JP4319805B2 - Music data impression system and program - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a music data impression system and program suitable for constructing a music database of a system for retrieving music data from music impressions.
[0002]
[Prior art]
2. Description of the Related Art In recent years, with the development of network infrastructure and the spread of mobile terminal devices such as mobile phones, the progress of music data distribution technology, such as mobile distribution of music data via the Internet, has been remarkable. As represented by a karaoke system or the like, in order to search desired music data from a music database, a text search such as a song code, a song name, a singer name, a composer name, etc. is generally used. On the other hand, when bibliographic items such as song names and singer names are unknown, the user can use fuzzy inference etc. by entering the impression of the song that the user wants such as “fun” or “sad” as a search key. Various music selection systems that enable music selection close to the user's preference have been proposed (for example, JP-A-5-257487).
[0003]
In such a system, it is necessary to make an impression to give the impression to the music data of each song constituting the music database. Conventionally, the impression given to such music data is determined by an average value such as the appearance frequency for the entire music, paying attention to various characteristic parameters such as the tempo, pitch, and pitch difference of the song.
[0004]
[Problems to be solved by the invention]
However, there is a climax in the song, and in the case of a song with a pleasant fast tempo, the tempo may be intentionally reduced in the first half. Therefore, when the impression of a song is obtained with the average value for the entire song as described above, there is a problem that changes in the middle of the song are completely ignored and the impression that the song originally has cannot be obtained.
[0005]
The present invention has been made in view of the above points, and provides a music data impression system and program capable of appropriately obtaining an impression of a song based on the structure of the song and adding the impression to the music data. For the purpose.
[0006]
[Means for Solving the Problems]
A music data impression system according to the present invention includes a group weighting processing unit that divides music data into a plurality of groups based on GTTM (Generative Theory of Tonal Music) and obtains a weight of each group, A feature parameter extraction unit that extracts a feature parameter indicating the feature of the music for each group, and a strength of relevance of one or more concepts associated by the feature parameter from the extracted feature parameter for each group A concept activity value calculator that calculates an activity value of the concept of the music data by combining the activity values of the concept of each group with the weights of the groups and combining the activity values indicating the activity values indicating the music data An impression imparting processing unit for imparting an impression to the music data based on the activity value of the concept of the music data obtained Characterized by comprising.
[0007]
A music data impression program according to the present invention includes: a group weighting step of dividing music data into a plurality of groups based on GTTM (Generative Theory of Tonal Music) to obtain weights of each group; A feature parameter extracting step for extracting a feature parameter indicating the music feature for each group; calculating an activity value of one or more concepts from the extracted feature parameters for each group; and A concept activity value calculating step of calculating an activity value of the concept of the music data by combining the activity values by weighting the weights of the groups, and the music based on the calculated activity value of the concept of the music data It was configured to cause the computer to execute an impression step for giving the impression to the data Than is.
[0008]
According to the present invention, music data is divided into groups, the weights of the songs in each group are obtained, the activity values of the concepts are obtained from the feature parameters extracted for each group, and the activity values of the concepts of each group are obtained. The weight of the music data is combined to determine the activity value of the total concept of the music data, and based on this, the impression is given to the music data, which reflects the evaluation of important parts of the music Thus, it is possible to make an impression that captures the characteristics of the music well.
[0009]
If the group weighting processing unit divides the music data into groups by hierarchy and determines the weight of each group so that the weight becomes higher in the lower hierarchy , the characteristic portion can be emphasized more. it can.
[0010]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings.
FIG. 1 is a diagram showing a configuration of a music data impression system according to the present embodiment.
Music data such as MIDI and note data input from an input device (not shown) or a receiving device is supplied to the group weighting processing unit 1 and the feature parameter extraction unit 2. The group weighting processing unit 1 divides input music data into groups and obtains weights for each group. In this system, music data is grouped by hierarchy and the weight of each group is obtained. The feature parameter extraction unit 2 extracts various feature parameters from the music data for each group obtained by the group weighting processing unit 1. The characteristic parameters include, for example, pitch average high to low, or low to high, pitch average long to short, or short to long, pitch variance high to low, or low to low. There are changes to high, tempo, etc.
