Title of Invention: Information Processing Apparatus and Method, and Program
Technical field
[0001]
　The present technology relates to an information processing device, method, and program, and more particularly to an information processing device, method, and program that enable 3D Audio content to be easily created.
Background technology
[0002]
　Conventionally, the MPEG (Moving Picture Experts Group)-H 3D Audio standard is known (see Non-Patent Document 1 and Non-Patent Document 2, for example).
[0003]
　3D Audio, which is handled by the MPEG-H 3D Audio standard, can reproduce the direction, distance, and spread of sound in three dimensions, enabling audio playback with a more realistic feel than conventional stereo playback. Become.
prior art documents
Non-patent literature
[0004]
Non-Patent Document 1: ISO/IEC 23008-3, MPEG-H 3D Audio
Non-Patent Document 2: ISO/IEC 23008-3:2015/AMENDMENT3, MPEG-H 3D Audio Phase 2
SUMMARY OF THE INVENTION
Problems to be Solved by the Invention
[0005]
　However, with 3D Audio, the time cost of producing content (3D Audio content) increases.
[0006]
　For example, in 3D Audio, the number of dimensions of object position information, ie, sound source position information, is higher than in stereo (3D Audio is three-dimensional, and stereo is two-dimensional). Therefore, in 3D Audio, the time cost is high, especially in the task of determining the parameters that make up the metadata for each object, such as the horizontal and vertical angles that indicate the position of the object, the distance, and the gain for the object. .
[0007]
　In addition, compared to stereo content, 3D Audio content is overwhelmingly fewer in terms of both content and creators. Therefore, the current situation is that there are few high-quality 3D Audio contents.
[0008]
　In view of the above, it is desired to make it possible to produce 3D Audio content of sufficient quality more easily, that is, in a short period of time.
[0009]
　The present technology has been developed in view of such circumstances, and enables 3D Audio content to be easily produced.
Means to solve problems
[0010]
　An information processing apparatus according to one aspect of the present technology includes a determination unit that determines one or more parameters that configure metadata of an object based on one or more attribute information of the object.
[0011]
　An information processing method or program according to one aspect of the present technology includes a step of determining one or more parameters constituting metadata of an object based on one or more attribute information of the object.
[0012]
　In one aspect of the present technology, one or more parameters configuring metadata of an object are determined based on one or more attribute information of the object.
Brief description of the drawing
[0013]
1 is a diagram illustrating determination of metadata using a decision tree; FIG.
FIG. 2 is a diagram for explaining distribution adjustment of metadata; FIG.
3 is a diagram for explaining distribution adjustment of metadata; FIG.
4 is a diagram showing a configuration example of an information processing apparatus; FIG.
5 is a flowchart for explaining metadata determination processing; FIG.
6 is a diagram showing a configuration example of a computer; FIG.
MODE FOR CARRYING OUT THE INVENTION
[0014]
　Hereinafter, embodiments to which the present technology is applied will be described with reference to the drawings.
[0015]

　This technology determines metadata for each object, more specifically, one or a plurality of parameters that constitute the metadata, thereby making it easier, that is, in a short time. It enables the production of sufficiently high-quality 3D Audio content.
[0016]
　In particular, the present technology has the following features (F1) to (F5).
[0017]
　Feature (F1): Determining metadata from information given to each object
　Feature (F2): Determining metadata from an audio signal for each object
　Feature (F3): Determining metadata from other information
　( F4): Modification of metadata to achieve desired distribution
　Feature (F5): Multiple decision patterns in metadata
[0018]
　With this technology, high-quality 3D Audio content can be produced in a short period of time by determining object metadata from the following information. As a result, it is expected that the number of creators of high-quality 3D Audio content and 3D Audio content will increase.
[0019]
　A specific example of determination of object metadata will be described below for each piece of information used for automatic calculation (automatic determination) of object metadata.
[0020]
　Note that in the present technology, the object may be any object, such as an audio object or an image object, as long as it has parameters such as position information and gain as metadata.
