of A1 and C1 in which the contents of the two audio signals are the same) and It can be seen that there is a large relative difference between the values in the abnormal sections (the section of C1 in which the 1 SEG broadcast is silent and the section of D1 in which the 1 SEG broadcast is white noise). That is, it is determined whether or not the materials of the two audio signals are the same in the case of “spectrum envelope + high-pass filter + MEL scale” rather than “frequency spectrum”, based on a predetermined threshold. Judgment can be made appropriately.

FIG. 11 shows a comparison between AM broadcast and FM broadcast when the speech comparison apparatus 1 according to the first embodiment of the present invention is used and when the speech comparison apparatus 1B according to the third embodiment of the present invention is used. It is a figure which shows the experimental data which were performed.
The broken line in the graph of FIG. 11 is the case where processing is performed using the speech comparison apparatus 1 according to the first embodiment of the present invention, similar to the experimental data shown in FIG. “Frequency spectrum” is described.
On the other hand, the solid line in the graph of FIG. 11 is the case where processing is performed using the speech comparison apparatus 1B according to the third embodiment of the present invention, as in the experimental data shown in FIG. Therefore, it is described as “spectrum envelope + high pass filter + MEL scale”.

  In the experimental data shown in FIG. 11, the amount of sample points differs between the case of “frequency spectrum” indicated by a broken line and the case of “spectrum envelope + high-pass filter + MEL scale” indicated by a solid line. Normalized based on quantity. Further, in the experimental data shown in FIG. 11, the shift width of the window function used in the frame acquisition means 102a and 102b (see FIGS. 2 and 8) is 100 msec, and the length of the window function is 66 msec. The sampling frequency was 48000 Hz, and the quantization bit was 16 bits. Further, in the experimental data shown in FIG. 11, the difference (distance) between the audio signals when the AM broadcast and the FM broadcast have the same broadcast content is set to 0 dB.

In the experimental data shown in FIG. 11, the sound level of the measurement sound is set to +4 dBm as a -18 dBFS sine wave signal when the full bit scale when quantized with 16 bits is set to 0 dBFS. The news program of the material was recorded at an average talk level of about 0 dBm and recorded by demodulating broadcast waves. Also, in the range indicated by B2, stationary noise (thermal noise) is mixed at about −16 dBm to −11 dBm into audio material having an average talk level of 0 dBm to +4 dBm. In addition, when 1 mW of electric power was consumed by adding a 1 kHz sine wave signal to a device having an impedance of 600Ω, 0 dBm was set, and when the voltage was 0.775 V, 0 VU = + 4 dBm.
Incidentally, the average sound level is a value read from a VU meter (not shown). Due to the electrical characteristics of the VU meter, when the response speed is 300 ms and a continuous signal of 300 ms or longer is received, an accurate level is indicated. In this case, the VU meter can indicate the average value of the input signal as a VU value (Volume Unit).

  In the section A2 to C2 in FIG. 11, the AM broadcast and the FM broadcast have the same broadcast contents (news program). However, assuming that a failure has occurred in the transmission of the AM broadcast, the AM broadcast is used in the section B2. Stationary noise was mixed, and pulse noise was mixed into the AM broadcast in the section C2. In the section D2, AM broadcasting is set as white noise. In the section E2, AM broadcast and FM broadcast have different contents (marathon relay in AM broadcast and talk program in FM broadcast).

  As shown in FIG. 11, in both the case of “frequency spectrum” indicated by a broken line and “spectrum envelope + high-pass filter + MEL scale” indicated by a solid line, B2 (when stationary noise is mixed in AM broadcasting) ), C2 (when pulse noise is mixed in AM broadcast), D2 (when the first audio signal is stationary noise), E2 (when the contents of the first audio signal and the second audio signal are different) The value of the waveform difference increases in the interval, indicating that an abnormality has been detected. That is, in both cases of the voice comparison device 1 (see “frequency spectrum” shown by a broken line) and the voice comparison device 1B (see “spectrum envelope + high-pass filter + MEL scale” shown by a solid line), It can be seen that the abnormality was properly detected.