[0011]
The weight of each group obtained by the group weight processing unit 1 and the feature parameter extracted by the feature parameter extraction unit 2 are input to the concept activation value calculation unit 3. The concept activity value calculation unit 3 calculates an activity value indicating the strength of relevance of one or more concepts associated with the feature parameters from the feature parameters extracted for each group, and also activates the activity values of the concepts of each group. Are weighted by the weight of each group and synthesized to calculate the activity value of the total concept of the music data. The calculated activity value of the concept of music data is input to the impression imparting processing unit 4. The impression imparting processing unit 4 imparts the impression to the music data from the activation value of the concept of the music data. The music data and the impression given thereto are stored in the music database 5. The group weighting processing unit 1, the feature parameter extraction unit 2, the concept activation value calculation unit 3 and the impression imparting processing unit 4 execute the processing of each unit by a computer, so that the group weighting step, the feature parameter processing step, the concept activation value are performed. You may comprise by the program provided with the calculation step and the impression provision step.
[0012]
Next, details of processing of each part of the system configured as described above will be described.
FIG. 2 is a diagram illustrating an example of input music data. The music data is composed of note data. The group weighting processing unit 1 performs grouping and weighting by analyzing the music structure of the input music data. As such a music structure analysis technique, for example, GTTM (Generative Theory of Tonal Music) is known. In this system, the group weighting process is executed by executing the following process based on GTTM.
[0013]
First, based on a predetermined rule, a beat strength is assigned to each note constituting the music. For example, if a beat having the highest appearance frequency in the music is a basic beat (in the example of FIG. 2, an eighth note is a basic beat), and then a note is present in the basic beat portion, it is relatively long. A strong beat is determined based on the beat and the interval between those beats. Based on the strength of these beats, weights w1 to w12 are assigned to the notes at the lowest level.
[0014]
Next, the rhythm contained in the note data is analyzed. In the example of FIG. 2, it is detected that there is a rhythm of quarter note−eighth note−eighth note, and based on this, a group of syllable structures is obtained as shown in FIG. 3. Next, in order from the group of the lowest hierarchy of the obtained syllable structure, weights are added to each group while leaving the most important note in each group in the upper group. Which note is considered important depends on the rule, but for example, a higher note, the first note in the group, etc. can be prioritized. Finally, as shown in FIG. 4, a tree structure in which branches gradually converge from the lower layer to the upper layer is obtained, and weights are obtained for the groups corresponding to the branches of each tree . In the example of FIG. 4, the weights w1, w2 are 8, 8, 4, 2, and 1 for the eighth note level, quarter note level, half note level, whole note level, and two bar level, respectively. , ... is required.
Next, a calculation example of the weight of each group will be described with reference to FIG.
The weight can be obtained, for example, by converting the weight of the note inherited from the lower group into a ratio in the new group. Here, the ratio of the new group is 1.0, 0.5, 0.25, etc. shown on the right side of FIG. Lowest level of the group, it is not possible to capture the features of the music and too short, the two measures approximately. In the illustrated example, the length of the group is 2 bars, 4 bars, and 8 bars in order from the lowest level. Here, it is considered that the feature of the music appears most in the group of two measures at the lowest level, and the total weight of the group of two measures is 1.0, the total weight of the group of four measures is 0.5, The total weight of a group of 8 bars is 0.25. In this example, the weight w of the group (7, 8) of the seventh bar and the eighth bar is the heaviest at 0.606811.
[0015]
On the other hand, the feature parameter extraction unit 2 analyzes the note data for each group (1, 2), (3, 4),..., (1, 8) as shown in FIG. Degree of transition from high to low, Degree of transition from low to high, Degree of average pitch length from short to long, Degree of shift from short to long, Degree of pitch dispersion from high to low, Degree of shift from low to high, Extract feature parameters such as ...
[0016]
The concept activity value calculation unit 3 calculates concept activity values such as “bright”, “powerful”, and “beautiful” by driving concept knowledge as shown in FIG. Such conceptual knowledge can be appropriately created by a conceptual system of CFS (Conceptual Fuzzy Sets). FIG. 8 is a diagram showing a table for calculating the activity value of a concept created based on such a concept dictionary. Now, paying attention to the concept of “stable” shown in FIG. 8, each term of the word vector extending in the horizontal direction of the “stable” item corresponds to the feature parameter, and the inner product of the word vector and the feature parameter. To obtain the activity value for the concept “stable”. Similarly, for other concepts “unstable”, “lightness”,..., The inner product of the word vector and the feature parameter is obtained.
