JP2014006692A - Hearing impression amount estimating device and program therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a hearing impression amount estimating device that can present an accurate hearing impression amount.SOLUTION: A hearing impression amount estimating device 1 includes: acoustic signal analysis means 10 for obtaining an acoustic analysis value by acoustic analysis; a database 20 for storing a mental state transition model 21 and a probability distribution model 23; mental state estimating means 30 for estimating a mental state of a listener by applying maximum likelihood estimation processing or random number generation processing with the acoustic analysis value to the mental state transition model; probability distribution model extracting means 40 for generating an extraction condition that conforms a mental state of a test subject with the mental state of the listener; and hearing impression amount estimating means 50 for calculating a hearing impression amount corresponding to the acoustic analysis value, from the probability distribution model.

Description

  The present invention relates to an auditory impression amount estimation device for estimating an auditory impression amount indicating a degree of a feature of a sound felt by a listener, and a program thereof.

  Along with the development of sound collection and reproduction technology, various acoustic systems with a high sense of realism, such as 22.2 multi-channel acoustic system and Wave Field Synthesis, have been proposed. For this reason, in these acoustic systems, it is required to objectively evaluate the quality.

  Conventionally, PEAQ (Perceptual Evaluation of Audio Quality) has been proposed as an objective evaluation method for encoded sound, and is standardized by ITU (International Telecommunication Union) (see Non-Patent Document 1). The technology described in Non-Patent Document 1 simulates a human auditory peripheral system, and substitutes the auditory central system with a simple neural network.

  In recent years, sensibility information called realism has attracted attention as an index for evaluating the quality of sound collection and reproduction technology. This sense of presence shows the feeling of being there, but it is known that it is influenced not only by the reproducibility of the acoustic space, but also by psychological effects (for example, “the heart is shaken”). (See Non-Patent Document 2). In addition, it is known that the auditory impression, which is the premise of a sense of reality, varies depending on various conditions such as the listener's preference and psychological state (see Non-Patent Document 3). Furthermore, it is known that the time constant of state transition differs depending on the type of emotion, such as emotions that change rapidly and immediately become a steady state, or emotions that last for a relatively long time. Furthermore, it is thought that the auditory impression that influences differs depending on the type of emotion.

  Here, for example, an invention described in Patent Document 1 has been proposed as an objective evaluation method of sound quality felt by humans. The invention described in Patent Document 1 uniquely determines an auditory impression amount from an acoustic feature amount, and estimates a sense of reality from the auditory impression amount.

JP 2011-250049 A

ITU-R BS. 1387 IEICE Technical Report HIP2008-132, “Image Survey on“ Realism ””, Hearing Society Document Vol.40, No.1, H-2010-1 Proceedings of the Acoustical Society of Japan 2010 Spring Meeting 2-1-7, “A study of influences of word and phone accuracies on unsupervised HMM-based speech synthesis”

  However, the invention described in Patent Document 1 has a problem that the psychological state is not reflected in the auditory impression amount because the auditory impression amount is uniquely determined from the acoustic feature amount, and the auditory impression amount is not accurate. For this reason, in the invention described in Patent Document 1, the estimated presence is not necessarily an accurate representation of the presence actually felt by the listener.

  Therefore, an object of the present invention is to solve the above-described problems and provide an auditory impression amount estimation device and a program thereof that can present an accurate auditory impression amount.

  In view of the above-described problem, the auditory impression amount estimation device according to the first invention of the present application estimates an auditory impression amount that indicates a degree of a sound feature felt by a listener when an acoustic signal is reproduced. An estimation device comprising: an acoustic signal analysis unit; a psychological state transition model database; a psychological state estimation unit; a probability distribution model database; an extraction condition generation unit; and an auditory impression amount calculation unit. To do.

  According to this configuration, the auditory impression amount estimation device obtains an acoustic analysis value, which is an acoustic feature amount of the inputted acoustic signal, by acoustic analysis of the acoustic signal by the acoustic signal analysis means.

  Here, as the acoustic feature amount, for example, time change pattern of fundamental frequency, frequency characteristics, frequency characteristic classification class, level time change pattern, loudness, roughness, sharpness, interaural time difference, interaural level difference, The binaural correlation degree and the width of the binaural correlation function can be given. Then, the auditory impression amount estimation device acoustically analyzes one or more of these acoustic feature amounts.

  In addition, the auditory impression amount estimation device uses a psychological state transition model in which a psychological state transition model is set in advance by performing a subjective evaluation experiment for each evaluation condition and setting a probability of transition of the subject's psychological state. It is remembered. Moreover, the auditory impression amount estimation device estimates the psychological state of the listener by the psychological state estimating means based on the psychological state transition probability set in the psychological state transition model and the random number generated in the random number generation process.

  The auditory impression amount estimation device stores a probability distribution model in which the auditory impression amount and the acoustic feature amount are associated in advance by performing a subjective evaluation experiment for each psychological state of the subject in the probability distribution model database. Therefore, the auditory impression amount estimation device uses the probability distribution model extraction unit to match the psychological state of the subject included in the probability distribution model with the psychological state of the listener estimated by the psychological state estimation unit. An extraction condition can be generated.

  The auditory impression amount estimation device extracts the auditory impression amount and the acoustic feature amount that match the extraction condition from the probability distribution model by the auditory impression amount calculation means, and the extracted acoustic impression amount and the acoustic feature amount have an acoustic feature. By applying a probability function set in advance for each section of the amount, an auditory impression amount corresponding to the acoustic analysis value is calculated. That is, the auditory impression amount estimation device extracts only model data that matches the psychological state of the listener from all model data (auditory impression amount and acoustic feature amount) of the probability distribution model. Therefore, the auditory impression amount presented by the auditory impression amount estimation device reflects the psychological state of the listener.

The psychological state indicates the feelings of the listener and the subject.
An auditory impression is a representation (labeling) of the characteristics of sounds felt by the listener or subject.
The auditory impression amount is a numerical value of the degree of the auditory impression, that is, the auditory impression.
A listener is a person who actually listens to an acoustic signal.
A subject is a subject of a subjective evaluation experiment when constructing a psychological state transition model or a probability distribution model.

Moreover, the auditory impression amount estimation device according to the second invention of the present application further includes setting parameter input means for inputting a listener's preference as a setting parameter depending on the listener, and the psychological state transition model database is set as an evaluation condition. A psychological state transition model including an acoustic feature amount and a subject's preference is stored.
According to this configuration, the auditory impression amount estimation device can reflect the listener's preference in the psychological state estimation result.
The preference indicates the preference of the listener or the subject.

Moreover, the auditory impression amount estimation device according to the third invention of the present application further includes biological information measuring means for measuring the biological information of the listener, the psychological state transition model database further includes the biological information of the subject in the evaluation condition. A psychological state transition model is stored.
According to this configuration, the auditory impression amount estimation device can reflect the listener's biological information in the psychological state estimation result.

Moreover, the auditory impression amount estimation device according to the fourth invention of the present application is characterized in that the psychological state transition model database stores a psychological state transition model in which the psychological state of the subject changes stepwise.
According to this configuration, the auditory impression amount estimation device can grasp in detail the transition of the listener's psychological state.

Moreover, the auditory impression amount estimation apparatus according to the fifth invention of the present application is characterized in that the psychological state transition model database stores a psychological state transition model in which an internal state included in the psychological state of the subject changes.
According to this configuration, the auditory impression amount estimation device can grasp in detail the transition of the listener's psychological state.
In addition, an internal state expresses (labels) the psychological state of a listener or a test subject with words. Here, the internal state is one unit (one unit) constituting the psychological state transition model.

  In addition, the auditory impression amount estimation apparatus according to the first invention of the present application can also be realized by an auditory impression amount estimation program that causes a general computer including hardware resources such as a CPU and a database to operate in cooperation with each other as described above. . This auditory impression amount estimation program may be distributed via a communication line, or may be distributed by writing in a recording medium such as a CD-ROM or a flash memory.

According to the present invention, the following excellent effects can be obtained.
According to the first invention of the present application, since the psychological state of the listener is reflected in the auditory impression amount, an accurate auditory impression amount can be presented.
According to the second invention of this application, in addition to the listener's psychological state, the listener's preference is reflected in the psychological state estimation result, so that a more accurate auditory impression amount can be presented.

