JP2018025644A - Music key estimation device, and music code progression estimation device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a music code progression device that estimates a Key and code progression of a music.SOLUTION: A music code progression estimation device comprises: a voice data acquisition unit (21 and 17) that acquires voice data on a music; and a code progression estimation unit (11) that analyzes the voice data to estimate a code progression of the music. The code progression estimation unit (11) is configured to: conduct a frequency analysis of the voice data to obtain a chroma vector; determine a Key of the music from the chroma vector; obtain a code in each of a plurality of prescribed small sections from the chroma vector; convert the code obtained for each small section on the basis of the determined Key of the music; and determine a code progression of the music on the basis of combinations of the prescribed number of the codes extracted from a converted code group.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an apparatus for estimating the key of music and an apparatus for estimating chord progression.

  In recent years, research in the field of automatic music analysis has been actively conducted (see Patent Document 1 and the like), and many practical applications have been developed. For example, there are applications such as similar music search software, software having a code tracking function, and music recording software. Automatic analysis of music can lead to quantitative evaluation of music and transcription. As main information obtained by automatic analysis, there are melody analysis, chorus section estimation, etc. Among them, chord progression is what makes the atmosphere of the music significant. By analyzing chord progression appropriately, it is possible to quantitatively evaluate ambiguous things such as the atmosphere of music.

JP 2009-015535 A JP 2009-186944 A JP 2008-096844 A JP 2007-248610 A

  However, high accuracy cannot be expected for complex signals only by using pitch analysis by frequency analysis to obtain chord progression. In addition, in order to evaluate the chord characteristics, it is necessary to determine the key of the music.

  The present invention provides an apparatus for estimating the key of music and an apparatus for estimating chord progression.

  The music key estimation apparatus according to the present invention includes an audio data acquisition unit that acquires audio data of a music, a storage unit that stores key information indicating a correspondence between a key and a combination of scales, and analyzes audio data, A key estimation unit that identifies the key of the key. The key estimation unit frequency-analyzes speech data of a predetermined section to obtain a chroma vector, selects a plurality of scales from the scales included in the chroma vector, and based on the combination of the selected plurality of scales and the key information. A candidate is determined, and the key of the music is specified from the determined key candidates.

The music chord progression estimation apparatus according to the present invention includes an audio data acquisition unit that acquires audio data of a music, and a chord progression estimation unit that analyzes the audio data and estimates the chord progression of the music.
The chord progression estimation unit frequency-analyzes the audio data to obtain a chroma vector, specifies a music key from the chroma vector, obtains a chord from each of the plurality of predetermined small sections from the chroma vector, and specifies the specified music key The chord obtained for each small section is converted based on the chord, and the chord progression of the music is specified based on a combination of a predetermined number of chords extracted from the chord group that has been converted.

  The first program according to the present invention is a program for causing the information processing apparatus to operate as an apparatus for specifying the key of music. The first program causes the control device of the information processing device to obtain a chroma vector by frequency analysis of audio data in a predetermined section, a function to select a plurality of scales from the scales included in the chroma vector, and a selection A program for executing a function for determining a key candidate based on a combination of a plurality of scales and the key information, and a function for specifying the key of the music from the determined key candidates.

  The second program according to the present invention is a program for causing the information processing apparatus to operate as an apparatus for specifying the chord progression of music. The second program has a function for obtaining a chroma vector by performing frequency analysis on audio data, a function for specifying the key of a song from the chroma vector, and a plurality of predetermined small sections from the chroma vector. A code for a music piece based on a combination of a function for obtaining a chord, a function for converting a chord obtained for each subsection based on the key of the specified music piece, and a predetermined number of chords extracted from the converted chord group It is a program that executes a function for identifying progress.

  According to the present invention, by analyzing music data, it is possible to automatically extract a key in a predetermined section of the music, and further to automatically extract a chord progression pattern for the music.