[0021]
　For example, the present technology can be applied to the case of determining the metadata of an image object based on attribute information indicating the attributes of the image object, using parameters indicating the position of the image object such as a 3D model in space as metadata. It is possible. The attribute information of the image object can be the type (type) and priority of the image object.
[0022]
　In the following, the case where the object is an audio object will be described as an example.
[0023]
　If the object is an audio object, the metadata consists of one or more parameters (information) used in processing for reproducing sound based on the audio signal of the object, more specifically in rendering processing of the object.
[0024]
　Specifically, the metadata includes, for example, a horizontal angle, a vertical angle, and a distance, which form position information indicating the position of the object in a three-dimensional space, and the gain of the audio signal of the object. Note that although an example in which the metadata consists of a total of four parameters, ie, the horizontal angle, the vertical angle, the distance, and the gain, is described below, the number of parameters in the metadata may be any number as long as it is one or more.
[0025]
　The horizontal angle is the angle indicating the horizontal position of the object viewed from a predetermined reference position such as the user's position, and the vertical angle is the angle indicating the vertical position of the object viewed from the reference position. . Also, the distance constituting the position information is the distance from the reference position to the object.
[0026]
(Determination from Information Assigned to
　Each Object) First, a method of determining metadata, more specifically parameters of metadata, from information assigned to each object will be described.
[0027]
　Metadata of an object is often determined based on information related to attributes of the object, such as instrument information, sound effect information, and priority information. However, the rules for determining metadata according to the musical instrument information and the like differ depending on the creator of the 3D Audio content.
[0028]
　Instrument information is information indicating the type of object (sound source) such as vocal "vocal", drums "drums", bass "bass", guitar "guitar", piano "piano", etc. That is, it is information indicating the sound source type. More specifically, the musical instrument information is information indicating the type of object such as musical instrument, voice part, gender of voice such as male or female, that is, information indicating attributes of the object itself serving as a sound source.
[0029]
　For example, in the case of a certain producer, for an object whose instrument information is "vocal", the horizontal angle that constitutes the metadata is often set to 0 degrees (0°), and the gain is set to a value greater than 1.0. tend to be Further, for example, in the case of a certain producer, the vertical angle constituting the metadata for an object whose instrument information is "bass" is often a negative value.
[0030]
　In this way, the value of the parameter that constitutes the metadata for the musical instrument information and the range that the value of the parameter that constitutes the metadata may take may be determined to some extent by the creator of the 3D Audio content. In such cases, it is possible to determine the object's metadata from the instrument information.
[0031]
　The acoustic effect information is information indicating acoustic effects such as effects added to the audio signal of the object, that is, applied to the audio signal. In other words, the sound effect information is information indicating attributes related to the sound effect of the object. In particular, among acoustic effect information indicating acoustic effects, information indicating reverberation effects as acoustic effects, ie, reverberation characteristics, is referred to as reverberation information, and information indicating acoustic effects other than reverberation effects is referred to as acoustic information.
[0032]
　Reverberation information is information that indicates the reverberation characteristics of the audio signal, such as dry "dry", short reverb "short reverb", long reverb "long reverb", etc. . For example, "dry" indicates that no reverberation effect is applied to the audio signal.
[0033]
　For example, in the case of a certain producer, for an object whose reverberation information is "dry", the horizontal angle that constitutes the metadata is often a value within the range of -90 degrees to 90 degrees, and the reverberation information is For objects that are "long reverb", the vertical angle that constitutes metadata is often positive.
[0034]
　In this way, for reverberation information as well as for musical instrument information, the values ​​of metadata parameters for reverberation information and the possible ranges of metadata parameter values ​​may be determined to some extent for each producer. Therefore, it is possible to determine the metadata even using the reverberation information.
[0035]
　Further, the acoustic information is information indicating acoustic effects other than reverberation to be added (given) to the audio signal of the object, such as natural "natural" and distortion "dist". Note that "natural" indicates that no effect is applied to the audio signal.
[0036]
　For example, in the case of a certain producer, for objects whose acoustic information is "natural", the horizontal angle that constitutes the metadata is often a value within the range of -90 degrees to 90 degrees, and the acoustic information is For objects that are 'dist', the vertical angle that constitutes the metadata is often a positive value. Therefore, it is possible to determine the metadata even using the acoustic information.