According to the third aspect of the present invention, since the listener's biological information is reflected in the psychological state estimation result, a more accurate auditory impression amount can be presented.
According to the fourth and fifth aspects of the present invention, since the transition of the psychological state of the listener can be grasped in detail, a more accurate auditory impression amount can be presented.

It is a block diagram which shows the structure of the auditory impression amount estimation apparatus which concerns on 1st Embodiment of this invention. It is a figure explaining the 1st example of the psychological state transition model memorize | stored in the database of FIG. It is a figure explaining the 2nd example of the psychological state transition model memorize | stored in the database of FIG. It is a figure explaining the 3rd example of the psychological state transition model memorize | stored in the database of FIG. (A)-(c) is a figure explaining the production | generation reason of extraction conditions in 1st Embodiment. (A)-(e) is a figure explaining the production | generation reason of extraction conditions in 1st Embodiment. It is a figure explaining the calculation of the auditory impression amount by the auditory impression amount calculating means of FIG. 1, and the calculation of the realistic feeling estimated value by the realistic feeling estimated value calculating means. It is a figure explaining the example which the auditory impression amount presentation means of FIG. 1 presented the auditory impression amount and the realistic sensation estimated value in a bar graph format. It is a figure explaining the example which the auditory impression amount presentation means of FIG. 1 presented the auditory impression amount and the realistic sensation estimated value in the form of a correlation diagram. In the auditory impression amount estimation device according to the second embodiment of the present invention, FIG. It is a figure explaining the example which presented the auditory impression amount and the realistic sensation estimated value in the bar graph format in the auditory impression amount estimation apparatus according to the second embodiment of the present invention. It is a figure explaining the example which presented the auditory impression amount and the realistic sensation estimated value in the radar chart format in the auditory impression amount estimation device according to the second embodiment of the present invention. It is a flowchart which shows the whole operation | movement of the auditory impression amount estimation apparatus which concerns on 2nd Embodiment of this invention. It is a flowchart which shows operation | movement of step S1 of FIG. It is a flowchart which shows operation | movement of step S2 of FIG. It is a flowchart which shows operation | movement of step S3 of FIG. It is a flowchart which shows operation | movement of step S4 of FIG. In the modification 1 of this invention, it is a figure explaining a psychological state transition model. In the modification 2 of this invention, it is a figure explaining a psychological state transition model. (A)-(d) is a figure explaining the setting of the transition probability in the modification 2 of this invention. (A) And (b) is a figure explaining selection of a transition probability in the modification 2 of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings as appropriate. In each embodiment, means having the same function are denoted by the same reference numerals and description thereof is omitted.

(First embodiment)
[Configuration of hearing impression estimation device]
With reference to FIG. 1, the structure of the auditory impression amount estimation apparatus 1 which concerns on 1st Embodiment of this invention is demonstrated.
The auditory impression amount estimation device 1 presents an auditory impression amount when a listener listens to a reproduced sound field acoustic signal (acoustic signal) and an estimated value of presence. For this reason, the auditory impression amount estimation device 1 includes an acoustic signal analysis unit 10, a database 20, a psychological state estimation unit 30, a probability distribution model extraction unit (extraction condition generation unit) 40, and an auditory impression amount estimation unit 50. Is provided.
The setting parameter input means 60 and the biological information measurement means 70 will be described later.

First, the reproduced sound field acoustic signal input to the acoustic signal analyzing means 10 will be described.
The reproduced sound field acoustic signal is an acoustic signal collected in the reproduced sound field (an acoustic signal measured in an actually reproduced acoustic space). For example, the reproduced sound field acoustic signal is an acoustic signal obtained by collecting an acoustic signal obtained by playing an orchestra live performance in a concert hall using an audio device in a living room.

At this time, for example, a measurement device simulating a human head (a microphone installed on a dummy head) can be used for measurement of the acoustic space (sound collection of a reproduced sound field acoustic signal).
For measurement of the acoustic space, a measuring device that can acquire spatial information using a plurality of microphones may be used. For example, the sound arrival direction can be calculated by arranging a plurality of microphones and comparing the volume and phase. At this time, by calculating the correlation between signals (in the case of a dummy head, the correlation between both ears), it is possible to acquire a sense of sound expansion as spatial information.

  The acoustic signal analysis means 10 receives a reproduced sound field acoustic signal and obtains an acoustic analysis value that is an acoustic feature amount of the input reproduced sound field acoustic signal by acoustic analysis of the reproduced sound field acoustic signal. For example, the acoustic signal analysis means 10 can calculate the estimated loudness value, sound pressure level, frequency characteristics, interaural cross-correlation, interaural level difference, interaural phase difference, time-varying pattern of fundamental frequency, and frequency characteristic classification. Acoustic features such as class, level temporal change pattern, roughness, sharpness, and width of interaural correlation function are calculated as acoustic analysis values. Then, the acoustic signal analysis unit 10 outputs the calculated acoustic analysis value to the probability distribution model extraction unit 40.

Here, the acoustic signal analysis unit 10 may obtain the direction of the sound image as the acoustic feature amount. For example, the angle of the sound image with respect to the listener is obtained from the interaural level difference and the interaural phase difference. When the direction of the sound image is represented by four directions, the acoustic signal analyzing means 10 determines that the direction of the sound image is “front” if the angle of the obtained sound image is within the range of 90 ° in front of the listener, and the angle of the sound image is If the angle is within the range of 90 ° to the left and right of the listener, the direction of the sound image is “side”, and if the angle of the sound image is within the range of 90 ° to the back of the listener, the direction of the sound image is “back”.
Furthermore, the acoustic signal analysis means 10 may represent the direction of the sound image in eight directions divided within a 45 ° range.

  Note that these acoustic analysis methods are general and will not be described. Needless to say, the acoustic feature amount required by the acoustic signal analyzing means 10 is not limited to this as long as it can be analyzed. Moreover, it can be set manually which acoustic feature-value which the acoustic signal analysis means 10 calculates | requires.

  The database 20 is a database in which various information used by a psychological state estimation unit 30 and an auditory impression amount estimation unit 50 described later are stored in advance. For example, the database 20 stores an acoustic signal for evaluation, a psychological state transition model 21, a probability distribution model 23, and a weighting coefficient.

  The evaluation acoustic signal is an acoustic signal heard by the subject in the subjective evaluation experiment when the psychological state transition model 21 and the probability distribution model 23 are constructed. The evaluation acoustic signal may include a reproduced sound field acoustic signal or may not include a reproduced sound field acoustic signal.

The psychological state transition model 21 uses the acoustic feature amount as an evaluation condition and performs a subjective evaluation experiment for each evaluation condition to obtain a transition probability between the psychological states of the subject.
The evaluation condition (metadata) is a condition for performing a subjective evaluation experiment for constructing the psychological state transition model 21.
The construction of the psychological state transition model 21 will be described later.

  The psychological state indicates the feelings of the listener and the subject, for example, “impression” that indicates a thrilling impression, “normal” that does not have a particular impression, and a impression that is a generous impression. There is something like "impressed (gene)".

Other psychological states include, for example, “fun”, “sad”, and “irritated”.
Further, the psychological state may be classified by category. For example, the psychological state of the category “depression” includes psychological states “not concerned”, “anxious”, “not confident”. Further, for example, the psychological state of the category “hostility” includes psychological states of “aggressive”, “hateful”, and “muty”. Further, for example, the psychological state of the category “Fatigue” includes psychological states of “boring”, “tired”, and “boring”. Further, for example, the psychological state of the category “active pleasure” includes psychological states of “lively”, “full of energy”, and “hot”. Further, for example, the psychological state of the category “inactive pleasure” includes psychological states of “relaxed”, “idle”, and “moist”. For example, the psychological state of the category “affinity” includes psychological states of “Ioioi”, “Adorable”, and “Nice” (see Reference 1).
Reference 1: “Creation of a multifaceted emotional state scale”, Terasaki et al., Psychological Research, No. 62, pp. 350-356, 1992

Apart from the psychological state described above, the setting parameter may include the presence or absence of interest.
The interest indicates the interest of the listener or the subject with respect to the evaluation target (reproduced sound field acoustic signal A and evaluation acoustic signal). For example, the presence / absence of interest may be set to a binary value such as “0” for “interested” and “1” for “not interested”. Further, as the presence / absence of interest, values corresponding to “very interested”, “appreciably interested”, “not interested at all”, and the like may be set stepwise.