The figure which shows the structure of the music analysis apparatus which is one Embodiment of the music chord progression estimation apparatus and music key estimation apparatus which concern on this invention. The flowchart which shows the process in a music analysis device Diagram for explaining calculation of chroma vector in analysis interval The flowchart which shows the key estimation process in a music analyzer A diagram explaining the correspondence between musical scales and labels that identify musical scales Diagram explaining the selection of five low values (scales with low chroma vector values) The figure which shows the structural example of the Key information which shows a response | compatibility with each Key and the combination of five low values. Diagram explaining scale circulation model The figure which showed each constituent sound about Key of "C" and "G" A flowchart showing chord progression estimation processing in the music analysis device The figure which showed the shift amount at the time of converting to Key "C" The figure explaining extraction of the chord progression pattern candidate (four chord strings) in the analysis section Diagram showing Markov model with code identification Diagram explaining the sequential extraction of six codes from codes in the analysis section

  Hereinafter, embodiments of a music chord progression estimation apparatus and a music key estimation apparatus according to the present invention will be described with reference to the drawings as appropriate.

(Embodiment 1)
1. Configuration of Music Analysis Device FIG. 1 is a diagram showing a configuration of a music analysis device that is an embodiment of a music chord progression estimation device and a music key estimation device according to the present invention. The music analysis device is a device that estimates the key (key) and chord progression of music (that is, automatically specifies). The music analysis apparatus estimates a chroma vector for an arbitrary section of the music, specifies a key based on the estimated chroma vector, and further specifies a chord progression.

  FIG. 1 shows the configuration of the music analysis apparatus 10. The music analysis apparatus 10 is configured by an information processing apparatus such as a personal computer. The music analysis apparatus 10 includes a controller 11 that controls the overall operation, a display unit 13 that performs screen display, an operation unit 15 that is operated by a user, and a storage unit 17 that stores data and programs.

  The display unit 13 is configured by, for example, a liquid crystal display or an organic EL display. The operation unit 15 is a device for a user to give an instruction, and includes a keyboard, a mouse, a touch panel, and the like.

  The storage unit 17 is a recording medium that stores parameters, data, and programs necessary for realizing the functions, and stores a control program executed by the controller 11 and various data. The storage unit 17 includes, for example, a hard disk (HDD), a semiconductor storage device (SSD), and a flash memory.

  The controller 11 includes a CPU and an MPU, and implements a predetermined function by executing a predetermined control program 17a stored in the storage unit 17. That is, the controller 11 functions as a key estimation unit and a chord progression estimation unit by executing the control program 17a. The control program 17a executed by the controller 11 may be provided via a network or may be provided by a recording medium such as a CD-ROM. The function of the controller 11 may be realized by cooperation of hardware and software, or may be realized only by a hardware circuit. That is, the controller 11 can be composed of not only a CPU and an MPU but also a DSP, an FPGA, an ASIC, and the like.

  The music analysis apparatus 10 includes a communication interface 19 for connecting to an external device such as a printer. The communication interface 19 is an interface circuit that communicates data and the like with an external device in accordance with USB, HDMI (registered trademark), IEEE1394, or the like. The music analysis apparatus 10 may further include an interface circuit that performs communication in accordance with standards such as IEEE 802.11 and WiFi for connection to a network. The music analysis apparatus 10 further includes an audio input interface 21 that inputs an audio signal from the microphone 20 that converts audio into an audio signal. An audio signal input via the audio input interface 21 is converted into audio data by an AD converter (not shown) and input to the controller 11.

2. Operation of Music Analysis Device The operation of the music analysis device 10 having the above configuration will be described. FIG. 2 is a flowchart showing processing related to the estimation of the key and chord progression of the music analysis apparatus 10. The process of the music analysis device 10 will be described with reference to FIG. The processing shown in FIG. 2 is executed by the controller 11 according to the control program 17a.

  In FIG. 2, the controller 11 of the music analysis apparatus 10 first extracts audio data of the analysis section using a partial section of the music as the analysis section (S11). In many cases, music is formed by stereo signals of two left and right channels. In this embodiment, in order to use both waveforms of the left and right channels as useful information, a combined signal of the signals of the left and right channels is used in the analysis. The analysis section may be set to an arbitrary section in the music. For example, the analysis section is set to a chorus section where the characteristics of the music appear more. The analysis interval includes a plurality of sampling intervals as shown in FIG.