[0037]
　Furthermore, priority information is information indicating the priority of an object. For example, the priority information is any value from 0 to 7, and the larger the value, the higher the priority of the object. It can be said that such a priority is also information indicating an attribute of an object.
[0038]
　For example, in the case of a certain creator, for objects whose priority information value is less than 6, the horizontal angle that constitutes the metadata is often set to a value outside the range of -30 degrees to 30 degrees. Unless the value of the degree information is 6 or more, the vertical angle tends to be less than 0 degree. Therefore, it is possible to determine the metadata using the priority information as well.
[0039]
　By using instrument information, reverberation information, acoustic information, and priority information in this way, it is possible to determine object metadata for each producer.
[0040]
　In addition, if the object was recorded with a certain speaker arrangement, ie, a certain channel configuration, the metadata is often determined based on the channel information of the object.
[0041]
　Here, the channel information corresponds to the speaker to which the object audio signal is supplied, such as stereo L and R, 5.1 channel (5.1ch) C, L, R, Ls, and Rs. Channel, that is, information indicating attributes related to the channel of an object.
[0042]
　For example, for a producer, for an R-channel object whose channel information is "stereo R" and an L-channel object whose channel information is "stereo L", metadata is provided in the L and R channels. Constituent horizontal angles have a positive/negative inversion relationship, and vertical angles are often the same angle.
[0043]
　For example, for an Ls channel object whose channel information is "5.1 channel Ls" and an Rs channel object whose channel information is "5.1 channel Rs", horizontal The angles have a positive/negative inversion relationship, and the vertical angles are often the same angle.
[0044]
　From these facts, it can be seen that it is possible to determine metadata even using channel information.
[0045]
　In this technology, metadata of each object, more specifically, parameters constituting the metadata are determined based on at least one of instrument information, reverberation information, acoustic information, priority information, and channel information.
[0046]
　Specifically, in the present technology, determination of metadata is performed using, for example, a decision tree, which is a supervised learning method.
[0047]
　In the decision tree, learning data (learning data ).
[0048]
　Learning of a decision tree model is performed by inputting the instrument information, the reverberation information, the acoustic information, the priority information, and the channel information and outputting the metadata. By using the decision tree model obtained in this way, it is possible to easily determine (predict) the metadata of each object.
[0049]
　FIG. 1 shows an example of a decision tree that determines the horizontal angle and vertical angle that constitute the metadata.
[0050]
　In the example shown in FIG. 1, for each object in the decision tree, it is first determined whether or not the instrument information is "vocal".
[0051]
　If it is determined that the instrument information is "vocal", then it is determined whether or not the "reverberation information" is "dry". At this time, if the "reverberation information" is determined to be "dry", it is determined that the horizontal angle of the object is 0 degrees and the vertical angle is 0 degrees, and the decision tree processing ends.
[0052]
　On the other hand, when it is determined that the reverberation information is not "dry", it is determined whether the reverberation information is "long reverb". Then, when the reverberation information is determined to be "long reverb", it is determined that the horizontal angle of the object is 0 degrees and the vertical angle is 30 degrees, and the processing of the decision tree ends.
[0053]
　In this way, in the decision tree, decisions are made continuously until the end of the decision tree according to the results of decisions based on each piece of information such as instrument information, reverberation information, acoustic information, priority information, and channel information. horizontal and vertical angles are determined.
[0054]
　If such a decision tree is used, the horizontal and vertical angles that make up the metadata for each object can be determined from information assigned to each object, such as instrument information, reverberation information, acoustic information, priority information, and channel information. It is possible to
[0055]
　Note that the metadata determination method is not limited to the decision tree, and other supervised learning methods such as linear decision, support vector machine, and neural network may be used.
[0056]
(Determination from Audio Signal for
　Each Object) Next, a method for determining metadata from the audio signal of each object will be described.