The probability distribution model 23 performs a subjective evaluation experiment for each subject's psychological state, and associates model data (auditory impression amount and acoustic feature amount) in advance.
The auditory impression is a word that expresses (labels) the characteristics of the sound felt by the listener or the subject, and includes, for example, “expansion” and “bright”.
The auditory impression amount is a numerical value of the degree of the auditory impression, that is, the auditory impression.
The construction of the probability distribution model 23 will be described later.

The weighting coefficient indicates the contribution ratio of the auditory impression amount to the realistic sensation estimated value, and can be obtained by multivariate analysis such as multiple regression analysis and quantification type I.
In the multiple regression analysis, when a variable (object variable) is predicted from a plurality of observed values (explanatory variables), the contribution rate is calculated so that the prediction error is minimized. In the present embodiment, a multiple regression analysis is performed in which the observed value (explanatory variable) is the auditory impression amount obtained from the probability distribution model 23 of the acoustic feature amount, and the variable (objective variable) is the estimated presence value. For this reason, the weighting coefficient depends on the observation value stored in the database 20.

  In addition, although the example which calculates | requires a weighting coefficient by multivariate analysis was demonstrated, the learning method is not limited to this. For example, the weighting factor can be obtained by machine learning such as a neural network or a genetic algorithm.

<Construction of psychological state transition model: first example>
Hereinafter, the first to third examples of the construction of the psychological state transition model 21 will be described with reference to FIGS. 2 to 4 (see FIG. 1 as appropriate).
As shown in FIG. 2, the psychological state transition model 21 of the first example is different from a certain psychological state, such as between “impression”, “normal”, and “impression”. Transition to psychological state. The psychological state transition model 21 maintains the same psychological state so as to maintain “normal”.

In FIG. 2, the psychological state of the subject is described as “impression”, “normal”, and “impression”.
Moreover, the character in a circle shows a psychological state, and the path | route in which a psychological state changes is shown by the arrow (FIG. 3 is also the same).
Also, _let P _x be the psychological state transition probability (x = 11, 12, 13, 21, 22, 31, 33). For example, the transition probability P ₁₁ indicates the probability of maintaining “normal” without making a transition from “normal”. Further, for example, the transition probability P ₁₂ indicates the probability of transition from “normal” to “impression (gene)”.

  The psychological state transition model 21 can be constructed by subjecting a plurality of subjects to subjective evaluation experiments. Specifically, a psychological state change (transition) is input by a fader while allowing the subject to listen to the evaluation acoustic signal. In this fader, the left end indicates “impression”, the center indicates “normal”, and the right end indicates “impression”. Moreover, the test subject can move the fader to input a change in the psychological state when the evaluation acoustic signal is heard.

Further, while the evaluation acoustic signal is heard by the subject, the “loudness” of the evaluation acoustic signal is measured. Then, from the relationship between the change in the psychological state and the “loudness”, the psychological state transition model 21 indicating the transition probability P _x between “impression”, “normal”, and “impression” is generated. Can ask

Here, it is known that when there is a sudden change in sound volume, a shocking impression can be obtained (see Reference 2). For this reason, in the psychological state transition model 21, when the difference between the estimated loudness values at the times t ₁ and t ₂ of the evaluation acoustic signal is greater than or equal to a preset threshold value, “normal” is changed to “impression (gene)”. the value of the transition probability P ₁₂ may be set high.

  Reference 2: O. Grewe, et al, “Listening to music as a re-creative process: Physiological, psychological, and psychoacoustical correlates of chills and strong emotions.” Music Perception, Vol. 24, No. 3, pp. 297 -314,2007.

<Construction of psychological state transition model: second example>
As shown in FIG. 3, the psychological state transition model 21 of the second example is between “impressed (gene) high”, “impressed (gene)”, and “excited (gene) low”. A certain psychological state transitions in stages.

  For example, in the psychological state transition model 21, according to the intensity of impression such as “gene”, “excitement (gene) high”, “excitement (gene)”, and “excitement (gene) low” 3 It is set in stages.

  Here, “impression (gene) high” indicates that the intensity of impression that is generous is high. Further, “low impression (gene)” indicates that the intensity of impression such as a gene is low. In addition, “in the middle of impression” indicates that the intensity of impression as a gene is intermediate between “impression (gene) high” and “impression (gene) low”.

As described above, in the second example, it is possible to construct the psychological state transition model 21 that reflects the strength of the psychological state and grasp the transition of the psychological state of the listener in detail.
In the second example, the transition probability P _x can be set in the same manner as in the first example, and thus the description and illustration are omitted.

<Construction of psychological state transition model: third example>
As shown in FIG. 4, the psychological state transition model 21 of the third example has the same psychological state between the internal states “indifference”, “interest”, and “high interest” included in “normal”. The internal state included in the state transitions. The psychological state transition model 21 also transitions even in internal states included in different psychological states, such as between “high interest” included in “normal” and internal states “low emotion” included in “impression”. To do.

This internal state expresses (labels) the psychological state of the listener or subject in words. Here, the internal state is one unit (one unit) constituting the psychological state transition model 21.
For example, if the psychological state is “normal”, the internal state is divided into “indifference” indicating that there is no interest, “interest” indicating that it is interested, and “high interest” that indicates high interest. Become.
For example, if the psychological state is “impressed”, the internal state is “impressed low” indicating that the emotion is low and “impressed high” indicating that the emotion is high.
Further, for example, if the psychological state is “disgust”, the internal state is “denial” indicating that it is denied, and “rough” indicating that it is violent.

  As described above, in the third example, by distinguishing the psychological state “normal” from the internal state, it is possible to change the probability of transition from each internal state to “impression”, which is another psychological state. Therefore, in the third example, it is possible to grasp in detail the transition of the listener's psychological state.

  Also, if the psychological state once becomes “impressed” and then returns to “normal”, the internal state at that time is considered to be “high interest” that is likely to return to “impressed” next. Therefore, as shown in FIG. 4, a psychological state transition model 21 that does not return to a certain internal state, such as “impression low” to “interest”, can be constructed.

In FIG. 4, the character inside the broken line indicates the psychological state, the character in the circle indicates the internal state, and the path through which the psychological state transitions is indicated by an arrow.
In the third example, since the transition probability P _x can be set in the same manner as in the first example, the description and illustration are omitted.
Further, for example, the psychological state transition model 21 can be constructed by any method among the first to third examples described above.

<Construction of probability distribution model>
The construction of the probability distribution model 23 will be described.
For example, it is considered to be influenced by the “spreading feeling” of the auditory impression, the degree of intercorrelation between both ears, the sound pressure level, the direction of the sound source, the spread of the immediately preceding sound, and the listener's preference. Therefore, the acoustic feature quantity is described as “interaural cross-correlation”, and the auditory impression is described as “expansion”.

  The probability distribution model 23 can be constructed by performing subjective evaluation experiments on a plurality of listeners (subjects). Specifically, the subject is made to listen to the evaluation acoustic signal, and the “interaural cross-correlation” of the evaluation acoustic signal is measured in advance. In addition, the subject's psychological state such as “normal” is also answered in advance by the subject.

  Here, the degree of “expansion” felt when listening to the evaluation acoustic signal in various psychological states is answered by the subject, and the auditory impression amount of “expansion” is obtained. Then, the “interaural cross-correlation degree” of the acoustic signal for evaluation is associated with the auditory impression amount of “expansion” answered by the subject to obtain model data. At this time, the psychological state when the model data is associated is added to each model data. This subjective evaluation experiment is performed on a plurality of subjects whose psychological states are “normal” and “impressed”, and model data in various psychological states is obtained. As a result, the probability distribution model 23 adds psychological states such as “normal” and “impression” to each model data.

Returning to FIG. 1, the description of the configuration of the auditory impression amount estimation device 1 will be continued.
The psychological state estimation means 30 estimates the psychological state of the listener based on the psychological state transition probability set in the psychological state transition model 21 and the random number generated by the random number generation process. That is, the psychological state estimation means 30 estimates the most likely listener's psychological state stochastically using random number generation processing.

Here, there is a possibility of transition from the same psychological state to different psychological states, such as “normal” to “impression (gene)” and “normal” to “impression (zoku)” (see FIG. 2). ). In this case, the psychological state estimating means 30 determines whether or not the transition is made from “normal” to “impression (gene)” and “impression (zock)”, the transition probabilities P ₁₂ and P ₁₃ and the transition probabilities P ₁₂ and P _13. Judgment is made by comparison with random numbers generated in each of ₁₃ . For example, the maximum value, average value, or median value of the transition probabilities P ₁₂ and P ₁₃ is used as a reference value, the reference value is compared with a random number, and the psychological state where the random number is farthest from the reference value is estimated. It is good. Then, the psychological state estimating unit 30 outputs the estimated psychological state of the listener to the probability distribution model extracting unit 40.