  The controller 11 frequency-analyzes the voice data in the analysis section to obtain a chroma vector, and estimates (ie specifies) the music key based on the chroma vector (S12). In this process, the audio signal of the music is subjected to frequency analysis, a chroma vector in the analysis section is obtained, and the key is estimated from the chroma vector. Further, the controller 11 estimates (specifies) the chord progression of the music based on the estimated key (S13).

  Hereinafter, the key estimation process (S12) and the chord progression estimation process (S13) will be described more specifically.

2-1. Key Estimation FIG. 4 is a flowchart showing details of the key estimation process (S12) in the flowchart shown in FIG. First, the controller 11 calculates a chroma vector in the analysis section (S21). More specifically, first, the chroma vector CHi (i = 1, 2,...) Is calculated by performing frequency analysis for each sampling interval in the analysis interval. For example, continuous wavelet transform is used as a frequency analysis method. Continuous wavelet transform is a technique called mother wavelet that expresses all waveforms by scaling and translation of the fundamental wave, and can hold time-axis information of the waveform of the analysis source. The chroma vector has as its component the signal intensity of each scale (sound) (C, C #, D,...). The chroma vector is generated by summing up each frequency (every scale) within each sampling interval.

  After obtaining a chroma vector in each sampling interval, a chroma vector obtained in the entire analysis interval is obtained. That is, the chroma vector for the entire analysis interval is obtained by adding the chroma vector values for each same scale (sound) (each chroma vector component). FIG. 5 shows an example of the chroma vector obtained in this way. In the present embodiment, a label indicating a scale (sound) is attached to each scale (sound). Numbers in parentheses in FIG. 5 indicate labels. For example, the scale “C” is represented as “0” and the scale “F” is represented as “5”.

  Next, the controller 11 refers to the chroma vector and selects five sounds (hereinafter referred to as “low value sounds”) from among the 12 sounds “C” to “B” having the weak signal intensity (S22). ). For example, in the example of the chroma vector shown in FIG. 5, five sounds of C #, D #, F #, G #, and A # are selected as low-value sounds as shown in FIG.

  When a bass value is selected, a key candidate is set from a combination of bass values (S23). FIG. 7 is a diagram showing the key information 17b indicating the correspondence between each key and a combination of five low-value sounds constituting the key. In the key information 17b, a sound is expressed using a label. In the key information 17b in FIG. 7, for example, Key “C” is associated with a combination of sounds with labels 1, 3, 6, 8, and 10. In specifying the key, it seems that it is common to use a sound with a high signal strength, but in this embodiment, the key is specified using the sound with the lower signal strength. The reason for using the low signal strength sound without using the high signal strength sound is that the key identification accuracy has been improved by the inventors' experiments. The reason why five sounds are used from the lowest signal intensity is particularly assumed in the present embodiment is popular music, and popular music often has a configuration using seven sounds (diatonic scale). Because. The controller 11 refers to the key information 17b and sets a key candidate from the selected combination of bass values.

  Key candidates are set as follows. The controller 11 selects the key of the combination that most closely matches the combination of the five low-value sounds among the combinations defined by the key information 17b as the key candidate. For example, when the combination of the five selected low-value sounds is (1, 3, 6, 8, 10), the key “C” is uniquely identified with reference to the key information 17b in FIG.

  On the other hand, when the key cannot be uniquely determined from the combination of the five low-value sounds, the key is determined by changing the conditions. Specifically, a combination pattern of four sounds is constructed from the obtained five low-value sounds, and it is determined which low-value sound combination defined in the key information 17b is close to the pattern. For example, if the key is originally a song with “C” (= (1, 3, 6, 8, 10)) and five low-value sounds are detected as (1, 3, 5, 6, 8), In the key information 17b, a combination corresponding to a combination of four sounds is calculated as a key candidate. In FIG. 7, three key candidates of “C” and “F” matching (1, 3, 6, 8) and “G” matching (1, 3, 5, 8) are obtained. The number of key candidates calculated by determining a key candidate by a combination of four sounds is 3 at the maximum and 1 at the minimum.