[0057]
　For example, metadata of an object may be determined based on information such as sound pressure (sound pressure information) and pitch (pitch information) obtained from the audio signal of the object. Since the information such as the sound pressure and pitch (pitch) represents the characteristics of the sound of the object, it can also be said that it is the information representing the attribute of the object.
[0058]
　Specifically, for example, in the case of a certain producer, the higher the sound pressure of the audio signal, the closer the vertical angle constituting the metadata is to a value close to 0 degrees, and the higher the sound pressure, the closer the gain constituting the metadata is. It is often taken as a value less than 1.0.
[0059]
　Furthermore, for example, in the case of a certain producer, when the audio signal is a low frequency signal, the vertical angle constituting the metadata tends to be a negative value, and conversely, when the audio signal is a high frequency signal, the vertical angle is positive. tend to be valued.
[0060]
　Therefore, by adding information about these sound pressures and pitches to the input of the above-described method of determining metadata from information assigned to each object (hereinafter also referred to as a metadata determination method), metadata can improve the determination accuracy of
[0061]
　Specifically, for example, for sound pressure and pitch, feature amounts calculated by the method described below may be added to the input of the above-described metadata determination method, that is, the input of a decision tree or the like.
[0062]
　For example, with respect to sound pressure, the feature amount level(i_obj) calculated by the following equation (1) may be used as one of the inputs for the metadata determination method.
[0063]
[Number 1]

[0064]
　Note that in equation (1), i_obj indicates an object index, and i_sample indicates an audio signal sample index.
[0065]
　In equation (1), pcm(i_obj, i_sample) indicates the sample value of the sample whose index is i_sample in the audio signal of the object whose index is i_obj, and n_sample indicates the total number of samples of the audio signal. showing.
[0066]
　Furthermore, regarding the pitch (pitch), for example, the feature amount level_sub(i_obj, i_band) calculated by the following equation (2) may be used as one of the inputs of the metadata determination method.
[0067]
[Number 2]

[0068]
　Note that the indexes i_obj, i_sample, and n_sample in equation (2) are the same as in equation (1), and i_band is an index indicating a band.
[0069]
　For example, by filtering the audio signal with a bandpass filter, the audio signal of each object is divided into audio signals of three bands of 0 kHz to 2 kHz, 2 kHz to 8 kHz, and 8 kHz to 15 kHz. Also, here, the audio signal of each band is represented as pcm_sub(i_obj, i_band, i_sample).
[0070]
　Furthermore, index i_band=1 indicates the band from 0 kHz to 2 kHz, index i_band=2 indicates the band from 2 kHz to 8 kHz, and index i_band=3 indicates the band from 8 kHz to 15 kHz.
[0071]
　In such a case, the feature quantity level_sub(i_obj,1), the feature quantity level_sub(i_obj,2), and the feature quantity level_sub(i_obj,3) are obtained by Equation (2) and used as the input for the metadata determination method. .
[0072]
(Determination from Other Information)
　Further, a method for determining metadata from other information will be described.
[0073]
　For example, object metadata may be determined based on information such as the number of 3D Audio content objects, metadata of other objects, object names, and genres of 3D Audio content made up of objects. Therefore, by adding such information to the input of the metadata determination method, the determination accuracy can be improved.
[0074]
　Object names often include information that substitutes for instrument information and channel information, such as the object's instrument name and corresponding channel, that is, information that indicates the attributes of the object. can do.
[0075]
　Information indicating the genre of 3D Audio content such as music, such as jazz, and the number of objects, which is the total number of objects forming the 3D Audio content, are information indicating attributes of content composed of objects. Therefore, information about content attributes, such as genre and the number of objects, can also be used as object attribute information for determining metadata.
[0076]
　For example, in the case of a certain creator, if the number of objects placed in the space (the number of objects) is large, each object will be placed at uneven intervals in the space. They are often placed at intervals.
[0077]
　Therefore, for example, the number of objects that compose the 3D Audio content can be added as one of the inputs for the metadata determination method. In this case, for example, the horizontal angle, vertical angle, and distance forming the metadata are determined so that the objects are arranged at regular or irregular intervals in the space.
[0078]
　Further, for example, in the case of a certain creator, it is often the case that other objects are not arranged at the same position as an object whose position in space has already been determined.