[0017]
As a result, as shown in FIG. 9, a conceptual activation value is obtained for each group (1, 2), (3,4),..., (1, 8). However, the active value of the concept of each group is multiplied by the weight w of that group. That is, the activation value of the concept of the group (1, 2) is a result of multiplying the inner product of the feature parameter and the word vector by the weight w = 0.146199 of the group (1, 2). When the activity values of the concepts of the groups (1, 2), (3, 4),. . The total obtained in the right column of FIG. 9 indicates the activity value of the concept of the entire music data.
[0018]
The activity values of these concepts may be added to the music data as they are as “impressions”, but in this system, these concepts are further positioned as intermediate concepts, and from the activity values of these intermediate concepts, further “joy”, The activity value of the general concept such as “sadness” is obtained. FIG. 10 shows an example of a table for obtaining the general concept from the intermediate concept. The active value of the general concept “joy” is obtained by calculating the inner product of the word vector corresponding to the general concept “joy” and the active value of the total intermediate concept shown in FIG. Further, as shown in FIG. 11, the activation value of the general concept may be obtained from the activation value of the intermediate concept for each group, and these may be added to all groups to obtain the activation value of the general concept. .
[0019]
The impression giving process 4 associates the activation value of the general concept obtained in this way, or the activation value of the general concept and the intermediate concept with the music data as “impression” and registers it in the music database 5.
[0020]
【The invention's effect】
As described above, according to the present invention, music data is divided into groups, the weights of the songs in each group are obtained, the activity values of the concepts are obtained from the feature parameters extracted for each group, and the activity of the concepts in each group is obtained. By combining the values weighted by the weights of each group, the activity value of the total concept of the music data is obtained, and based on this, the impression is given to the music data. Reflecting the evaluation of the part, there is an effect that it is possible to make an impression that satisfactorily captures the characteristics of the music.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of a music data impression system according to an embodiment of the present invention.
FIG. 2 is a diagram showing music data to be impressed by the system.
FIG. 3 is a diagram for explaining grouping of the music data.
FIG. 4 is a diagram showing a group structure of the music data.
FIG. 5 is a diagram showing the weight of each group in the music data group structure;
FIG. 6 is a diagram showing feature parameters of each group extracted from the music data.
FIG. 7 is a diagram for explaining concept knowledge for obtaining a concept from a feature parameter.
FIG. 8 is a diagram showing a table for obtaining a concept activation value from feature parameters.
FIG. 9 is a diagram showing the activity value of each group concept obtained.
FIG. 10 is a diagram showing a table for obtaining the activation value of the general concept from the activation value of the intermediate concept.
FIG. 11 is a diagram showing the activity value of the general concept obtained for each group.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Group weighting process part 2 ... Feature parameter extraction part 3 ... Concept activity value calculation part 4 ... Impression imparting process part 5 ... Music database

Claims (3)

A group weighting processing unit that divides music data into a plurality of groups based on GTTM (Generative Theory of Tonal Music) and obtains a weight of each group;
A feature parameter extracting unit that extracts a feature parameter indicating a feature of the music for each group from the music data;
For each group, an activity value indicating the strength of relevance of one or more concepts associated with the feature parameters is calculated from the extracted feature parameters, and the activity values of the concepts of each group are calculated for each group. A concept activity value calculation unit that calculates an activity value of the concept of the music data by weighting and combining with a weight;
An impression imparting system for music data, comprising: an impression imparting processing unit for imparting an impression to the music data based on the calculated activity value of the music data concept. 2. The music according to claim 1, wherein the group weight processing unit divides the music data into groups by hierarchy, and determines the weight of each group so that the weight is increased in a lower hierarchy. Data impression system. A group weighting step for dividing the music data into a plurality of groups based on GTTM (Generative Theory of Tonal Music) and calculating the weight of each group;
A feature parameter extracting step for extracting a feature parameter indicating a feature of the music for each group from the music data;
The active value of one or more concepts is calculated from the extracted feature parameters for each group, and the active values of the concepts of each group are weighted with the weight of each group and synthesized. A concept activity value calculating step for calculating an activity value of the concept;
A music data impression program configured to cause a computer to execute an impression imparting step of imparting an impression to the music data based on the calculated activity value of the music data concept.

Images