  In some cases, five sounds cannot be selected as the low-value sound. For example, when there are a plurality of sounds having the same chroma vector value and five are selected from those having a weak signal intensity, more than five sounds may be selected. In such a case, the sound is selected so that the number of selected sounds is 5 or less. For example, in the chroma vector, when the sound level is selected from the one with the weaker signal intensity, “0.09”, “0.117”, “0.147”, “0.191”, “0.23” , “0.23” corresponds to six sounds. In this case, four of the low signal strengths are selected as bass values. Similarly, only two or three sounds may be selected as the bass value. In this case, a key candidate is set by referring to the key information 17b from a combination of four or two sounds selected as the bass value. That is, a key including 4 or 2 bass values as constituent sounds is set as a key candidate. If there is only one bass value, it is processed as an error.

  When the key candidate is set as described above, the controller 11 determines the key (hereinafter referred to as “estimated key”) of the music from the key candidate (S24). Specifically, the estimated key is determined from the key candidates as follows.

  The method for determining the estimated key differs depending on the number of key candidates. Each case will be described below.

(1) When the number of key candidates = 1, the key candidate is determined as an estimated key.

(2) When the number of key candidates = 3 When the present inventors are required to have three key candidates, when the correct key is N, the required candidates are N, N-7, and N + 7. I found Therefore, when there is a common key candidate whose distance from the two key candidates is 7, the common key is determined as the estimated key. If there is no such common key, it is processed as an error. FIG. 8 is a diagram for explaining a scale circulation model. Scale addition / subtraction is performed according to this model. For example, D-2 returns from “D” by “2” counterclockwise to “C”. D + 2 advances from “D” by 2 clockwise and becomes “E”. Thus, for example, when three key candidates are obtained and they are “C”, “F”, and “G”, since F + 7 = C and G−7 = C, the estimated key is “C”. Become.

(3) When the number of key candidates = 2 The processing differs depending on whether or not the distance between two key candidates is two. The processing in each case will be described below.

a) When the distance between two key candidates is 2 A key whose distance is 7 from the two key candidates is set as an estimated key.
b) When the distance between two key candidates is not two: Three types of voting are performed on the two key candidates according to a predetermined determination condition, and the key candidate having the larger vote value is determined as the estimated key. Hereinafter, three types of voting will be described.

<Voting 1>
The two key candidates are compared in strength for the same scale, and the key candidate having the higher scale is voted.

  For example, “C” and “G” are obtained as key candidates, the value of the chroma vector of “C” obtained in the analysis interval is 0.5, and the value of the chroma vector of “G” is 0.3. If it is, vote for “C”, which is stronger.

<Voting 2>
Among the scales constituting each candidate, the signal intensities of the different scales between the two key candidates are compared, and as a result of the comparison, the key candidate having the lower scale of the signal intensity is voted.

  There is a case where only one of the five low-value sounds has a different scale between the two key candidates for the scale constituting the key candidate. For example, when “C” and “G” are obtained as key candidates, the combination of bass values constituting “C” is (1, 3, 6, 8, 10) and constitutes “G”. The combination of bass values is (1, 3, 5, 8, 10). Here, the only scale that differs by one is F (5) and F # (6). Therefore, the signal intensities of F (5) and F # (6) are compared, and a vote is given to the key candidate including the constituent sound with the lower signal intensity. For example, in the chroma vector obtained in the analysis section, when the value of “F” is 0.6 and the value of “F #” is 0.3, “F #” is smaller, so Vote for “C” containing “F #” (6).

<Voting 3>
The intensities of the sounds (scales) immediately before the sound indicating each Key candidate (ie, the sound one step lower) are compared, and the higher Key candidate is voted.

  For example, it is assumed that “C” and “G” are obtained as two key candidates. FIG. 9 is a diagram showing the constituent sounds of the “C” and “G” keys. In the figure, the sound with “◯” is a constituent sound of each key. Sounds not marked with “◯” indicate the low-value sound described above.