[0079]
　Therefore, for example, metadata of other objects whose metadata has already been determined may also be used as input for the metadata determination method.
[0080]
　Note that the information given to each object described above, the information obtained from the audio signal, and other information such as the number of objects may be used alone as input for the metadata determination method. information may be combined and used as input for metadata determination techniques.
[0081]

　By the way, using the above-described information, it is possible to determine the metadata of the object. However, in 3D Audio content with a small number of objects (hereinafter also simply referred to as content), the parameters of the determined metadata may be determined biased in one place. Such an example is shown in FIG.
[0082]
　In FIG. 2, the horizontal axis indicates the horizontal angle forming the metadata, and the vertical axis indicates the vertical angle forming the metadata.
[0083]
　Also, in FIG. 2, one circle indicates one object, and the pattern attached to each circle differs for each instrument information assigned to the object corresponding to those circles.
[0084]
　Here, circles C11 and C12 indicate objects given "vocal" as instrument information, and circles C13 and C14 indicate objects given "bass" as instrument information. Circles C15 to C20 indicate objects to which piano "piano" is assigned as instrument information.
[0085]
　Each of these circles is placed at a position determined by the horizontal and vertical angles determined by prediction for the corresponding object. That is, the horizontal position of each circle is the position indicated by the horizontal angle of the object corresponding to each circle, and the vertical position of each circle is the vertical angle of the object corresponding to each circle. The position indicated by the angle.
[0086]
　The size of each circle indicates the magnitude (height) of the sound pressure of the audio signal of the object, and the size of the circle increases in proportion to the sound pressure.
[0087]
　Therefore, FIG. 2 shows the distribution of the parameters (metadata) of each object in the parameter space (parameter space) with the horizontal angle and the vertical angle as axes, and the magnitude of the sound pressure of the object signal of each object. It can be said.
[0088]
　For example, content with a small number of objects often includes only important musical instruments such as vocals, piano, and bass as objects, as shown in FIG. The placement of these instruments tends to be front-center for some authors, resulting in a bias in the determined metadata.
[0089]
　In this example, the circles C11 to C18 are concentrated in the center of FIG. 2, and it can be seen that the metadata of the objects corresponding to those circles have close values. In other words, the distribution of the metadata of each object is a distribution that concentrates at positions close to each other in the parameter space. In such a case, if the determined metadata is used as it is for rendering, the resulting content will be of low quality without three-dimensional sound direction, distance, and spread.
[0090]
　Therefore, in this technology, by adjusting the distribution of objects, that is, the distribution of object metadata, it is possible to obtain high-quality content with three-dimensional sound direction, distance, and spread.
[0091]
　In the distribution adjustment, as an input, metadata that has already been determined by the input of the creator or metadata that has been determined by prediction using a decision tree or the like is used. Therefore, it can be applied independently of the metadata determination method described above. In other words, distribution adjustment of metadata can be performed regardless of the method of determining metadata.
[0092]
　Metadata distribution adjustment may be performed by either a manual method (hereinafter referred to as a manual adjustment method) or an automatic method (hereinafter referred to as an automatic adjustment method). Each method will be described below.
[0093]
(Manual adjustment method)
　First, the manual adjustment method of metadata will be described.
[0094]
　In the manual adjustment method, the value of the parameter in the metadata of the object is added by a predetermined value for addition, multiplied by a predetermined value for multiplication, or both addition and multiplication are performed. distribution adjustment of metadata is performed.
[0095]
　For example, the value to be added in the addition process of the manual adjustment method and the value to be multiplied in the multiplication process can be adjusted by operating bars, etc. on the 3D Audio content creation tool of the GUI (Graphical User Interface). .
[0096]
　This makes it possible to adjust the distribution of all objects, that is, the distribution of metadata, by widening or narrowing it while maintaining the positional relationship of objects. can.
[0097]
　Here, for example, when adjusting the distribution of metadata only by addition processing, negative values ​​are added to parameters with negative values ​​among metadata parameters, and parameters with positive values ​​are added. By adding a positive value to , the metadata distribution can be adjusted (corrected) to have a more spatial spread.