  The probability distribution model extraction unit 40 generates an extraction condition that is a psychological state that matches between the psychological state of the subject included in the probability distribution model 23 and the psychological state of the listener estimated by the psychological state estimation unit 30. To do.

  For example, in the probability distribution model 23, the psychological state of the subject is “impressed” and “normal”. In addition, it is assumed that the psychological state of the listener estimated by the psychological state estimating unit 30 is “impression”. In this case, among the psychological states of all the model data included in the probability distribution model 23, the psychological state that matches the estimation result of the psychological state estimating means 30 is “impression”. Accordingly, the probability distribution model extraction unit 40 generates, for example, an extraction condition indicating “impression” and outputs the extraction condition to the auditory impression amount estimation unit 50. Further, the probability distribution model extraction unit 40 outputs the acoustic analysis value input from the acoustic signal analysis unit 10 to the auditory impression amount estimation unit 50.

<Reason for generating extraction conditions>
The reason for generating the extraction condition will be described with reference to FIGS.
As shown in FIG. 5A, the acoustic feature amount changes according to the time of the acoustic signal, and the auditory impression amount also changes accordingly. For this reason, a psychological state such as impression may change in the middle of an acoustic signal (for example, while listening to music).
In FIG. 5A, the acoustic feature quantity at each time is shown by a solid line, and the average value of the acoustic feature quantity is shown by a broken line.

In the conventional probability distribution model 90, as shown in FIG. 5B, the average value or the maximum value of the acoustic analysis result is used as the acoustic feature amount of each model data, and the auditory impression amount of each model data is the amount after hearing. Evaluation values are used.
In FIG. 5B, the model data at each time of the acoustic signal is illustrated by black circles. That is, in the probability distribution model 90 of FIG. 5B, each black circle indicates model data at a different time.

However, as shown in FIG. 5C, even if the acoustic feature amount is the same, the auditory impression amount may differ depending on the psychological state of the subject.
In FIG. 5 (c), when listening to a certain acoustic signal, the average value of the auditory impression obtained from the subject who has been impressed is illustrated by a solid line, the average value of all listeners is illustrated by a broken line, The average value of the amount of auditory impression obtained from the subject who answered that he / she did not was shown by a one-dot chain line.

For example, as shown in FIG. 6A, a regression line is drawn from all model data of the probability distribution model 23.
In FIG. 6, model data at each time of the evaluation acoustic signal is illustrated by black circles. That is, in the probability distribution model 23 of FIG. 6, each black circle represents model data at a different time.

  In addition, as shown in FIG. 6B, in the probability distribution model 23, model data (black circle) obtained from the subject who answered that he / she was impressed after listening to the evaluation acoustic signal and the evaluation acoustic signal were not impressed. Model data (white circles) obtained from subjects who answered, were classified. For example, even if the subject replied that they were impressed, they may not be impressed in the first half of the evaluation acoustic signal, and the dispersion of the model data becomes large. As a result, as shown in FIG. 6C, the error between the regression line (solid line) obtained from all black circles in FIG. 6 (b) and the regression line obtained from all white circles (broken line) increases.

  Further, as shown in FIG. 6D, in the probability distribution model 23, model data (black circles) obtained from a subject in a psychological state that is moved and model data (black circles) obtained from a subject in a psychological state that is not moved ( White circle). Here, as shown in FIG. 6 (b), model data obtained from a subject who answered that he / she was impressed after listening to the evaluation acoustic signal is treated as time series data as shown in FIG. 6 (d). The model data obtained from the subject who was in an unimpressed psychological state is included (see reference numeral 91). Therefore, as shown in FIG. 6E, when a regression line obtained from the all black circles in FIG. 6D and a regression line obtained from the all white circles are drawn, the error between the two becomes small. In other words, the error of the regression line is reduced by narrowing down the model data of the probability distribution model 23 under the extraction condition.

Returning to FIG. 1, the description of the configuration of the auditory impression amount estimation device 1 will be continued.
The auditory impression amount estimation means 50 calculates and presents the auditory impression amount and the realistic sensation estimated value. The auditory impression amount estimating means 50 includes an auditory impression amount calculating means 51, a realistic sensation estimated value calculating means 53, and an auditory impression amount presenting means 55.

The auditory impression amount calculation means 51 extracts model data that matches the extraction conditions generated by the probability distribution model extraction means 40 from the probability distribution model 23, and is preset in the extracted model data for each section of the acoustic feature amount. By applying the obtained probability function, the auditory impression amount corresponding to the acoustic analysis value input from the probability distribution model extraction means 40 is calculated.
The realistic sensation estimated value calculating unit 53 calculates a value obtained by multiplying the auditory impression amount calculated by the auditory impression amount calculating unit 51 by a preset weighting coefficient as the realistic sensation estimated value.

<Calculation of auditory impression and realistic sensation>
With reference to FIG. 7, calculation of the auditory impression amount by the auditory impression amount calculating unit 51 and calculation of the realistic sensation estimated value by the realistic sensation estimated value calculating unit 53 will be described (see FIG. 1 as appropriate).
The auditory impression amount calculation means 51 extracts model data that matches the extraction condition from the probability distribution model 23 including all model data. For example, consider a case where the psychological states of all model data included in the probability distribution model 23 are “impressed” and “normal”, and the extraction condition is “impressed”. In this case, the auditory impression amount calculation unit 51 extracts “impression” model data from the probability distribution model 23.

  Further, as shown in FIG. 7, the auditory impression amount calculation unit 51 divides the acoustic feature amount (horizontal axis direction) of the probability distribution model into predetermined sections. And the auditory impression amount calculation means 51 calculates | requires the area in which an acoustic analysis value is contained among each divided | segmented area. Furthermore, the auditory impression amount calculation unit 51 applies a probability function (eg, normal distribution, binomial distribution) of the obtained section to the extracted model data, and calculates an auditory impression amount corresponding to the acoustic feature amount. Here, the auditory impression amount calculating means 51 calculates the existence probability of the auditory impression amount according to a probability function. For example, the auditory impression amount calculation means 51 generates a random number, regards the random value as the cumulative frequency of existence probability, and calculates the auditory impression amount.

  At this time, in order to improve the reliability of the realistic sensation estimation value, the auditory impression amount calculation means 51 generates a random number a plurality of times to calculate a cumulative frequency a plurality of times, and calculates the average value of all the calculated cumulative frequencies as an auditory impression. It may be calculated as a quantity. Further, the auditory impression amount calculation means 51 may calculate the temporal change of the auditory impression amount within a preset range when calculating the realistic sensation estimated value in real time.

  The realistic sensation estimated value calculation means 53 reads the weighting coefficient corresponding to the probability distribution model 23 from the database 20. Then, the realistic sensation estimated value calculating unit 53 calculates the realistic sensation estimated value by multiplying the calculated auditory impression amount and the read weight coefficient.

Returning to FIG. 1, the description of the auditory impression amount estimation means 50 will be continued.
The auditory impression amount presenting means 55 presents the auditory impression amount calculated by the auditory impression amount calculating means 51 and the realistic feeling estimated value calculated by the realistic feeling estimated value calculating means 53. Here, the auditory impression amount presenting means 55 presents the auditory impression amount and the realistic sensation estimated value in a graphical format.

<Presentation of auditory impression and realistic sensation>
The presentation of the auditory impression amount and the realistic sensation estimated value by the auditory impression amount presenting means 55 will be described with reference to FIGS.
As shown in FIG. 8, the auditory impression amount presenting means 55 presents, for example, the auditory impression amount and the realistic sensation estimated value in a bar graph format. Specifically, the auditory impression amount presenting unit 55 presents the auditory impression amount (for example, “expansion”) calculated by the auditory impression amount calculating unit 51 as a bar graph 96. The auditory impression amount presenting means 55 presents the realistic sensation estimated value calculated by the realistic sensation estimated value calculating means 53 as a bar graph 97. At this time, the auditory impression amount presenting means 55 may present the extraction condition 98 generated by the probability distribution model extracting means 40.

