JP5733044B2 - Masking analysis device, masker sound selection device, masking device and program - Google Patents

Description

  The present invention relates to a technique for evaluating the effect of masking using various masker sounds.

  Conventionally, a sound masking technique for preventing leakage of a target sound by superimposing a masker sound on a target sound (maskee) such as a highly confidential conversation sound has been proposed. In addition to various types of noise such as white noise, the sound obtained by processing the target sound is also used as a masker sound. For example, Patent Literature 1 and Patent Literature 2 disclose a technique for generating a masker sound by reversing the time waveform of each section obtained by dividing the target sound on the time axis and changing the order of each section.

  Quantitative evaluation of the masking effect is important for the generation and selection of masker sounds that can effectively prevent voice leakage. A typical method for evaluating the masking effect is a subjective evaluation that measures the rate at which subjects who listened to the masked speech can understand the target sound (speech intelligibility). Has the problem of being very time consuming. Therefore, in the techniques of Non-Patent Document 1 and Non-Patent Document 2, the correlation value of the narrowband envelope of the speech before and after masking (hereinafter referred to as “narrowband envelope correlation”) is adopted as a quantitative evaluation index of the effect of masking. Is done. The narrow-band envelope is an envelope of a speech waveform in each band (for example, a quarter octave band) corresponding to the critical band of human hearing.

JP 2008-233671 A JP 2010-217883 A Houtgast T et al. "Predicting speech intelligibility in rooms from the Modulation Transfer Function. I. General room acoustics", Acustica, 46: 60-72, 1980 Drullman R. "Temporal envelope and fine structure cues for speech intelligibility", J. Acoust. Soc. Am 97: 585-592, 1995

  By the way, the action of sound masking includes energy masking and information masking. Energy masking is an action that obstructs listening of a target sound by superimposing a masker sound generated independently of the target sound on the target sound with relatively high energy, and information masking is disclosed in Patent Document 1 described above. As in the technique of Japanese Patent Application Laid-Open No. H11-133260, the target sound is disturbed by superimposing a masker sound (disturbance sound) whose acoustic characteristics are similar to the target sound on the target sound. A typical example of a masker sound effective for energy masking is white noise, and a typical example of a masker sound effective for information masking is an inverted voice obtained by inverting the voice waveform of the speaker of the target sound in the direction of the time axis.

  FIG. 11 is a graph showing the relationship between the calculated value of the narrowband envelope correlation and the actually measured value of the intelligibility for a plurality of cases where the energy ratio of the target sound to the masker sound (hereinafter referred to as “T / M ratio”) is different. It is. In FIG. 11, a case where white noise effective for energy masking is used as a masker sound and a case where inverted voice effective for information masking is used as a masker sound are separately illustrated.

  When white noise is used as a masker sound, as shown by the line Z1 in FIG. 11, the conversation intelligibility changes sensitively to changes in the narrowband envelope correlation, and the greater the narrowband envelope correlation, the higher the conversation intelligibility. The trend is noticeable. However, when the reverse speech is used as a masker sound, as shown by the line Z2 in FIG. 11, especially in the range from 0.3 to 0.8 of the narrowband envelope correlation, the conversation with respect to the change of the narrowband envelope correlation is performed. The tendency that the intelligibility does not change clearly is confirmed. That is, the narrowband envelope correlation disclosed in Non-Patent Document 1 and Non-Patent Document 2 is appropriate as an evaluation index for energy masking, but is not necessarily appropriate as an evaluation index for information masking. In view of the above circumstances, an object of the present invention is to appropriately evaluate the effect of information masking.

  Means employed by the present invention to solve the above problems will be described. In order to facilitate the understanding of the present invention, in the following description, the correspondence between the elements of the present invention and the elements of the embodiments described later will be indicated in parentheses, but the scope of the present invention will be exemplified in the embodiments. It is not intended to be limited.

  The masking analysis apparatus of the present invention is an apparatus for analyzing masking of a target sound by a masker sound, and an autocorrelation sequence of a line spectrum sequence (for example, a line spectrum sequence Li [m]) corresponding to each peak of a spectrum of an acoustic signal. (For example, autocorrelation sequence Ai [m]), a first acoustic signal (for example, acoustic signal s1 (t)) indicating the target sound and a second acoustic signal (for example, acoustic signal s2) indicating the mixed sound of the target sound and masker sound. (t)) and autocorrelation calculating means (for example, autocorrelation calculating unit 22) for each frame on the time axis, and first for each frame corresponding to each other in the first acoustic signal and the second acoustic signal. Cross-correlation calculating means (cross-correlation calculating unit 24) for calculating a cross-correlation coefficient (for example, cross-correlation coefficient ρ [m]) between the autocorrelation sequence of the acoustic signal and the autocorrelation sequence of the second acoustic signal; Comprising the index calculation means for calculating a calculated representative value of the plurality of cross-correlation coefficients as effect index for masking (e.g., effect index alpha) (e.g. the index calculator 26) for beam. In the above configuration, since the effect index is calculated according to the cross-correlation coefficient between frames corresponding to each other in the autocorrelation sequence of the first acoustic signal and the autocorrelation sequence of the second acoustic signal, the narrowband envelope correlation is calculated. It is possible to appropriately evaluate the effect of information masking compared with the case of using.

  The type of representative value calculated as an effect index from a plurality of cross-correlation coefficients is arbitrary. For example, a configuration for calculating an average value (for example, arithmetic mean) of a plurality of cross-correlation coefficients as an effect index, A configuration in which a predetermined percentile value (for example, 75th percentile value) of the cross-correlation coefficient is calculated as an effect index is suitable. The method of using the effect index calculated by the index calculating means is arbitrary in the present invention. For example, a configuration including display control means (for example, the display control unit 28) for displaying the effect index on the display device is suitable. is there.