[0098]
　Also, for example, when adjusting the distribution of metadata only by addition processing, by adding the same value to each parameter, it is possible to move the objects in parallel while maintaining the positional relationship of each object. metadata distribution adjustment can be realized.
[0099]
(Automatic Adjustment Method)
　In the automatic adjustment method, each object is viewed as a vector indicated by a horizontal angle, a vertical angle, and a distance, which constitutes metadata. A vector having such a horizontal angle, vertical angle, and distance as elements is hereinafter referred to as an object vector.
[0100]
　In the automatic adjustment method, the mean value of the object vectors of all objects is obtained as the object mean vector.
[0101]
　Then, difference vectors between the object average vector and each of the object vectors are obtained, and a vector having the mean square values ​​of the difference vectors as elements is obtained. That is, for each of the horizontal angle, vertical angle, and distance, a vector having as elements the mean square value of the difference between the respective values ​​of the object from the average value is obtained.
[0102]
　The mean square value for each of the horizontal angle, vertical angle and distance thus obtained corresponds to the variance for each of the horizontal angle, vertical angle and distance, and the horizontal angle, vertical angle and distance is called an object variance vector. It can be said that the object distribution vector indicates the distribution of metadata of a plurality of objects.
[0103]
　Furthermore, the metadata is adjusted so that the object variance vector obtained by the above calculation becomes a desired value, that is, a target variance value. When adjusting the metadata, one parameter (element) such as a horizontal angle that constitutes the metadata may be adjusted, or a plurality of parameters may be adjusted. Also, all parameters that make up the metadata may be adjusted.
[0104]
　Here, the desired target value of the object variance vector may be obtained, for example, by obtaining object variance vectors for a plurality of 3D Audio contents in advance and using the average value of those object variance vectors.
[0105]
　Similarly, in the auto-tuning method, the metadata may be adjusted so that the object mean vector is the target value, or the metadata may be adjusted so that both the object mean vector and the object variance vector are target values. may be adjusted.
[0106]
　In the automatic adjustment method, the values ​​of the object mean vector and object variance vector targeted for adjustment should be obtained in advance by learning, etc. for each genre of 3D Audio content, each creator, and each number of objects in 3D Audio content. can be By doing so, it is possible to realize distribution adjustment suitable for the content genre and distribution adjustment that reflects the creator's likeness.
[0107]
　Also, the sound pressure of each object may be weighted with respect to the object vector. That is, the object vector obtained for the object may be multiplied by a weight according to the sound pressure of the audio signal of the object, and the resulting vector may be used as the final object vector.
[0108]
　In this case, the sound pressure distribution can be set to a desired value, that is, a target sound pressure distribution, and the metadata can be adjusted (corrected) with higher quality. This is because audio content with moderate sound pressure distribution is considered to be of good quality.
[0109]
　In addition, in the distribution adjustment of metadata by these manual adjustment methods and automatic adjustment methods, some objects may be excluded from adjustment targets.
[0110]
　For objects excluded from distribution adjustment, the object's metadata is not used in calculating the object mean vector. However, the metadata of excluded objects may be used in calculating the object mean vector.
[0111]
　For example, an object whose instrument information is "vocal" is often important in the content, and the quality may be higher if the distribution of the metadata is concentrated in one place. In such a case, objects whose instrument information is "vocal" may be excluded from metadata distribution adjustment.
[0112]
　Objects that are not subject to metadata distribution adjustment may be objects that indicate predetermined information (values, etc.) assigned to each object, such as musical instrument information. It may be an object specified by, for example.
[0113]
　By the distribution adjustment described above, the distribution shown in FIG. 2 becomes, for example, as shown in FIG. In FIG. 3, portions corresponding to those in FIG. 2 are denoted by the same reference numerals, and description thereof will be omitted as appropriate. Also in FIG. 3, the horizontal axis indicates the horizontal angle that constitutes the metadata, and the vertical axis indicates the vertical angle that constitutes the metadata.