  Moreover, as shown in FIG. 9, the auditory impression amount presenting means 55 may present the auditory impression amount and the realistic sensation estimated value in the form of a correlation diagram. In this correlation diagram, the vertical axis represents the realistic sensation estimation value, and the horizontal axis represents the amount of auditory impression (for example, “expansion”). Then, a point 99 indicating the auditory impression amount calculated by the auditory impression amount calculating means 51 and the realistic feeling estimated value calculated by the realistic feeling estimated value calculating means 53 is plotted in this correlation diagram.

As described above, in the auditory impression amount estimation device 1 according to the first embodiment of the present invention, the auditory impression amount calculation unit 51 matches the psychological state of the listener among all model data of the probability distribution model 23. Extract model data only. Thus, the auditory impression amount estimation device 1 can reflect the psychological state of the listener in the auditory impression amount and present an accurate auditory impression amount. Furthermore, since the auditory impression amount estimation device 1 calculates an estimated value of the realistic sensation from the auditory impression amount, the realistic sensation actually felt by the listener can be accurately presented.
The operation of the auditory impression amount estimation device 1 is the same as that of the second embodiment, and will be described later.

(Second Embodiment)
[Configuration of hearing impression estimation device]
With reference to FIG. 10, a different point from 1st Embodiment is demonstrated about the structure of the auditory impression amount estimation apparatus 1A which concerns on 2nd Embodiment of this invention (refer FIG. 1 suitably).
The auditory impression amount estimation device 1A differs from the first embodiment in calculating j types of auditory impression amounts from i types of acoustic analysis values (however, an integer satisfying i> 1 and j> 1). Therefore, the auditory impression amount estimation device 1A includes an acoustic signal analysis unit 10, a database 20, a psychological state estimation unit 30, a probability distribution model extraction unit 40A, and an auditory impression amount estimation unit 50A.
In the present embodiment, it is assumed that a probability distribution model in which different acoustic analysis values and auditory impression amounts are associated one-to-one is stored (that is, i = j).

  The acoustic signal analyzing means 10 includes an estimated loudness value, sound pressure level, frequency characteristics, interaural cross-correlation, interaural level difference, interaural phase difference, fundamental frequency time change pattern, frequency characteristic classification class, Among acoustic feature quantities such as level temporal change pattern, roughness, sharpness, and width of interaural correlation function, i types are obtained as acoustic analysis values.

Database 20 stores the evaluation sound signal, the psychological state transition model 21, and j different probability distribution models Q _j, and a weighting coefficient W _j for each probability distribution model Q _j.
That is, as shown in FIG. 10, the database 20 has a probability distribution model Q _{1 in} which the first type of acoustic feature quantity (acoustic analysis value 1) and the first type of auditory impression quantity (auditory impression quantity 1) are associated with _{each other.} Remember.
Furthermore, database 20, the acoustic feature quantity of the second type (acoustic analysis value 2) and auditory impression of second kind (auditory impression of 2) stores a probability distribution model Q ₂ to which associated.
The database 20 stores a probability distribution model Q ₃ which acoustic features of third type (acoustic analysis 3) and auditory impression of third type (auditory impression of 3) is associated.
The database 20 stores a probability distribution model Q ₄ of acoustic features of fourth type (acoustic analysis value 4) and auditory impression of fourth type (auditory impression of 4) are associated.
In addition, the database 20 stores a probability distribution model Q _{j in} which the i-th acoustic feature value (acoustic analysis value i) and the j-th auditory impression amount (auditory impression amount j) are associated with each other.

The auditory impression amount estimating unit 50A includes an auditory impression amount calculating unit 51A, a sense of presence estimated value calculating unit 53A, and an auditory impression amount presenting unit 55A.
The auditory impression amount calculating unit 51A extracts model data that matches the extraction condition from each of the j types of probability distribution models _Qj , and calculates the j types of auditory impression amounts with reference to the extracted model data.
Note that the method of calculating the auditory impression amount is the same as that of the auditory impression amount calculating unit 51, and thus description thereof is omitted.

The realistic sensation estimated value calculation means 53A is the total estimated value obtained by multiplying the j types of auditory impression amounts calculated by the auditory impression amount calculation means 51A and the weighting factor W _{j for} each probability distribution model Q _j. Calculate as
Specifically, realistic estimate calculation unit 53A calculates the auditory impression of the first type, the multiplication value of the weighting factor W ₁ of the probability distribution model Q _1. Also, realistic estimate calculation unit 53A calculates the auditory impression of the second type, the multiplication value of the weight coefficient W ₂ of the probability distribution model Q _2. Also, realistic estimate calculation unit 53A calculates the auditory impression of third type, the multiplication value of the weighting factor W ₃ of the probability distribution model Q _3. Also, realistic estimate calculation unit 53A calculates the auditory impression of fourth type, the multiplication value of the weighting factor W ₄ of the probability distribution model Q _4. Also, realistic estimate calculation unit 53A calculates the auditory impression of j types th, the multiplication value of the weighting coefficient W _j of the probability distribution model Q _j. Then, the realistic sensation estimated value calculation unit 53A adds the j multiplied values to calculate the realistic sensation estimated value.

<Presentation of auditory impression and realistic sensation>
With reference to FIG. 11 and FIG. 12, the presentation of the auditory impression amount and the realistic sensation estimated value by the auditory impression amount presenting means 55A will be described.
The auditory impression amount presenting means 55A presents the auditory impression amount and the realistic sensation estimated value in a graphical format. Here, it is assumed that “movement feeling” and “expansion feeling” are obtained as the auditory impression amount.

As shown in FIG. 11, the auditory impression amount presenting means 55 </ b> A may present “movement feeling”, “expansion feeling”, and a sense of presence estimated value in a bar graph format.
Also, as shown in FIG. 12, the auditory impression amount presenting means 55A may present the “movement feeling”, “expansion feeling”, and the realistic sensation estimated value in a radar chart format.

[Overall operation of hearing impression estimation device]
With reference to FIG. 13, the overall operation of the hearing impression estimation device 1A will be described (see FIG. 1 as appropriate).
In FIG. 13, the auditory impression amount estimation device 1A calculates the auditory impression amount and the realistic sensation estimation value in real time from the input reproduced sound field acoustic signal.

The auditory impression amount estimation apparatus 1A acoustically analyzes the reproduced sound field acoustic signal by the acoustic signal analysis means 10 to obtain an acoustic analysis value (step S1).
The auditory impression amount estimating device 1A estimates the psychological state of the listener by the psychological state estimating means 30 (step S2).
The auditory impression amount estimation device 1A generates an extraction condition by the probability distribution model extraction unit 40A (step S3).

The auditory impression amount estimation device 1A calculates the auditory impression amount by the auditory impression amount calculation means 51A, and calculates the realistic feeling estimated value by the realistic feeling estimated value calculation means 53A (step S4).
The auditory impression amount estimation device 1A presents the auditory impression amount presenting means 55A and the auditory impression amount and the realistic sensation estimation value during real-time processing (step S5).

The auditory impression amount estimation device 1A determines whether or not to end the entire process depending on whether or not the reproduced sound field acoustic signal has reached the end (step S6).
When the reproduced sound field acoustic signal does not reach the end (No in step S6), the auditory impression amount estimation device 1A determines that the overall processing is not ended, increments counters ci and cj described later, and performs the processing in step S1. Return to.

When the reproduced sound field acoustic signal reaches the end (Yes in step S6), the auditory impression amount estimation device 1A determines that the entire process is finished, and proceeds to the process of step S7.
The auditory impression amount estimation device 1A presents the auditory impression amount presentation unit 55A with the auditory impression amount presentation means 55A and the auditory impression amount and the realistic sensation estimated value after the entire process is completed (step S7).

[Acoustic signal analysis processing]
The acoustic signal analysis process (step S1) in FIG. 13 will be described with reference to FIG. 14 (see FIG. 1 as appropriate).
The acoustic signal analysis means 10 cuts out a predetermined section from the reproduced sound field acoustic signal (step S11).

The acoustic signal analysis means 10 reads the reproduced sound field acoustic signal of the section cut out in the past from a memory (not shown) (step S12).
Note that the acoustic signal analysis means 10 performs the process of step S12 because the reproduced sound field acoustic signal of the past section is required when the acoustic feature quantity such as the level temporal change pattern is used.

The acoustic signal analysis means 10 determines whether or not the counter ci is equal to or less than the number N of acoustic feature quantity types (step S13).
When the counter ci is equal to or less than the number N of acoustic feature quantity types (Yes in step S13), the acoustic signal analysis unit 10 proceeds to the process of step S14.
The acoustic feature quantity type number N indicates the number of acoustic feature quantity types to be subjected to acoustic analysis, and is preset (N = i).