  In a preferred aspect of the present invention, the autocorrelation calculating means differs in at least one of the autocorrelation sequence of the first acoustic signal and the type of masker sound and the energy ratio (T / M ratio) of the target sound and masker sound. The autocorrelation number sequence of each of the plurality of second acoustic signals is calculated for each frame, and the cross-correlation calculating means calculates the autocorrelation number sequence of the first acoustic signal and the second acoustic signal for each of the plurality of second acoustic signals. The cross-correlation coefficient with the autocorrelation sequence is calculated for each frame, and the index calculation means has, for each of the plurality of second acoustic signals, an effect corresponding to the plurality of cross-correlation coefficients calculated for the second acoustic signal. Calculate the indicators. In the above aspect, since the effect index is calculated for a plurality of masker sounds of different types and sound pressures, it is optimal from the viewpoint of the effectiveness of information masking by comparing each effect index of the plurality of masker sounds. A masker sound can be selected.

  A masking analysis apparatus according to a preferred aspect of the present invention includes a correlation transition image (for example, correlation transition image 62) indicating a time series of an autocorrelation sequence in a region in which a frequency axis and a time axis are set, and each autocorrelation sequence. At least one of a correlation distribution image (for example, a correlation distribution image 64) indicating a distribution on the frequency axis of a numerical value obtained by summing up correlation values for each frequency for a plurality of frames is used as each of the first acoustic signal and the second acoustic signal. Display control means (for example, the display control unit 28) for displaying on the display device. In the above aspect, by comparing the correlation transition image between the first acoustic signal and the second acoustic signal, the user can adjust the degree to which the temporal transition of the harmonic structure changes before and after masking (that is, information masking). It is possible to grasp the degree) intuitively. Also, by comparing the correlation distribution image between the first acoustic signal and the second acoustic signal, the user can intuitively grasp the long-term change in the harmonic structure over a plurality of frames. is there.

  The present invention is also realized as a masker sound selection device that selects any one of a plurality of types of masker sounds using the masking analysis device according to each of the above aspects. The masker sound selection apparatus of the present invention uses the autocorrelation sequence of the line spectrum sequence corresponding to each peak of the spectrum of the acoustic signal, the first acoustic signal indicating the target sound, and the mixture of the different types of masker sound and target sound. Autocorrelation calculating means for calculating each frame on the time axis for each of the plurality of second acoustic signals indicating sound, and for each of the plurality of second acoustic signals, the autocorrelation number sequence of the second acoustic signal and the second A cross-correlation calculating means for calculating a cross-correlation coefficient with an autocorrelation sequence of one acoustic signal for each corresponding frame; and for each of a plurality of second acoustic signals, a calculation is performed for each frame of the second acoustic signal. The index calculation means for calculating the representative value of multiple cross-correlation coefficients as the masking effect index, and one of multiple types of masker sound is selected according to the effect index calculated by the index calculation means. To that and a selection means (e.g., selecting section 40). Even with the above configuration, the same operation and effect as the masking analysis apparatus of the present invention can be realized.

  The present invention is also realized as a masking device (for example, masking device 200) that masks a target sound using any one of a plurality of types of masker sounds. The masking device of the present invention uses a first acoustic signal indicating a target sound, a mixed sound of different types of masker sound and target sound, an autocorrelation number sequence of a line spectrum sequence corresponding to each peak of the spectrum of the acoustic signal. Autocorrelation calculating means for calculating each of the plurality of second acoustic signals shown for each frame on the time axis, and for each of the plurality of second acoustic signals, the autocorrelation sequence of the second acoustic signal and the first acoustic A cross-correlation calculating means for calculating a cross-correlation coefficient with the autocorrelation sequence of the signal for each frame corresponding to each other, and a plurality of the plurality of second acoustic signals calculated for each frame of the second acoustic signal An index calculation means for calculating a representative value of the cross-correlation coefficient as an effect index for masking, and selecting one of multiple types of masker sounds according to the effect index calculated by the index calculation means Selecting means for sound from the sound emitting device (for example, the selection unit 40); and a. Even with the above configuration, the same operation and effect as the masking analysis apparatus of the present invention can be realized.

  The masking analysis apparatus according to each aspect described above is realized by hardware (electronic circuit) such as a DSP (Digital Signal Processor) dedicated to speech synthesis, and general-purpose arithmetic processing such as a CPU (Central Processing Unit). It is also realized by cooperation between the device and the program. In order to analyze masking of a target sound by a masker sound, the program of the present invention uses an autocorrelation sequence of a line spectrum sequence corresponding to each peak of a spectrum of an acoustic signal, a first acoustic signal indicating the target sound, and a target sound. And autocorrelation calculation processing for each frame for each of the second acoustic signal indicating the mixed sound of the masker sound, and the self of the first acoustic signal for each frame corresponding to each other in the first acoustic signal and the second acoustic signal. Cross correlation calculation processing for calculating the cross correlation coefficient between the correlation number sequence and the autocorrelation number sequence of the second acoustic signal, and representative values of a plurality of cross correlation coefficients calculated for each frame are used to mask the target sound by masker sound. The computer executes an index calculation process for calculating as an effect index of the computer. According to the above program, the same operation and effect as the masking analysis apparatus of the present invention are realized. The program of the present invention is provided to a user in a form stored in a computer-readable recording medium and installed in the computer, or provided from a server device in a form of distribution via a communication network and installed in the computer. Is done.