  The key component includes a sound obtained by subtracting 1 from the key sound. For example, when the key is “C”, the sound obtained by subtracting 1 from “C” is “B”, and “B” is included in the constituent sounds of the key “C” as shown in FIG. The signal intensity of the sound obtained by subtracting 1 from the sound that is the key is not a low value because it constitutes the key. Therefore, when the correct key is “C”, the signal intensity of “B”, which is a sound obtained by subtracting 1 from “C”, is not low. On the other hand, when “G” is also listed as a key candidate, the sound obtained by subtracting 1 from “G” is F #, but this is not included in the constituent sound of “C”. That is, for two key candidates, the signal strength of a sound obtained by subtracting 1 from each key candidate is compared, and it is considered that the higher the signal strength, the higher the possibility of being a key candidate. Therefore, in the music analysis apparatus 10, when the key candidates are N1 and N2, the strengths of the sounds (signal strength) of N1-1 and N2-1 are compared, and a vote is given to the key candidate having the higher signal strength. Put in. For example, in each chroma vector, when the signal strength of “B” obtained by subtracting 1 from “C” is 0.4 and the signal strength of F # obtained by subtracting 1 from “G” is 0.1, Since “B” has a higher signal strength, vote for “C”.

  As described above, three types of voting are performed on the two key candidates, and the key candidate having the larger total vote value is determined as the estimated key.

  As described above, the key of the music can be determined from the chroma vector of the predetermined section of the music.

2. Next, the chord progression estimation process (S13) will be described. FIG. 10 is a flowchart showing details of chord progression estimation processing (S13) in the flowchart shown in FIG. The chord progression estimation process will be described with reference to the flowchart of FIG.

  The controller 11 obtains a code for each predetermined small section in the analysis section (S31). Here, the small section is set to a section of one beat (one quarter note). The music analysis device 10 performs BPM (Beat Per Minute) tracking during frequency analysis in a music analysis section. As a result, the number of samples in a small section, that is, a section of one beat (one quarter note) can be calculated. The controller 11 obtains a code for each sample section from the chroma vector, and obtains a code for one beat (for a small section). That is, a chord candidate is calculated from a chroma vector for each sampling, and a chord candidate having a frequency equal to or higher than a threshold within a one-beat section (small section) is determined as the code of the small section.

  Next, the controller 11 shifts the scale (pitch) of the chord of each small section in the analysis section so that the key becomes “C” (S32). Specifically, the code of each small section is shifted by the difference between Key and “C” obtained in advance according to the flowchart of FIG. FIG. 11 shows the shift amount when Key is converted to “C”. Referring to FIG. 11, for example, when the key obtained in advance is “F”, the shift amount is −5. Therefore, when the key obtained in advance is “F”, the code “E” in the small section is shifted by −5 and converted to “B”. Similarly, the code “G” in the small section is shifted by −5 and converted to “D”.

  Next, the controller 11 extracts a combination of 6 codes from the head of the code group in the analysis section shifted as described above (S33). For example, as shown in FIG. 12, when the code in the analysis section after the shift is “FGEFAGEAFGEA...”, First, six codes “FGEFAG” are extracted from the head.

  Next, the controller 11 extracts all possible combinations of the four chords from the extracted six chord combinations without changing the appearance order, and determines chord progression pattern candidates (S34). For example, in the example shown in FIG. 12, “FGEF”, “FGEA”, “FGEG”, and the like are extracted from “FGEFAG” as chord progression pattern candidates.

  Next, the controller 11 calculates the occurrence probability of each extracted chord progression pattern candidate based on the Markov model (S35).

  FIG. 13 is a diagram showing a Markov model used for calculating the probability of occurrence of a chord progression pattern. In general, by acquiring the key, it is possible to acquire which characteristic a code at a certain moment has. The characteristic of chords is something like "conversion" in a story, which is widely known as music theory. There are three types of code characteristics handled in the present invention: “Tonic (T)”, “Subdominant (SD)”, and “Dominant (D)”. The chord progression generally used is configured so that it can be heard comfortably in consideration of this chord characteristic. The inventor of the present application thought that this chord characteristic may be used as an evaluation to assist in estimating an appropriate chord progression, and has come to use a Markov model as shown in FIG.

  For example, when the chord progression pattern candidate is “FGEF”, using a Markov model, the probability of transition from F to G (0.5), the probability of transition from G to E (0.5), and E The probability of transition from F to F (0.3) is obtained, and 0.075 (= 0.5 × 0.5 × 0.3) is obtained as the occurrence probability of the chord progression pattern candidate “FGEF” using them.