  The acoustic signal analysis means 10 calculates an acoustic analysis value from the ci-th acoustic feature amount. And the acoustic signal analysis means 10 returns to the process of step S11 (step S14).

If the counter ci exceeds the number N of acoustic feature quantity types (No in step S13), the acoustic signal analysis means 10 proceeds to the process of step S15.
The acoustic signal analyzing means 10 stores the reproduced sound field acoustic signal cut out in step S11 in the memory, and ends the process (step S15).
That is, the acoustic signal analysis means 10 calculates N types of acoustic analysis values.

[Psychological state estimation processing]
The psychological state estimation process (step S2) in FIG. 11 will be described with reference to FIG.
The psychological state estimating means 30 reads the psychological state estimated in the past from a memory not shown (step S21).

The psychological state estimation means 30 determines whether or not the counter cj is equal to or less than the number L of auditory impression types (step S22).
If the counter ci is less than or equal to the number L of auditory impression types (Yes in step S22), the psychological state estimating means 30 proceeds to the process of step S23.
The psychological state estimation means 30 estimates the psychological state from the psychological state transition probability set in the psychological state transition model 21 and the random number generated in the random number generation process (step S23).
The number L of auditory impressions indicates the number of types of auditory impressions to be estimated (L = j).

If the counter ci exceeds the number L of auditory impression types (No in step S22), the psychological state estimating means 30 proceeds to the process of step S24.
The psychological state estimating means 30 stores the psychological state estimated in step S23 in the memory and ends the process (step S24).
Since the psychological state estimating means 30 needs to refer to the path from the past psychological state to the current psychological state in the psychological state estimating model 21 in order to estimate the current psychological state, the process of step S24 I do.

[Extraction condition generation process]
The extraction condition generation process (step S3) in FIG. 13 will be described with reference to FIG. 16 (see FIG. 1 as appropriate).
The probability distribution model extraction unit 40A determines whether or not the counter ci is equal to or smaller than the addition value of the number N of acoustic feature quantity types and the number M of parameter setting types (step S31).
The parameter setting type number M is the number of parameter types set in the setting parameter, and is zero in this embodiment because the setting parameter is not used.

When the counter ci is equal to or smaller than the addition value (N + M) (Yes in step S31), the probability distribution model extraction unit 40A proceeds to the process of step S32.
The probability distribution model extraction unit 40A generates an extraction condition that is a psychological state that matches between the psychological state of the subject included in the probability distribution model and the psychological state of the listener estimated by the psychological state estimation unit 30. (Step S32).
The probability distribution model extraction unit 40A stores the ci-th extraction condition in the memory and returns to the process of step S21 (step S33).

When the counter ci exceeds the added value (N + M) (No in step S31), the probability distribution model extracting unit 40A proceeds to the process of step S34.
The probability distribution model extraction unit 40A outputs the extraction condition stored in step S33 to the auditory impression amount estimation unit 50 (step S34).

[Hearing impression / realistic estimated value calculation processing]
With reference to FIG. 17, the auditory impression amount / realistic feeling estimated value calculation process (step S4) of FIG. 13 will be described (see FIG. 1 as appropriate).
The auditory impression amount estimation unit 50A determines whether or not the counter cj is equal to or smaller than the number L of auditory impression types (step S41).
When the counter cj is less than or equal to the number L of auditory impression types (Yes in step S41), the auditory impression amount estimation unit 50A proceeds to the process of step S42.

The auditory impression amount calculating means 51A extracts model data that matches the extraction condition from the probability distribution model (step S42).
The auditory impression amount calculating unit 51A refers to the extracted model data, calculates the jth auditory impression amount corresponding to the acoustic analysis value, and returns to the process of step S31 (step S43).

When the counter cj exceeds the number L of auditory impression types (No in step S41), the auditory impression amount estimation unit 50A proceeds to the process of step S44.
The realistic sensation estimated value calculation means 53A calculates a value obtained by multiplying each auditory impression amount by the weighting coefficient of the probability distribution model for calculating each auditory impression amount as the realistic sensation estimated value (step S44).

  As described above, the auditory impression amount estimation device 1A according to the second embodiment of the present invention, even when a plurality of acoustic feature amounts contribute to one auditory impression, as in the first embodiment, The feeling can be presented accurately.

(Third embodiment)
Returning to FIG. 1, the auditory impression amount estimation device 1 </ b> B according to the third embodiment of the present invention will be described while referring to differences from the first embodiment.

  The auditory impression amount estimation device 1B is different from the first embodiment in that preference is further included in the evaluation condition of the psychological state transition model 21B. Therefore, the auditory impression amount estimation device 1B includes the acoustic signal analysis means 10, the database 20B, the psychological state estimation means 30B, the probability distribution model extraction means 40, the auditory impression amount estimation means 50, and the setting parameter input means 60. With.

The database 20B stores a psychological state transition model 21B in which preference is further included in the evaluation condition.
Here, in the psychological state transition model 21B, the psychological state transition probability P _x is set using an acoustic feature quantity (for example, an estimated loudness value) as an evaluation condition, as in the first embodiment.
The transition probability P _x using the estimated loudness value as an evaluation condition is referred to as “transition probability P _{x | L} ”.

In this case, the subject's preference such as “gentle” as an evaluation condition is also made to be answered in advance by the subject. For example, it is known that it is easy to be impressed with respect to an object of interest (see Reference 3). Therefore, the preference of the listener's configuration parameters to be described later, when the preference of the subject matches, sets a high transition probability P ₁₂ from "normal" to "impress (Gene)". In this way, a unique psychological state transition model 21B can be constructed for each listener.
Reference 3: Togashi, “Mechanism of“ Kan ””, Cognitive Science, Vol. 8, No. 4, pp. 360-368, 2001
The transition probability P _x with preference as an evaluation condition is referred to as “transition probability P _{x | p} ”.

  The preference indicates the preference of the listener or the subject, and examples thereof include “like”, “dislike”, “gentle”, and “severe”. The preference may be a listener's preference for the type of content, such as “I like classic”. Here, the preference can be defined in advance by a subjective evaluation experiment described later.

Here, the setting parameter input means 60 will be described prior to the psychological state estimation means 30B.
The setting parameter input means 60 inputs setting parameters depending on the listener. For example, the listener can change the preference (for example, “gentle”) that the listener himself has from a predefined preference such as “gentle” or “severe” via a mouse or keyboard (not shown). Enter (select). Then, the setting parameter input unit 60 generates a setting parameter indicating that the listener's preference is “gentle”. Then, the setting parameter input unit 60 outputs the generated setting parameter to the psychological state estimation unit 30B.

Apart from the psychological state described above, the setting parameter may include the presence or absence of interest.
The interest indicates the interest of the listener or the subject with respect to the evaluation target (reproduced sound field acoustic signal A and evaluation acoustic signal). For example, the presence / absence of interest may be set to a binary value such as “0” for “interested” and “1” for “not interested”. Further, as the presence / absence of interest, values corresponding to “very interested”, “appreciably interested”, “not interested at all”, and the like may be set stepwise.

In addition, even when listening to the same sound, if the listening time zone is different, the auditory impression may be different due to the influence of the listener's biorhythm and fatigue. Therefore, the setting parameter may include a listening time zone in order to reflect the influence of this biorhythm and fatigue.
The listening time zone indicates a time zone in which the listener or the subject listens to sound (music). For example, there are “weekdays from 21:00 to 22:00” and “holiday from 10:00 to 12:00”. .
Note that the listener can input (select) the setting parameters such as the listener's interest and the listening time zone to the setting parameter input means 60 as well as the listener's preference.

  The psychological state estimating means 30B estimates the psychological state of the listener based on the psychological state transition probability set in the psychological state transition model 21B and the random number generated in the random number generation process.

As described above, even if the psychological state transition model 21B is the same route from “normal” to “impressed (gene)”, the evaluation condition includes two types of the acoustic feature amount and the subject's preference. There are types of transition probabilities P _{12 | L} and P _{12 | p} . For this reason, the psychological state estimating means 30B estimates the psychological state as described below.