[0114]
　In the example of FIG. 3, the objects whose instrument information is "vocal", that is, the objects indicated by circles C11 and C12, are excluded from metadata distribution adjustment.
[0115]
　As shown in FIG. 3, by adjusting the distribution of metadata, each object, that is, the metadata of each object is distributed more appropriately apart than in the case shown in FIG. As a result, it is possible to obtain high-quality content with three-dimensional sound direction, distance, and spread.
[0116]

　Next, an information processing apparatus for determining metadata by the above-described metadata determination method and performing distribution adjustment of the determined metadata will be described.
[0117]
　For example, when a method of determining metadata using a decision tree is used as the method of determining metadata, the information processing apparatus is configured as shown in FIG.
[0118]
　The information processing device 11 shown in FIG. 4 has a metadata determining section 21 and a distribution adjusting section 22 .
[0119]
　The metadata determining unit 21 determines metadata of each object by prediction based on information about the attribute of the object supplied from the outside, that is, one or more attribute information of the object, and determines the determined metadata of each object. to output Note that the number of objects whose metadata is to be determined may be one or more than one, but here the metadata is determined for a plurality of objects.
[0120]
　The object attribute information is at least one of instrument information, reverberation information, acoustic information, priority information, channel information, number of objects, metadata of other objects, object name, and information indicating genre. is. The metadata determination unit 21 is also supplied with an audio signal for calculating feature amounts relating to sound pressure and pitch as object attribute information.
[0121]
　The metadata determination unit 21 also has a decision tree processing unit 31 . The metadata determining unit 21 appropriately calculates feature amounts relating to sound pressure and pitch as attribute information of the object based on the audio signal, and determines the calculated feature amount and the attribute information of the object supplied from the outside. is input to the decision tree processing unit 31 . The attribute information input to the decision tree processing unit 31 may be one, or may be plural.
[0122]
　The decision tree processing unit 31 performs a process of determining metadata using a decision tree based on the input attribute information of the object, and supplies the metadata of each object obtained as the determination result to the distribution adjustment unit 22. . The decision tree processing unit 31 holds a decision tree (decision tree model) obtained by learning in advance.
[0123]
　Although an example in which the horizontal angle, the vertical angle, and the distance are determined as metadata parameters in the decision tree processing unit 31 will be described here, the determined parameters may include the gain. Alternatively, any one or more parameters among a plurality of parameters forming the metadata may be determined by the decision tree processing section 31 .
[0124]
　The distribution adjustment unit 22 performs the distribution adjustment described above on the metadata of each of the plurality of objects supplied from the decision tree processing unit 31, and outputs the metadata after the distribution adjustment as the final metadata of each object to a later stage. supply (output).
[0125]
　The distribution adjustment unit 22 has an object dispersion vector calculation unit 32 , a coefficient vector calculation unit 33 and a coefficient vector application unit 34 .
[0126]
　The object variance vector calculation unit 32 sets a vector having elements of the horizontal angle, the vertical angle, and the distance that constitute the metadata of each object supplied from the decision tree processing unit 31 as an object vector, and calculates the object vector of each object. Find the object mean vector based on Furthermore, the object variance vector calculation unit 32 calculates object variance vectors based on the obtained object average vector and each object vector, and supplies the object variance vectors to the coefficient vector calculation unit 33 .
[0127]
　The coefficient vector calculation unit 33 converts each element of a predetermined value vector having predetermined values ​​obtained in advance for each of the horizontal angle, the vertical angle, and the distance into the object dispersion vector supplied from the object dispersion vector calculation unit 32. By dividing by each of the elements of , a coefficient vector having coefficients for each of the horizontal angle, the vertical angle, and the distance as elements is calculated and supplied to the coefficient vector application unit 34 .
[0128]
　Here, the predetermined value vector obtained in advance is the target object variance vector, and is obtained by learning for each genre or each creator, for example. Specifically, for example, the value of the target object variance vector is a vector having, as an element, an average value for each element of object variance vectors obtained for a plurality of 3D Audio contents of the same genre.