  In the calculation of the probability of occurrence, a weight may be assigned when the chord progression first is a dominant chord and when the chord progression last is a tonic chord. This is because a general chord progression starts with a chord other than the dominant chord and ends with a tonic chord. For example, when the first is the dominant code, the obtained occurrence probability is decreased, and when the last is the tonic code, the respective weights may be set so as to double the obtained occurrence probability.

  Next, the controller 11 calculates the appearance frequency of each chord progression pattern candidate (S36).

  When the above processing (S34-S36) is completed for one set of codes (six codes) extracted from the code group included in the entire analysis section, the controller 11 determines that all the codes included in the analysis section , It is determined whether all combinations of 6 codes have been extracted (S37).

  If all the combinations of 6 chords have not been extracted, the controller 11 sets a plurality of chord progression pattern candidates from the set 6 chords while setting new 6 chords by shifting the extraction start position one by one (S40). Then, the occurrence probability and the appearance frequency are obtained for each chord progression pattern (S35 to S36). For example, as shown in FIG. 14, the 6 codes to be set can be composed of 4 codes for each 6-code combination while sequentially shifting to “FGEFAG”, “GEFAGE”, “EFAGEA”,... The chord progression pattern is obtained, and the occurrence probability and appearance frequency are obtained for each chord progression pattern.

When a plurality of chord progression pattern candidates consisting of four chords are obtained as described above, an evaluation value is calculated from the occurrence probability and the appearance frequency for each chord progression pattern candidate (S38). The evaluation value of each chord progression pattern candidate is calculated by the following equation.
Evaluation value = chord progression pattern candidate occurrence probability × 10 ^α
α = Appearance frequency of chord progression pattern candidate−1

  The controller 11 sets the chord progression pattern candidate showing the highest evaluation value among all chord progression pattern candidates as the estimated chord progression pattern (S39).

  As described above, the music analysis apparatus 10 can specify the chord progression of the music.

3. Summary As described above, the music analysis device 10 of the present embodiment operates as a chord progression estimation device. The music analysis apparatus 10 as a chord progression estimation apparatus includes a voice input interface 21 (an example of a data acquisition unit) that obtains voice data of a music, and a controller 11 (code) that analyzes the voice data and estimates the chord progression of the music. An example of a progress estimation unit). The controller 11 frequency-analyzes the audio data to obtain a chroma vector, specifies the key of the song from the chroma vector, obtains a code in each of a plurality of predetermined small sections from the chroma vector (S31), and Based on the key, the chord obtained for each subsection is converted (S32), and the chord progression of the music is specified based on the combination of a predetermined number of chords extracted from the converted chord group (S39).

  In addition, the music analysis device 10 according to the present embodiment also operates as a key estimation device. The music analysis device 10 as the key estimation device analyzes the voice data, the voice input interface 21 that acquires the voice data of the music, the storage unit 17 that stores the key information 17b indicating the correspondence between the key and the scale combination, and the voice data. And a controller 11 (an example of a key estimation unit) that identifies the key of the music. The controller 11 frequency-analyzes speech data in an analysis section (an example of a predetermined section) to obtain a chroma vector (S21), selects a plurality of scales from the scales included in the chroma vector (S22), and selects the selected plurality of scales. Key candidates are determined based on the scale combinations and the key information 17b (S23), and the key of the music is specified from the determined key candidates (S24).

  According to the music analysis apparatus 10 having the above configuration, the key and chord progression of a music can be automatically specified by analyzing the audio data of the music. Such a technique of the music analysis apparatus 10 can be applied to a music providing system that uses a chord progression as an index for calculating the degree of similarity of music and selects music having similar atmospheres as similar music. Furthermore, the present invention can be applied to an automatic musical score generation system for music.

  In the above example, the controller 11 of the music analysis apparatus 10 acquires the audio data (audio signal) to be analyzed via the audio input interface 21, but the acquisition destination of the audio data is not limited to the audio input interface 21. . When the sound data of the music of the analysis data is stored in the storage unit 17, the controller 11 may read out and acquire the sound data from the storage unit 17. Or you may acquire the audio | voice data of the music of analysis data from a network via the communication interface 19. FIG. That is, the data acquisition unit of the present invention can be configured by the voice input interface 21, the communication interface 19, the storage unit 17, and the like.