1 is a block diagram of a masking analysis apparatus according to a first embodiment of the present invention. It is a block diagram of an autocorrelation calculation part. It is explanatory drawing of operation | movement of a masking analyzer. It is a flowchart of the operation | movement which produces | generates a line spectrum sequence. It is a graph which shows the relationship between the effect index of 1st Embodiment, and the measured value of conversation intelligibility. It is a block diagram of the masking analysis apparatus in 2nd Embodiment. It is a graph which shows the relationship between the effect index of 3rd Embodiment, and the measured value of conversation intelligibility. It is a schematic diagram which shows the example of a display in 3rd Embodiment. It is a schematic diagram which shows the example of a display in 4th Embodiment. It is a block diagram of the masking apparatus which concerns on 5th Embodiment. It is a graph which shows the relationship between the calculated value of a narrow-band envelope correlation, and the measured value of conversation intelligibility.

<First Embodiment>
FIG. 1 is a block diagram of a masking analysis apparatus 100 according to the first embodiment of the present invention. The masking analysis device 100 is an acoustic processing device that analyzes the effect of masking the target sound VT using the masker sound VM, and includes an arithmetic processing device 12, a storage device 14, and a display device 16, as shown in FIG. Realized in a computer system. The display device 16 is composed of a liquid crystal display panel, for example, and displays an image instructed from the arithmetic processing device 12.

  The storage device 14 stores a program PGM executed by the arithmetic processing device 12 and various data used by the arithmetic processing device 12. For example, a known recording medium such as a semiconductor recording medium or a magnetic recording medium or a combination of a plurality of types of recording media may be employed as the storage device 14.

  The storage device 14 stores an acoustic signal s1 (t) and an acoustic signal s2 (t). The acoustic signal s1 (t) is an audio signal indicating a time waveform of the target sound VT to be masked. On the other hand, the acoustic signal s2 (t) is a signal indicating a time waveform of a sound obtained by superimposing (adding) the masker sound VM on the target sound VT indicated by the acoustic signal s1 (t) (that is, a signal after masking). That is, the acoustic signal s1 (t) corresponds to the sound before masking. For example, an acoustic signal s1 (t) and an acoustic signal s2 (t) recorded in advance using a sound collection device are stored in the storage device 14. Note that the sound signal collected by the sound collecting device is acquired sequentially (for example, for each section of a predetermined time length) as the acoustic signal s1 (t) or the acoustic signal s2 (t) and processed in substantially real time. It is also possible.

  The arithmetic processing unit 12 calculates and outputs an index value (hereinafter referred to as “effect index”) α indicating the effect of masking by the masker sound VM by executing the program PGM stored in the storage device 14. Are realized (autocorrelation calculation unit 22, cross-correlation calculation unit 24, index calculation unit 26, display control unit 28). The effect index α is a suitable variable as an index of the effectiveness of the masker sound VM for information masking, and is roughly calculated by comparing the acoustic signal s1 (t) and the acoustic signal s2 (t) before and after masking. Is done. A configuration in which a dedicated electronic circuit (DSP) realizes a part of the functions of the arithmetic processing device 12 or a configuration in which the functions of the arithmetic processing device 12 are distributed over a plurality of integrated circuits may be employed.

  By the way, the effect of the information masking is remarkable when the reverse voice obtained by reversing the voice waveform of the same speaker as the target sound VT in the time axis direction is applied as the masker sound VM. Since the reverse voice and the target sound VT are the same for the speaker, the long-term harmonic structure of the voice before and after masking using the reverse voice as the masker sound VM (sequence of fundamental and multiple harmonic components) Hardly changes. Considering the above tendency, it is surmised that the action of information masking is related to the time transition of the harmonic structure being different before and after masking. That is, the effect of information masking is greater as the time transition of the harmonic structure changes before and after masking. From the above knowledge, in this embodiment, the effect index α is calculated by comparing the time transition of the harmonic structure of the acoustic signal s1 (t) with the time transition of the harmonic structure of the acoustic signal s2 (t). To do.

  The autocorrelation calculation unit 22 in FIG. 1 performs the autocorrelation sequence A1 [m] (A1 [1] to A1 [M]) of the acoustic signal s1 (t) and the acoustic signal for each of M frames having a predetermined time length. The autocorrelation sequence A2 [m] (A2 [1] to A2 [M]) of s2 (t) is calculated (m = 1 to M). The autocorrelation sequence A1 [m] is a numeric sequence reflecting the harmonic structure in the mth frame of the acoustic signal s1 (t), and the autocorrelation sequence A2 [m] is the acoustic signal s2 (t). Is a numerical string reflecting the harmonic structure in the m-th frame. The autocorrelation calculation unit 22 performs the same processing for each of the acoustic signal s1 (t) and the acoustic signal s2 (t). Therefore, in the following description, each of the acoustic signal s1 (t) and the acoustic signal s2 (t) is represented as an acoustic signal si (t) for convenience by the suffix i (i = 1, 2), and the acoustic signal s1. Matters common to both (t) and the acoustic signal s2 (t) will be comprehensively described.

  FIG. 2 is a detailed block diagram of the autocorrelation calculation unit 22. As shown in FIG. 2, the autocorrelation calculation unit 22 includes a section setting unit 32, a frequency analysis unit 34, and a correlation analysis unit 36. The section setting unit 32 multiplies the acoustic signal si (t) by a predetermined time window, so that the acoustic signal si (t) is converted into M section signals qi corresponding to different frames as shown in FIG. [m] (qi [1] to qi [M]). Each frame is set to a time length of about 20 milliseconds to 30 milliseconds, for example, and overlaps each other on the time axis. Note that the time length of each frame can be variably controlled in accordance with, for example, the fundamental frequency of the acoustic signal si (t).