For example, the transition probabilities P _{12 | L} and P _{12 | p} are normalized with values from 0.1 to 1.0, the transition probability P _{12 | L} = 0.6, and the transition probability P _{12 | p} = 0.4. Suppose that In this case, a random number is generated, and it is determined whether or not the generated random number exceeds the transition probability P _{12 | L.} Here, when the random number exceeds the transition probability P _{12 | L} , the psychological state estimation means 30B makes a transition from “normal” to “impression (gene)”. On the other hand, if the random number does not exceed the transition probability P _{12 | L} , the psychological state estimation unit 30B generates the random number again, and sets “normal” depending on whether the re-generated random number exceeds the transition probability P _{12 | p.} It is determined whether or not to make a transition to “impression (gene)”.

  As described above, the auditory impression amount estimation device 1B according to the third embodiment of the present invention reflects the listener's preference in the psychological state estimation result in addition to the listener's psychological state. The amount of auditory impression can be presented.

(Fourth embodiment)
The auditory impression amount estimation device 1 </ b> C according to the fourth embodiment of the present invention will be described with respect to differences from the third embodiment.

  The auditory impression amount estimation device 1C is different from the first embodiment in that ecological information is further included in the evaluation condition of the psychological state transition model 21C. For this reason, the auditory impression amount estimation device 1C includes the acoustic signal analysis means 10, the database 20C, the psychological state estimation means 30C, the probability distribution model extraction means 40, the auditory impression amount estimation means 50, and the setting parameter input means 60. And biological information measuring means 70.

  The database 20C stores a psychological state transition model 21C in which biological information is further included in the evaluation condition. For example, the biometric information is a brain wave measurement value, a heart rate, or a sweating amount of a listener or a subject.

  Here, similarly to the first embodiment, the psychological state transition model 21 </ b> C is set with the transition probability of the psychological state using the acoustic feature amount (for example, estimated loudness value) and the subject's preference as evaluation conditions. At this time, the biological information of the subject who is listening to the evaluation acoustic signal is measured in advance as an evaluation condition. In this way, the psychological state transition model 21C can be constructed.

Here, the biological information measuring means 70 will be described prior to the psychological state estimating means 30C.
The biological information measuring means 70 measures the biological information of the listener. For example, the biological information measuring unit 70 measures the biological information of the listener when listening to the reproduced sound field acoustic signal. Then, the biological information measuring unit 70 outputs the measured biological information to the psychological state estimating unit 30C.

  The psychological state estimation means 30C estimates the listener's psychological state from the psychological state transition probability set in the psychological state transition model 21C and the random number generated in the random number generation process.

  As described above, the psychological state transition model 21 </ b> C includes three types of the evaluation condition, that is, the acoustic feature amount, the subject's preference, and the ecological information even in the same route from “normal” to “impression (gene)”. Therefore, there are three types of transition probabilities. Accordingly, the psychological state estimating means 30C generates random numbers for each of the three types of transition probabilities, and determines whether or not the generated random numbers exceed the transition probability by a preset reference number (for example, 3) or more. When a random number equal to or greater than the reference number exceeds the transition probability, the psychological state estimation unit 30C causes the psychological state to transition. On the other hand, when a random number equal to or greater than the reference number does not exceed the transition probability, the psychological state estimation unit 30C does not cause the psychological state to transition.

  As described above, the auditory impression amount estimation device 1C according to the fourth embodiment of the present invention can present a more accurate auditory impression amount because the biological information of the listener is reflected in the estimation result of the psychological state. it can.

  As mentioned above, although each embodiment of this invention was described, this invention is not limited to this, It can implement in the range which does not change the meaning. The modification of embodiment is shown below.

(Modification 1)
With reference to FIG. 18, the difference from the first embodiment will be described regarding the auditory impression amount estimation device 1 according to the first modification of the present invention.
As shown in FIG. 18, the psychological state transition model 21 has a first point in which two types of psychological states are changed in stages according to the strength, such as “impression (gene)” and “impression (zoku)”. Different from the embodiment.

  Here, in the psychological state transition model 21, each of “impression (gene)” and “impression” is set in three stages according to the strength. In the psychological state transition model 21, “impression (gene)” and “impression (high)” are combined at each stage. For example, “impression (gene low) (slow / low)” indicates a psychological state in which the intensity of emotion such as jean is low and the intensity of emotion such as jerk is low.

(Modification 2)
With reference to FIG. 19, the auditory impression amount estimation device 1 according to the second modification of the present invention will be described with respect to differences from the first embodiment (see FIG. 1 as appropriate).
The second modification is different from the first embodiment in that the psychological state estimating unit 30 selects the transition probability of the psychological state transition model 21 according to the acoustic analysis value input from the acoustic signal analyzing unit 10.

  In the present modification, the psychological state transition model 21 is a simple model in which the psychological state transitions between “normal” and “impressed” as shown in FIG. The acoustic analysis value is assumed to be “sound pressure level”.

Psychological state transition model 21 may be the same path from "normal" to "excitement", according to the difference between the sound pressure level can be set a plurality of transition probabilities P _12. For example, the psychological state transition model 21, the transition probability P ₁₂ when the difference between the sound pressure level flat _| and _flat, the transition probability P ₁₂ when the difference between the sound pressure level increases _| and _Noboru, the difference in sound pressure level It is possible to set the transition probability P _{12 |}

  This difference in sound pressure level is called a “level difference”. In this modification, the difference in sound pressure level changes greatly between “rising” when the sound pressure level changes greatly from low to high and “falling” when the sound pressure level changes greatly from high to low. Not classified as “flat”.

<Setting transition probability>
A method for setting a transition probability in the psychological state transition model 21 will be described with reference to FIG.
In FIG. 20 (b), the change in the impression of the first subject is shown by a solid line, the change in the impression of the second subject is shown by a broken line, and the change of the impression of the third subject is shown by a one-dot chain line.
In FIG.20 (c), each time of FIG.20 (b) was illustrated as sample points 1-8.

  FIG. 20D illustrates the relationship between the level difference and the change in the impression of the subject at the sample points 1 to 8. In FIG. 20D, the level difference is abbreviated as “flat”, the level difference as “rise”, and the level difference as “down”. In FIG. 20D, a state where the subject's impression is high is illustrated as “high”, and a state where the subject's impression is low is illustrated as “low”. In FIG. 20D, the impression state of the first subject is illustrated as “impression state 1”, the impression state of the second subject is illustrated as “impression state 2”, and the impression state of the third subject is “impression”. Stated 3 ".

First, as shown in FIG. 20A, the sound pressure level of the evaluation acoustic signal is calculated in a predetermined time unit to obtain a level difference indicating a change in the sound pressure level. At this time, as shown in FIG. 20 (b), a change in emotion is input by a fader while allowing a plurality of (for example, three) subjects to listen to an evaluation acoustic signal.
In the following description, it is assumed that the impression of all subjects is low at the start of listening to the evaluation acoustic signal.

As shown in FIG. 20 (d), at sample points 1 and 2, the level difference is “flat” and the impressions of all the subjects have not changed from “low”. For this reason, it is considered that the psychological state is highly likely to maintain “normal”. Therefore, the transition probability P _{11 | flat} when the level difference is “flat” is 3/3 based on the ratio between the number of all subjects and the number of subjects whose impression has not changed from “low”.

Further, at the sample point 3, the level difference is “increased”, and the impression has changed from “low” to “high” in two of the subjects. For this reason, it is considered that there is a high possibility that the psychological state transitions from “normal” to “impressed”. Therefore, the transition probability P _{12 | rise} when the level difference is “rise” is 2/3 based on the ratio between the number of all subjects and the number of subjects whose impression has changed from “low” to “high”.

On the other hand, at the sample point 3, the impression of the remaining one of the subjects has not changed from “low”. Therefore, the transition probability P _{11 | rising} when the level difference is “rising” is 1/3 when expressed by the ratio of all the subjects and the subjects whose impressions have not changed from “low”.

With the above procedure, in the psychological state transition model 21, the level difference is “increased” with the transition probabilities P _{11 | flat} , P _{12 | flat} , P _{21 | flat} , P _{22 | flat} when the level difference is “ _flat ”. Transition probabilities P _{11 | ascending} , P _{12 | ascending} , P _{21 | ascending} , P _{22 | ascending,} and transition probabilities P _{11 | descending} , P _{12 | descending} , P _{21 | descending, P 22 |} it is preferable to set all the _descending and.
It should be noted that the accuracy of the transition probability can be improved by replacing the evaluation acoustic signal with another signal or repeatedly performing subjective evaluation experiments.