  As described above, the first embodiment has been described as an embodiment of the present invention. However, the technology in the present invention is not limited to this, and various modifications, replacements, and additions may be made within the scope of the claims or equivalents thereof. Can be omitted.

DESCRIPTION OF SYMBOLS 10 Music analyzer 11 Controller 13 Display part 15 Operation part 17 Storage part 17a Control program 17b Key information 19 Communication interface 20 Microphone 21 Voice input interface

Claims (11)

An audio data acquisition unit for acquiring audio data of the music;
A storage unit for storing key information indicating a correspondence between a key and a combination of scales;
A key estimation unit that analyzes voice data and identifies the key of the music,
The key estimation unit
Obtain the chroma vector by frequency analysis of the audio data of the predetermined section,
Selecting a plurality of scales from the scales included in the chroma vector;
A key candidate is determined based on the selected combination of scales and the key information,
The key of the music is specified from the determined key candidates.
Music key estimation device. The Key estimation unit selects a predetermined number of scales from the lower signal strength in the scales included in the chroma vector.
The music key estimation apparatus according to claim 1. The key estimation unit
If there is one key candidate, that key candidate is identified as the key of the song,
When there are two key candidates, one key is selected from the key candidates based on a predetermined determination condition, the selected key is specified as the key of the music,
2. The music key estimation apparatus according to claim 1, wherein when there are three key candidates, one key is selected from the key candidates using a scale circulation model, and the selected key is specified as the key of the music. The predetermined determination condition is:
A first condition that the intensity of the same scale between two key candidates is higher;
A second condition in which the intensities of the different scales between the two key candidates are lower;
A third condition in which the intensity of the scale one level lower than the scale indicated by each Key candidate is higher,
The music key estimation apparatus according to claim 3. An audio data acquisition unit for acquiring audio data of the music;
A chord progression estimation unit that analyzes voice data and estimates the chord progression of the music,
The chord progression estimation unit
Frequency analysis of the audio data to obtain a chroma vector;
The key of the music is specified from the chroma vector,
From the chroma vector, a chord is obtained in each of a plurality of predetermined subsections, the chord obtained for each subsection is converted based on the key of the specified music, and a predetermined number of codes extracted from the converted chord group A music chord progression estimation device that identifies the chord progression of the musical piece based on a combination of chords. The chord progression estimation unit converts the chord obtained for each small section based on the difference between the identified key and “C”.
The music chord progression estimation apparatus according to claim 5. The chord progression estimation unit
Using a predetermined number of chord combinations as chord progression patterns, extracting a plurality of chord progression patterns from a chord group obtained in a plurality of small sections,
One chord progression pattern is identified based on the occurrence probability and appearance frequency of each chord progression pattern, and the chord progression of the music is identified based on the identified chord progression pattern.
The music chord progression estimation apparatus according to claim 5. The chord progression estimation unit calculates the occurrence probability of the chord progression pattern using a Markov model indicating a transition probability between three types of sound characteristics of tonic, dominant, and subdominant.
The music chord progression estimation apparatus according to claim 7. The chord progression estimation unit
Obtain the chroma vector by frequency analysis of the audio data of the predetermined section,
Selecting a plurality of scales from the scales included in the chroma vector;
A key candidate is determined based on the key information indicating the correspondence between the selected combination of scales and the combination of key and scale,
The key of the music is specified from the determined key candidates.
The music chord progression estimation apparatus according to any one of claims 5 to 8. A program for operating an information processing device as a device for specifying the key of music,
In the control device of the information processing device,
A function for obtaining a chroma vector by performing frequency analysis on audio data in a predetermined section;
A function of selecting a plurality of scales from the scales included in the chroma vector;
A function for determining a key candidate based on a combination of a plurality of selected scales and the key information;
A function for specifying the key of the music from the determined key candidates;
program. A program for operating the information processing device as a device for specifying the chord progression of music,
In the control device of the information processing device,
A function that obtains chroma vectors by frequency analysis of audio data;
A function for specifying the key of a song from the chroma vector, and a function for obtaining a code in each of a plurality of predetermined small sections from the chroma vector;
A function of converting the code obtained for each of the small sections based on the key of the specified music;
A function of specifying the chord progression of the music based on a combination of a predetermined number of chords extracted from the converted chord group,
program.

Images