1, 1A, 1B Speech comparison apparatus 101 Synchronization means 102a, 102b Frame acquisition means 103a, 103b Frequency conversion means 104a, 104b Normalization means 105 Distance calculation means 106 Comparison means 107a, 107b Spectral envelope conversion means 108a, 108b High-pass filter Means 109a, 109b Mel scale conversion means 2 Interface device 201a, 201b A / D converter 3 Display device

Claims (4)

An audio comparison device that determines whether or not a first audio signal and a second audio signal, which are demodulated signals when modulated by different modulation schemes, are the same material,
Synchronizing means for detecting a time difference at which a cross-correlation function between the first audio signal and the second audio signal is maximized, and synchronizing the first audio signal and the second audio signal according to the time difference. When,
Frame acquisition means for multiplying each of the first audio signal and the second audio signal synchronized by the synchronization means by a window function of a predetermined time width and acquiring a frame corresponding to the time width; ,
Frequency conversion means for converting each of the first audio signal and the second audio signal input as frames corresponding to a predetermined time width by the frame acquisition means into a frequency domain by Fourier transformation;
Spectrum envelope conversion means for converting each of the first audio signal and the second audio signal input after being converted into a frequency domain by the frequency conversion means into a spectrum envelope;
For each of the first audio signal and the second audio signal input after being converted into a spectrum envelope by the spectrum envelope conversion means, a signal in a frequency band lower than a predetermined frequency is removed, thereby High-pass filter means for performing processing to extract a signal in a frequency band above the frequency;
A Mel scale is used for each of the first audio signal and the second audio signal that are input by processing to extract a signal in a frequency band equal to or higher than the predetermined frequency by the high-pass filter means. Mel scale conversion means for performing weighting on the frequency corresponding to human auditory sensitivity characteristics;
Normalization means for performing a normalization process on each of the first sound signal and the second sound signal input by weighting on the frequency by the mel scale conversion means ;
Distance calculating means for calculating a distance between the first audio signal and the second audio signal normalized by the normalizing means;
Comparing means for judging whether or not the value of the distance, which is a calculated value of the distance calculating means, is equal to or greater than a predetermined reference value, and outputting the result. The distance calculation means calculates the square of the absolute value of the difference between the first audio signal and the second audio signal at a predetermined sample frequency, and further calculates the absolute value of the difference calculated at each sample frequency. The speech comparison apparatus according to claim 1, wherein the square root of the sum of squares is calculated as a distance. Mel scale function used in the mel scale conversion unit, when the frequency of the audio signal is f, claim 1 or claim, characterized by being represented by Mel (f) = 2595log10 (1 + f / 700) 2. The voice comparison device according to 2. In order to determine whether the first audio signal and the second audio signal, which are demodulated signals when modulated by different modulation schemes, are the same material,
Synchronizing means for detecting a time difference at which a cross-correlation function between the first audio signal and the second audio signal is maximized, and synchronizing the first audio signal and the second audio signal according to the time difference. ,
Frame acquisition means for multiplying each of the first audio signal and the second audio signal synchronized by the synchronization means by a window function of a predetermined time width, and acquiring a frame corresponding to the time width;
Frequency conversion means for converting each of the first audio signal and the second audio signal input as a frame corresponding to a predetermined time width by the frame acquisition means into a frequency domain by Fourier transformation;
A spectral envelope converting means for converting each of the first audio signal and the second audio signal input after being converted into a frequency domain by the frequency converting means into a spectral envelope;
For each of the first audio signal and the second audio signal input after being converted into a spectrum envelope by the spectrum envelope conversion means, a signal in a frequency band lower than a predetermined frequency is removed, thereby High-pass filter means for performing processing to extract a signal in a frequency band above the frequency,
A Mel scale is used for each of the first audio signal and the second audio signal that are input by processing to extract a signal in a frequency band equal to or higher than the predetermined frequency by the high-pass filter means. Mel scale conversion means for performing weighting on the frequency corresponding to human auditory sensitivity characteristics,
Normalization means for performing a normalization process on each of the first audio signal and the second audio signal input by weighting on the frequency by the mel scale conversion means;
Distance calculating means for calculating a distance between the first audio signal and the second audio signal normalized by the normalizing means;
Comparing means for determining whether or not the distance value, which is a calculated value of the distance calculating means, is equal to or greater than a predetermined reference value, and outputting the result;
Voice comparison program to function as

Images