  The frequency analysis unit 34 in FIG. 2 performs a line spectrum sequence Li [m] (Li [1] to Li [M] corresponding to each peak of the spectrum Qi [m] of the section signal qi [m] for each of M frames. ]). As shown in FIG. 2, the line spectrum sequence Li [m] is arranged at each of the LN frequencies Fp at which the amplitude value (absolute value) of the spectrum Qi [m] of the section signal qi [m] peaks. It is a series of spectral lines whose intensity is normalized to a predetermined value (1).

  FIG. 4 is a flowchart of processing in which the frequency analysis unit 34 generates a line spectrum sequence Li [m] for the mth frame (section signal qi [m]) of the acoustic signal si (t). The process shown in FIG. 4 is executed for each of the M section signals qi [1] to qi [M] of each acoustic signal si (t).

  The frequency analysis unit 34 initializes a variable x indicating one spectral line to 1 (SA1), and determines whether the variable x falls below a predetermined value LN (SA2). At the stage immediately after the start of the process of FIG. 4, the variable x is below the predetermined value LN. When the variable x falls below the predetermined value LN, the frequency analysis unit 34 calculates the spectrum (complex spectrum) Qi [m] of the section signal qi [m] (SA3). For calculation of the spectrum Qi [m], a known frequency analysis such as discrete Fourier transform is arbitrarily employed.

  The frequency analysis unit 34 specifies and stores the frequency Fp of one peak having the maximum amplitude value in the amplitude spectrum | Qi [m] | of the spectrum Qi [m] calculated in step SA3 (SA4), and step SA3. A spectrum Ri [m] in which the intensity of each frequency other than the frequency Fp specified in step SA4 is set to zero among the spectrum Qi [m] calculated in step S4 is generated (SA5). Then, the frequency analysis unit 34 converts the spectrum Ri [m] into an acoustic signal ri [m] in the time domain by, for example, inverse Fourier transform (SA6), and the converted acoustic signal ri [m] is the current section signal. Subtract from qi [m] (SA7).

  The frequency analysis unit 34 adds 1 to the variable x and proceeds to step SA2 (SA8). If the variable x after addition still falls below the predetermined value LN (SA2: YES), the immediately preceding step The processing from step SA3 to step SA8 is repeated for the section signal qi [m] after the processing at SA7. That is, the amplitude of the spectrum Qi [m] is sequentially excluded while the acoustic components of the frequency Fp are sequentially excluded from the section signal qi [m] until the total number of frequencies Fp specified for the section signal qi [m] reaches a predetermined value LN. The process of specifying the peak frequency Fp of the value is repeated.

  When the total number of the frequencies Fp reaches the predetermined value LN (SA2: NO), the frequency analyzing unit 34 selects the section signal qi [in step SA4 among the K frequencies (frequency bands) discretely set on the frequency axis. A line spectrum sequence Li [m] in which a spectrum line normalized to an intensity of 1 is set to each of the LN frequencies Fp specified for m] is generated (SA9). Among the K frequencies, the intensity of each frequency other than LN frequencies Fp is set to zero. The above is the calculation method of the line spectrum sequence Li [m]. For example, Y. Hara, M. Matsumoto, and K. Miyoshi, “Method for controlling pitch independently from power spectrum envelope for speech and music signal”, J. Temporal Design in Architecuture. and the Environment 9 (1) 121-124 (2009).

  As shown in FIG. 3, the correlation analysis unit 36 in FIG. 2 uses the autocorrelation number sequence Ai [m] (for the line spectrum sequence Li [m] generated by the frequency analysis unit 34 for each frame of each acoustic signal si (t). Ai [1] to Ai [M]) are calculated. The autocorrelation sequence (autocorrelation function) Ai [m] is an autocorrelation coefficient pi [m, k] (pi [m, 1] to pi [m, K) corresponding to each of the K frequencies on the frequency axis. ]) Series (Kth order vector).

  Since the line spectrum sequence Li [m] generated by the frequency analysis unit 34 is composed of spectral lines arranged at the respective frequencies Fp at which the amplitude value peaks in the section signal qi [m], the line spectrum sequence Li [m]. ] Of the autocorrelation sequence Ai [m] approximates a spectrum that emphasizes the harmonic structure in each frame of the acoustic signal si (t). That is, peaks appear at intervals corresponding to the fundamental frequency of the acoustic signal si (t) in the autocorrelation coefficient pi [m, 1] to pi [m, K] series of the autocorrelation sequence Ai [m]. . For each of the acoustic signal s1 (t) and the acoustic signal s2 (t), the above processing is executed for each frame (for each section signal qi [m]), thereby corresponding to each frame of the acoustic signal s1 (t). M autocorrelation sequence A1 [1] to A1 [M] and M autocorrelation sequence A2 [1] to A2 [M] corresponding to each frame of the acoustic signal s2 (t) are generated.

As shown in FIG. 3, the cross-correlation calculating unit 24 in FIG. 1 performs the acoustic signal s1 (t) between frames corresponding to each other on the time axis in the acoustic signal s1 (t) and the acoustic signal s2 (t). The cross-correlation coefficient ρ [m] (ρ [1] to ρ [M]) between the autocorrelation sequence A1 [m] and the autocorrelation sequence A2 [m] of the acoustic signal s2 (t) Calculate for each. The cross-correlation coefficient ρ [m] is an autocorrelation sequence A1 [m] of the mth frame of the acoustic signal s1 (t) and an autocorrelation sequence A2 [m] of the mth frame of the acoustic signal s2 (t). ] (Ie, the degree of temporal transition similarity of the harmonic structure between the acoustic signal s1 (t) and the acoustic signal s2 (t)). The cross-correlation calculating unit 24 calculates the cross-correlation coefficient ρ [m] by, for example, the following equation (1).
The operator E {} in Equation (1) means an average (typically an arithmetic average) over K frequencies on the frequency axis. Further, the symbol δi [m, k] (δ1 [m, k], δ2 [m, k]) in the equation (1) means a deviation calculated by the following equation (2). The symbol .mu.i (.mu.1, .mu.2) in Equation (2) is the autocorrelation coefficient pi [m, k] (pi [m, 1] to pi [m, K] over K frequencies in the mth frame. Series).