<Selection of transition probability>
A method for selecting a transition probability according to an acoustic analysis value will be described with reference to FIG.
In FIG. 21, the case where the psychological state is normal is abbreviated as “normal”, and the case where the psychological state is impressed are abbreviated as “feel”.

  As shown in FIG. 21A, the psychological state estimation means 30 divides the sound pressure level of the evaluation acoustic signal input in time series into divided sections in which the level difference is the same. And the psychological state estimation means 30 selects a transition probability for every division | segmentation area according to a level difference.

That is, since the level difference is “flat” in the first divided section, the psychological state estimation means 30 has transitions corresponding to “flat” among the 12 types of transition probabilities set in the psychological state transition model 21. Probabilities P _{11 | Plane} , P _{12 | Plane} , P _{21 | Plane} , P _{22 | Plane} are selected. Then, the psychological state estimating means 30 is the same as in the first embodiment in the first divided section based on the selected transition probabilities P _{11 | flat} , P _{12 | flat} , P _{21 | flat} , P _{22 |} Estimate the listener's psychological state.

The psychological state estimation means 30 selects the transition probabilities P _{11 | ascending} , P _{12 | ascending} , P _{21 | ascending} , P _{22 |} As in the first embodiment, the psychological state of the listener in the second divided section is estimated. Further, the psychological state estimating means 30 selects the transition probabilities P _{11 | down} , P _{12 | down} , P _{21 | down} , P _{22 | down} because the level difference is “down” in the fourth divided section. As in the first embodiment, the psychological state of the listener in the fourth divided section is estimated.
In addition, since the transition probability is selected for the third and fifth divided sections in the same manner as the first divided section, the description is omitted.

As described above, it is known that when there is a sudden change in volume, impression is aroused (see Reference 2). Therefore, the psychological state estimating means 30 can select the transition probabilities P _{12 |} normal, P _{12 | ascending} , P _{12 | descending} from “normal” to “impressed” according to the level difference. As a result, psychological state estimation means 30, for example, at the timing when the sound pressure level has changed greatly, can increase the transition probabilities P ₁₂ to the "impression" from the "normal".

  As shown in FIG. 21 (b), the condition for selecting the transition probability is not limited to only one type of sound pressure level, and other acoustic feature quantities such as an average sound pressure level, biological information, or realistic feeling estimated values are combined. May be. That is, the auditory impression amount estimation device 1 according to the modified example 2 can be applied to the second to fourth embodiments as in the first embodiment.

(Other variations)
In each of the above-described embodiments, the database storing the psychological state transition model 21 and the probability distribution model 23 has been described as being integrated. However, the present invention is not limited to this. That is, according to the present invention, the psychological state transition model database that stores the psychological state transition model 21 and the probability distribution model database that stores the probability distribution model 23 may be configured separately.

  In each of the above-described embodiments, it has been described that the realistic sensation estimated value is calculated. However, the present invention may not calculate the realistic sensation estimated value. In this case, the auditory impression amount estimation device 1 does not include the realistic sensation estimated value calculation unit 53, and the auditory impression amount presentation unit 55 presents only the auditory impression amount.

  In each of the above-described embodiments, the psychological state is described as the extraction condition, but the present invention is not limited to this. That is, the probability distribution model 23 may be further finely classified according to acoustic feature amounts such as sound pressure level and interaural cross-correlation and subject preference. Then, the probability distribution model extraction unit 40 generates an extraction condition that can extract model data that satisfies the acoustic analysis value of the acoustic signal analysis unit 10 and the setting parameters of the setting parameter input unit 60.

  In the second embodiment, the probability distribution model in which the acoustic analysis value and the auditory impression amount are associated one-to-one has been described. However, the probability distribution model of the present invention is not limited to this. That is, the present invention may use a probability distribution model in which one acoustic analysis value is associated with a plurality of acoustic analysis values as conditional probabilities.

Furthermore, in the present invention, when evaluating the overall acoustic quality from a plurality of auditory impression amounts, a plurality of probability distribution models 23 can be constructed using a neural network or multiple regression analysis. In this case, regarding the degree of influence of each auditory impression amount on the overall impression, an objective evaluation of the acoustic quality reflecting the psychological state can be performed by switching the probability distribution model 23 depending on the psychological state of the listener.
The overall impression is an overall impression with respect to individual impressions, and can be said to be favorable or appropriate.

1, 1A, 1B, 1C Auditory impression estimation device 10 Acoustic signal analysis means 20, 20B, 20C Database (psychological state transition model database, probability distribution model database)
30, 30B, 30C Psychological state estimation means 40, 40A Probability distribution model extraction means (extraction condition generation means)
50, 50A Auditory impression amount estimation means 51, 51A Auditory impression amount calculation means 53, 53A Presence sense value calculation means 55, 55A Auditory impression amount presentation means 60 Setting parameter input means 70 Biological information measurement means

Claims (6)

A hearing impression amount estimation device for estimating an auditory impression amount indicating a degree of a characteristic of a sound felt by a listener when an acoustic signal is reproduced,
The acoustic signal is input, and an acoustic analysis value that is an acoustic feature value of the input acoustic signal is obtained by acoustic analysis of the acoustic signal; and
A psychological state transition model database that prestores a psychological state transition model in which the acoustic feature amount is an evaluation condition, and a psychological state transition model in which a subjective evaluation experiment is performed for each of the evaluation conditions to set a psychological state transition probability of the subject
A psychological state estimating means for estimating the psychological state of the listener based on a transition probability of the psychological state set in the psychological state transition model and a random number generated by random number generation processing;
A probability distribution model database for storing a probability distribution model in which a subjective evaluation experiment is performed for each psychological state of the subject and the auditory impression amount and the acoustic feature amount are associated in advance;
Extraction condition generation means for generating an extraction condition that is a psychological state that matches between the psychological state of the subject included in the probability distribution model and the psychological state of the listener estimated by the psychological state estimation means;
The auditory impression amount and the acoustic feature amount that match the extraction condition are extracted from the probability distribution model, and a probability function set in advance for each section of the acoustic feature amount in the extracted auditory impression amount and the acoustic feature amount. By applying the auditory impression amount calculating means for calculating the auditory impression amount corresponding to the acoustic analysis value,
An auditory impression amount estimation device comprising: A setting parameter input means for inputting the preference of the listener as a setting parameter depending on the listener;
The auditory impression amount estimation apparatus according to claim 1, wherein the psychological state transition model database stores a psychological state transition model in which the evaluation feature includes the acoustic feature amount and the subject's preference. Further comprising biological information measuring means for measuring the biological information of the listener,
The auditory impression amount estimation apparatus according to claim 2, wherein the psychological state transition model database further stores a psychological state transition model in which the biological condition information of the subject is included in the evaluation condition.   The auditory impression amount estimation according to any one of claims 1 to 3, wherein the psychological state transition model database stores a psychological state transition model in which the psychological state of the subject gradually changes. apparatus.   The auditory impression according to any one of claims 1 to 3, wherein the psychological state transition model database stores a psychological state transition model in which an internal state included in the psychological state of the subject changes. Quantity estimation device. In order to estimate the auditory impression amount indicating the degree of the feature of the sound felt by the listener when the acoustic signal is reproduced, the acoustic feature amount is used as an evaluation condition, and a subjective evaluation experiment is performed for each of the evaluation conditions. A psychological state transition model database for preliminarily storing a psychological state transition model for which a transition probability between psychological states of the subject is stored, and subjecting the auditory impression amount and the acoustic feature amount in advance by performing a subjective evaluation experiment for each psychological state of the subject A computer comprising a probability distribution model database for storing the attached probability distribution model,
Acoustic signal analysis means for inputting the acoustic signal and obtaining an acoustic analysis value which is an acoustic feature value of the inputted acoustic signal by acoustic analysis of the acoustic signal;
Psychological state estimation means for estimating the psychological state of the listener based on the transition probability of the psychological state set in the psychological state transition model and the random number generated in the random number generation process;
Extraction condition generating means for generating an extraction condition that is a psychological state that matches between the psychological state of the subject included in the probability distribution model and the psychological state of the listener estimated by the psychological state estimating means;
The auditory impression amount and the acoustic feature amount that match the extraction condition are extracted from the probability distribution model, and a probability function set in advance for each section of the acoustic feature amount in the extracted auditory impression amount and the acoustic feature amount. By applying the auditory impression amount calculating means for calculating the auditory impression amount corresponding to the acoustic analysis value,
Auditory impression estimation program to function as

Images