[0129]
　The coefficient vector application unit 34 calculates distribution-adjusted metadata by multiplying the metadata supplied from the decision tree processing unit 31 by the coefficient vector supplied from the coefficient vector calculation unit 33 for each element. , outputs the obtained metadata to a later stage. The coefficient vector application unit 34 multiplies the metadata by the coefficient vector element by element to adjust the distribution of the metadata. As a result, the distribution of the metadata becomes a distribution corresponding to the target object variance vector.
[0130]
　For example, after the coefficient vector application unit 34, rendering processing is performed based on the audio signal and metadata of each object, and the metadata is manually adjusted by the creator.
[0131]
　Note that not only metadata but also object attribute information such as instrument information is supplied from the decision tree processing unit 31 to the object variance vector calculation unit 32 and the coefficient vector application unit 34, and based on the object attribute information, , objects to be excluded from distribution adjustment may be determined. In this case, the metadata distribution adjustment is not performed for the excluded object, and the metadata determined by the decision tree processing unit 31 is output as it is as the final metadata.
[0132]
　Further, as the metadata distribution adjustment, the object mean vector may be adjusted, or both the object variance vector and the object mean vector may be adjusted. Furthermore, here, an example in which the distribution adjustment unit 22 performs distribution adjustment by an automatic adjustment method has been described. good too.
[0133]
　In such a case, for example, the distribution adjustment unit 22 adds or multiplies the metadata of the object by a predetermined value specified by the creator or the like, performs an operation based on the predetermined value and the metadata, and distributes the distribution. Ask for adjusted metadata. Also in this case, an object specified by the creator or an object determined by the attribute information of the object or the like may be excluded from the distribution adjustment.
[0134]

　Next, the operation of the information processing apparatus 11 shown in FIG. 4 will be described. That is, the metadata determination processing by the information processing device 11 will be described below with reference to the flowchart of FIG.
[0135]
　In step S11 , the decision tree processing unit 31 determines metadata based on the attribute information of the object, and supplies the determination result to the object dispersion vector calculation unit 32 and the coefficient vector application unit 34 .
[0136]
　That is, the metadata determination unit 21 calculates the feature amounts of the sound pressure and the pitch by performing the calculations of the above-described formulas (1) and (2) based on the audio signal supplied as necessary. . Then, the metadata determination unit 21 inputs the calculated feature amount, musical instrument information supplied from the outside, etc. to the decision tree processing unit 31 as object attribute information.
[0137]
　The decision tree processing unit 31 performs processing for determining metadata using a decision tree based on the supplied attribute information of the object. The metadata determination unit 21 also supplies object attribute information to the object dispersion vector calculation unit 32 and the coefficient vector application unit 34 as necessary.
[0138]
　In step S12, the object variance vector calculation unit 32 obtains an object average vector based on the metadata of each object supplied from the decision tree processing unit 31, and calculates an object variance vector from the object average vector and the object vector, It is supplied to the coefficient vector calculator 33 .
[0139]
　In step S13, the coefficient vector calculation unit 33 calculates a vector having predetermined values ​​obtained in advance for each of the horizontal angle, the vertical angle, and the distance as elements, that is, a target object dispersion vector obtained in advance. A coefficient vector is calculated by dividing each element by the object variance vector supplied from the unit 32 , and supplied to the coefficient vector application unit 34 .
[0140]
　In step S14, the coefficient vector application unit 34 adjusts the distribution of the metadata supplied from the decision tree processing unit 31 based on the coefficient vector supplied from the coefficient vector calculation unit 33. is output, and the metadata determination process ends.
[0141]
　For example, the coefficient vector application unit 34 adjusts the distribution of metadata by multiplying the metadata by the coefficient vector for each element. Note that, as described above, predetermined objects may be excluded from metadata distribution adjustment.
[0142]
　As described above, the information processing apparatus 11 determines the metadata of each object based on the attribute information of the object, and adjusts the distribution of the metadata. By doing so, the creator does not need to specify (input) the metadata of each object one by one, so it is possible to easily create high-quality 3D Audio content in a short time.
[0143]

　A program for causing a computer to execute a process including 　determining one or more parameters constituting metadata of an object based on one or more attribute information of the object .