In addition, the symbol Pi [m] (P1 [m], P2 [m]) in Equation (1) is defined as K deviations δi corresponding to the mth frame, as defined by Equation (3) below. [m, k] (Δi [m, 1] to Δi [m, K]) is an average of squares, and is calculated for each frame by the calculation of Equation (3). Accordingly, the lower the correlation between the autocorrelation sequence A1 [m] and the autocorrelation sequence A2 [m], the smaller the cross-correlation coefficient ρ [m] in Equation (1).

  The index calculation unit 26 of FIG. 1 calculates the masking effect index α using the masker sound VM using the M cross-correlation coefficients ρ [1] to ρ [M] calculated by the cross-correlation calculation unit 24. The index calculation unit 26 of the first embodiment calculates an average (for example, an arithmetic average) of M cross-correlation coefficients ρ [1] to ρ [M] as an effect index α. Therefore, generally speaking, the lower the correlation of the time transition of the harmonic structure between the acoustic signal s1 (t) and the acoustic signal s2 (t) (before and after masking) (that is, the higher the effect of information masking), the more effective it is. The index α tends to be a small numerical value. The display control unit 28 causes the display device 16 to display the effect index α calculated by the index calculation unit 26.

  FIG. 5 shows the effect index α calculated in the first embodiment and the conversation understanding for a plurality of cases where the energy ratio of the target sound VT to the masker sound VM (hereinafter referred to as “T / M ratio”) is changed. It is a graph which shows the relationship with the measured value of a degree. Similarly to FIG. 11, the case where the white noise is used as the masker sound VM and the case where the reverse sound is used as the masker sound VM are shown in FIG. Each numerical value is connected in ascending order of T / M ratio.

  As can be understood from FIG. 5, when white noise effective for energy masking is used as the masker sound VM, the change of the speech intelligibility with respect to the change of the effect index α is slow as indicated by the line Z1 in FIG. On the other hand, when the reverse voice is used as the masker sound VM, as shown by the line Z2 in FIG. 5, the speech intelligibility changes sensitively with respect to the change of the effect index α over the entire numerical range of the effect index α. That is, for the reverse voice effective for information masking, the tendency that the greater the effect index α is, the higher the degree of intelligibility of the conversation is noticeable. From the above tendency, it is understood that the effect index α of the first embodiment is appropriate as a quantitative evaluation index of information masking when compared with the narrowband envelope correlation of Non-Patent Document 1 and Non-Patent Document 2. . That is, according to the first embodiment, referring to the effect index α displayed on the display device 16, there is an advantage that the user can appropriately evaluate the effect of information masking using the masker sound VM.

Second Embodiment
A second embodiment of the present invention will be described below. In addition, about the element in which an effect | action and a function are equivalent to 1st Embodiment in each form illustrated below, the code | symbol referred by description of 1st Embodiment is diverted, and each detailed description is abbreviate | omitted suitably.

  FIG. 6 is a block diagram of the masking analysis apparatus 100 according to the second embodiment. As shown in FIG. 6, the storage device 14 of the second embodiment has a mixed sound (masking) of the target sound VT and the masker sound VM in addition to the acoustic signal s1 (t) indicating the target sound VT (sound before masking). Two kinds of acoustic signals s2 (t) (s2 (t) _A, s2 (t) _B) indicating the later voice) are stored. The type (generation method) of the masker sound VM_A of the acoustic signal s2 (t) _A and the masker sound VM_B of the acoustic signal s2 (t) _B are different. For example, the masker sound VM_A of the acoustic signal s2 (t) _A is a reverse sound, and the masker sound VM_B of the acoustic signal s2 (t) _B is white noise.

  In the second embodiment, an effect index αA between the acoustic signal s1 (t) and the acoustic signal s2 (t) _A, and an effect index αB between the acoustic signal s1 (t) and the acoustic signal s2 (t) _B. And are calculated separately. Specifically, the autocorrelation calculation unit 22 calculates the autocorrelation sequence Ai [m] for each of the acoustic signal s1 (t), the acoustic signal s2 (t) _A, and the acoustic signal s2 (t) _B, and The correlation calculation unit 24 performs cross-correlation coefficients ρA [1] to ρA [M] between the acoustic signal s1 (t) and the acoustic signal s2 (t) _A, the acoustic signal s1 (t), and the acoustic signal s2 (t ) _B, the cross-correlation coefficients ρB [1] to ρB [M] are calculated. The index calculator 26 calculates the effect index αA of the masker sound VM_A from the M cross-correlation coefficients ρA [1] to ρA [M], and M cross-correlation coefficients ρB [1] to ρB [M]. To calculate the effect index αB of the masker sound VM_B. The display control unit 28 causes the display device 16 to display the effect index αA and the effect index αB.

  In the second embodiment, the same effect as in the first embodiment is realized. In the second embodiment, since the effect index αA of the masker sound VM_A and the effect index αB of the masker sound VM_B are calculated and displayed separately, the masker sound effective for information masking among the masker sound VM_A and the masker sound VM_B. There is an advantage that the user can easily confirm VM (a masker sound having a small effect index α).

<Third Embodiment>
The storage device 14 of the third embodiment stores the acoustic signal s1 (t), the acoustic signal s2 (t) _A, and the acoustic signal s2 (t) _B, as in the second embodiment. The cross-correlation calculating unit 24 includes M cross-correlation coefficients ρA [1] to ρA [M] between the acoustic signal s1 (t) and the acoustic signal s2 (t) _A, and the acoustic signal s1 (t). M cross-correlation coefficients ρB [1] to ρB [M] with the acoustic signal s2 (t) _B are calculated by the same method as in the first embodiment.

  The index calculation unit 26 of the third embodiment sets the σ percentile value of the M cross-correlation coefficients ρA [1] to ρA [M] calculated by the cross-correlation calculation unit 24 (in the following example, the variable σ is set to 75 (% ) Is calculated as an effect index αA of the masker sound VM_A. That is, when M cross-correlation coefficients ρA [1] to ρA [M] are arranged in ascending order, the number-th cross-correlation coefficient ρA [m] corresponding to M σ% is calculated as the effect index αA. Is done. Similarly, the index calculation unit 26 calculates the σ percentile value of the M cross-correlation coefficients ρB [1] to ρB [M] as the effect index αB of the masker sound VM_B. The display control unit 28 causes the display device 16 to display the effect index αA and the effect index αB calculated by the index calculation unit 26.

  FIG. 7 is a graph showing the relationship between the effect index α (αA, αB) calculated in the third embodiment and the measured value of the intelligibility in the same manner as in FIG. Similar to the effect index α of the first embodiment, the effect index α (σ percentile value) of the third embodiment is based on the change of the effect index α when the reverse voice effective for information masking is a masker sound VM. There is a tendency that conversation intelligibility changes sensitively. Therefore, the third embodiment also has an advantage that the effect of information masking can be appropriately evaluated as in the first embodiment.

  The display control unit 28 of the third embodiment calculates the frequency for each predetermined class for the M cross-correlation coefficients ρA [1] to ρA [M] calculated by the cross-correlation calculation unit 24 and M mutual values. The frequencies are similarly calculated for the correlation coefficients ρB [1] to ρB [M], and as shown in FIG. 8, the frequency distribution 50A of the cross-correlation coefficients ρA [1] to ρA [M] and the cross-correlation coefficient The frequency distribution 50B of ρB [1] to ρB [M] is displayed on the display device 16.

  The user can intuitively compare the effect of information masking by each of the masker sound VM_A and the masker sound VM_B by comparing the frequency distribution 50A and the frequency distribution 50B displayed on the display device 16. . For example, the lower the correlation of the autocorrelation sequence Ai [m] between the acoustic signal s1 (t) and the acoustic signal s2 (t) (that is, the higher the effect of information masking by the masker sound VM), the cross-correlation coefficient ρ [m ] Tend to be unevenly distributed in a small range. When the frequency distribution 50A and the frequency distribution 50B in FIG. 8 are compared, the frequency distribution 50A has a more prominent tendency that the frequencies are unevenly distributed in a small numerical value range. Therefore, the user can intuitively determine that the masker sound VM_A (reverse sound) corresponding to the frequency distribution 50A is more effective for information masking than the masker sound VM_B (white noise).

  In addition, as shown in FIG. 8, the display control unit 28 of the third embodiment has a cumulative frequency distribution 52A of cross-correlation coefficients ρA [1] to ρA [M] and cross-correlation coefficients ρB [1] to ρB [ The cumulative frequency distribution 52B of M] is displayed on the display device 16, and a straight line 54 indicating the frequency corresponding to M σ% is arranged so as to overlap the cumulative frequency distribution 52A and the cumulative frequency distribution 52B. The class value corresponding to the intersection CA of the cumulative frequency distribution 52A and the straight line 54 corresponds to the effect index αA (that is, the σ percentile value of the cross-correlation coefficient ρA [m]), and the intersection CB of the cumulative frequency distribution 52B and the straight line 54 The class value corresponding to is equivalent to the effect index αB.

  The user can intuitively grasp the effect of information masking by the masker sound VM_A and the masker sound VM_B by comparing the cumulative frequency distribution 52A and the cumulative frequency distribution 52B. For example, since the intersection CA corresponds to a small class value compared to the intersection CB, the tendency that the frequency is unevenly distributed in a small numerical value range is the cross-correlation coefficient ρA [m] rather than the cross-correlation coefficient ρB [m]. Is more prominent. Therefore, the user can intuitively determine that the masker sound VM_A (reverse sound) corresponding to the cumulative frequency distribution 52A is more effective for information masking than the masker sound VM_B (white noise).

<Fourth embodiment>
The storage device 14 of the fourth embodiment stores the acoustic signal s1 (t), the acoustic signal s2 (t) _A, and the acoustic signal s2 (t) _B, as in the second embodiment and the third embodiment. As shown in FIG. 9, the display control unit 28 generates a correlation transition image 62 and a correlation distribution image 64 for each of the acoustic signal s1 (t), the acoustic signal s2 (t) _A, and the acoustic signal s2 (t) _B. It is displayed on the display device 16.

  The correlation transition image 62 is an image representing a time series of autocorrelation sequence Ai [m] (that is, time transition of a harmonic structure) in a region in which a frequency axis (vertical axis) and a time axis (horizontal axis) are set. It is. The numerical value of each autocorrelation coefficient pi [m, k] in the autocorrelation sequence Ai [m] is expressed by the gradation and color of each point in the correlation transition image 62. That is, the gradation and color of the coordinate point corresponding to the point corresponding to the mth frame on the time axis and the point corresponding to the kth frequency on the frequency axis are the autocorrelation of the mth frame. Numerical value of the autocorrelation coefficient pi [m, k] corresponding to the kth frequency among the K autocorrelation coefficients pi [m, 1] to pi [m, K] constituting the sequence Ai [m]. It is decided according to.

  The correlation distribution image 64 is an image showing the total value (cumulative frequency) of the autocorrelation coefficients pi [1, k] to pi [M, k] over M frames on the frequency axis (vertical axis). That is, the correlation distribution image 64 is configured to include straight lines corresponding to each of K frequencies on the frequency axis, and the total length of one straight line corresponding to the kth frequency is the self-phase relationship of the frequencies. The number pi [m, k] is selected in accordance with a numerical value (pi [1, k] + pi [2, k] +... + Pi [M, k]) obtained by summing up the M frames.

  As the time series of the autocorrelation sequence Ai [m] (time transition of the harmonic structure) changes before and after masking, the information masking effect tends to increase. Therefore, the user compares the correlation transition image 62 of each of the acoustic signal s2 (t) _A and the acoustic signal s2 (t) _B with the correlation transition image 62 of the acoustic signal s1 (t), so that the masker sound VM_A and It is possible to visually grasp the effect of information masking by each of the masker sounds VM_B (the time series difference of the autocorrelation sequence Ai [m]). For example, in the illustration of FIG. 9, the correlation transition image 62 of the acoustic signal s2 (t) _A is compared with the correlation transition image 62 of the acoustic signal s1 (t) as compared to the correlation transition image 62 of the acoustic signal s2 (t) _B. The difference is great. That is, the acoustic signal s2 (t) _A is more prominent than the acoustic signal s2 (t) _B to the extent that the time series of the autocorrelation sequence Ai [m] changes before and after masking. Therefore, the user can intuitively determine that the masker sound VM_A (reverse sound) is more effective for information masking than the masker sound VM_B (white noise). Further, by comparing the correlation distribution image 64 of each of the acoustic signal s2 (t) _A and the acoustic signal s2 (t) _B with the correlation distribution image 64 of the acoustic signal s1 (t), the user can obtain M frames. It is possible to intuitively grasp the long-term change in the harmonic structure over a case where the masker sound VM_A is used and a case where the masker sound VM_B is used.

<Fifth Embodiment>
FIG. 10 is a block diagram of a masking apparatus 200 according to the fifth embodiment of the present invention. The masking device 200 of the fifth embodiment is a device that selects and emits a plurality of types of masker sounds VM (VM_A, VM_B) having different generation methods and magnitudes (sound pressures). This is a configuration in which a selection unit 40 and a sound emitting device 42 are added to the masking analysis device 100 of the embodiment or the third embodiment. The storage device 14 stores a masker sound signal v (t) _A indicating the sound waveform of the masker sound VM_A and a masker sound signal v (t) _B indicating the sound waveform of the masker sound VM_B.

  The index calculation unit 26 of the fifth embodiment calculates the effect index αA of the masker sound VM_A and the effect index αB of the masker sound VM_B, as in the second or third embodiment. The selection unit 40 selects either the masker sound VM_A or the masker sound VM_B according to the effect index α (αA, αB) calculated by the index calculation unit 26. Specifically, the selection unit 40 selects a masker sound VM having a small effect index αA (that is, a masker sound VM effective for information masking). Then, the selection unit 40 acquires the masker sound signal v (t) (v (t) _A, v (t) _B) corresponding to the masker sound VM selected according to the effect index α from the storage device 14 and releases it. The sound device 42 is supplied. The sound emitting device 42 (for example, a speaker device) radiates a masker sound VM (VM_A, VM_B) as a sound wave according to the masker sound signal v (t) supplied from the selection unit 40.

  In the fifth embodiment, the same effect as in the first embodiment is realized. In the fifth embodiment, since the masker sound VM effective for information masking is automatically selected and emitted according to the effect index α, the burden on the user who tries to mask the target sound VT is reduced. Is possible.

  In the fifth embodiment, the display control unit 28 and the display device 16 can be omitted. A configuration in which the selection unit 40 selects a masker sound VM corresponding to the effect index α and notifies the user by, for example, display on the display device 16 (that is, a masker sound selection device that does not require the sound emission of the masker sound VM) is also adopted. Can be done. In the above description, the case where one of the two types of masker sounds VM (VM_A, VM_B) is selected has been exemplified, but the number of types of masker sounds VM as selection candidates is arbitrary.

<Modification>
Each of the above forms can be variously modified. Specific modifications are exemplified below. Two or more modes arbitrarily selected from the following examples can be appropriately combined.

(1) In each embodiment described above, the effect index α (αA, αB) is displayed on the display device 16, but the method of using the effect index α is not limited to display to the user. Specifically, in addition to the configuration in which the effect index α is applied to the selection of the masker sound VM as in the fifth embodiment, the configuration in which the effect index α calculated by the index calculation unit 26 is output by voice and the effect index α A configuration for printing on a network or a configuration for transmitting to another communication terminal via a communication network is also employed.

(2) In the second to fifth embodiments, two types of acoustic signals s2 (t) (s2 (t) _A, s2 (t) _B) are illustrated, but three or more types of acoustic signals s2 (t) Even if the configuration is prepared, it is possible to evaluate the effect of the information masking by the masker sound VM of each acoustic signal s2 (t) by executing the same processing as in each of the above embodiments for each acoustic signal s2 (t). It is.

(3) In the second to fifth embodiments, the type of masker sound VM is different between the acoustic signal s2 (t) _A and the acoustic signal s2 (t) _B, but the acoustic signal s2 (t) _A A configuration in which the T / M ratio is different from that of the acoustic signal s2 (t) _B is also employed. For example, when the acoustic signal s2 (t) _A and the acoustic signal s2 (t) _B are generated by applying the same type of masker sound VM to the masking of the target sound VT with different T / M ratios, the same as the above-described embodiments In addition, by calculating and evaluating the effect index α for each acoustic signal s (t), it is possible to specify the optimum T / M ratio from the viewpoint of enabling information masking. That is, it is preferable to calculate an effect index between the acoustic signal s1 (t) and each of the plurality of acoustic signals s2 (t) having different masker sound types and / or T / M ratios.

DESCRIPTION OF SYMBOLS 100 ... Masking analysis apparatus, 200 ... Masking apparatus, 12 ... Arithmetic processing apparatus, 14 ... Memory | storage device, 16 ... Display apparatus, 22 ... Autocorrelation calculation part, 24 ... Cross correlation calculation part, 26 ... ... index calculation part, 28 ... display control part, 32 ... section setting part, 34 ... frequency analysis part, 36 ... correlation analysis part.

Claims (8)

A device that analyzes masking of target sound by masker sound,
The autocorrelation number sequence of the line spectrum sequence corresponding to each peak of the spectrum of the acoustic signal is determined for each of the first acoustic signal indicating the target sound and the second acoustic signal indicating the mixed sound of the target sound and the masker sound. An autocorrelation calculation means for calculating each frame on the time axis;
Cross-correlation calculation for calculating a cross-correlation coefficient between the autocorrelation sequence of the first acoustic signal and the autocorrelation sequence of the second acoustic signal for each frame corresponding to each other in the first acoustic signal and the second acoustic signal. Means,
A masking analysis apparatus comprising: index calculation means for calculating a representative value of a plurality of cross-correlation coefficients calculated for each frame as the masking effect index. The masking analysis apparatus according to claim 1, wherein the index calculation unit calculates an average value of the plurality of cross-correlation coefficients as the effect index. The masking analysis apparatus according to claim 1, wherein the index calculation unit calculates a predetermined percentile value of the plurality of cross-correlation coefficients as the effect index. The autocorrelation calculating means includes an autocorrelation sequence of the first acoustic signal, and an autocorrelation sequence of each of a plurality of second acoustic signals in which at least one of the type of masker sound and the energy ratio of the target sound and the masker sound is different. For each frame,
The cross-correlation calculating means calculates, for each frame, a cross-correlation coefficient between an autocorrelation sequence of the first acoustic signal and an autocorrelation sequence of the second acoustic signal for each of the plurality of second acoustic signals.
The index calculation means calculates, for each of the plurality of second acoustic signals, the effect index according to a plurality of cross-correlation coefficients calculated for the second acoustic signal. Masking analysis device. Correlation transition image showing the time series of the autocorrelation sequence in the area where the frequency axis and time axis are set, and a numerical value on the frequency axis that is the sum of each correlation value of the autocorrelation sequence for each frequency for multiple frames The masking analysis apparatus according to claim 1, further comprising: a display control unit configured to display a correlation distribution image indicating a distribution on a display device for each of the first acoustic signal and the second acoustic signal. .
A masker sound selection device for selecting one of a plurality of masker sounds used for masking a target sound ,
The autocorrelation sequence of the line spectrum sequence corresponding to each peak of the spectrum of the acoustic signal is represented by a plurality of second sounds indicating a mixed sound of the first acoustic signal indicating the target sound, a different type of masker sound, and the target sound. For each of the signals, an autocorrelation calculating means for calculating for each frame on the time axis,
For each of the plurality of second acoustic signals, a cross-correlation calculating means for calculating a cross-correlation coefficient between the autocorrelation sequence of the second acoustic signal and the autocorrelation sequence of the first acoustic signal for each corresponding frame When,
For each of the plurality of second acoustic signals, index calculation means for calculating representative values of a plurality of cross-correlation coefficients calculated for each frame of the second acoustic signal as the masking effect index;
A masker sound selecting device comprising: selecting means for selecting any of the plurality of types of masker sounds according to the effect index calculated by the index calculating means.
A masking device for masking a target sound using any of a plurality of types of masker sounds,
The autocorrelation sequence of the line spectrum sequence corresponding to each peak of the spectrum of the acoustic signal is represented by a plurality of second sounds indicating a mixed sound of the first acoustic signal indicating the target sound, a different type of masker sound, and the target sound. For each of the signals, an autocorrelation calculating means for calculating for each frame on the time axis,
For each of the plurality of second acoustic signals, a cross-correlation calculating means for calculating a cross-correlation coefficient between the autocorrelation sequence of the second acoustic signal and the autocorrelation sequence of the first acoustic signal for each corresponding frame When,
For each of the plurality of second acoustic signals, index calculation means for calculating representative values of a plurality of cross-correlation coefficients calculated for each frame of the second acoustic signal as the masking effect index;
A masking device comprising: selecting means for selecting any one of the plurality of types of masker sounds according to the effect index calculated by the index calculating means and emitting sound from the sound emitting device. To analyze the masking of the target sound by the masker sound,
The autocorrelation number sequence of the line spectrum sequence corresponding to each peak of the spectrum of the acoustic signal is determined for each of the first acoustic signal indicating the target sound and the second acoustic signal indicating the mixed sound of the target sound and the masker sound. Autocorrelation calculation processing for each frame on the time axis,
Cross-correlation calculation for calculating a cross-correlation coefficient between the autocorrelation sequence of the first acoustic signal and the autocorrelation sequence of the second acoustic signal for each frame corresponding to each other in the first acoustic signal and the second acoustic signal. Processing,
A program for executing index calculation processing for calculating representative values of a plurality of cross-correlation coefficients calculated for each frame as the masking effect index.