JP5196550B2 - Code detection apparatus and code detection program - Google Patents

Description

  The present invention relates to a chord detection device and a chord detection program, and in particular, chord detection suitable for accurately detecting chords (chords) from a music acoustic signal in which a plurality of musical sounds are mixed (hereinafter simply referred to as “acoustic signal”). The present invention relates to an apparatus and a code detection program.

  Chord is a very important element in popular music, and when playing music of such a genre in a small band, do not use the score with the individual notes to play, but chord or lead It is common to use a musical score in which only a melody called a sheet and chord progression is written. Therefore, it is necessary to record the chord progression of a song in order to play a song such as a commercially available music CD in a band, but this work can only be performed by an expert with special musical knowledge. It was impossible. Therefore, various code detection devices and code detection programs for detecting codes by performing arithmetic processing on a sound signal mixed with a plurality of musical sounds using a fast Fourier transform method (FFT) using a commercially available personal computer have been studied. Yes.

  For example, if one measure is composed of multiple chords, the measure is divided into the first half and the second half, and the bass sound is detected in each of them. A code detection device is known in which a code is detected in each case (Patent Document 1).

  However, in the chord detection apparatus described in Patent Document 1, when a plurality of different chords contain the same bass sound, only one chord is detected in one whole measure, and the detection accuracy is insufficient. In addition, since a strong sound in the whole measure is used as a bass sound, there is a possibility that the bass sound cannot be correctly detected in bass running or the like in which sounds are connected with quarter notes like jazz.

In view of this, a chord detection apparatus has been proposed that can detect a correct chord even when different chords containing the same bass sound are present in a measure (Patent Document 2). In this chord detection apparatus, measures are divided not only according to the bass sound but also according to the degree of change in chords. That is, when the bass sound is different or the change degree of the chord is large, the chord is detected by dividing the bar.
JP 2007-52394 A JP 2008-40283 A

  In the apparatus described in Patent Literature 2, first, a chord detection section that is considered to contain a single chord is determined using the degree of change between the base sound and the chord constituent sound candidates, and within this chord detection section, The chord (chord name) is determined using the bass sound and the chord component sound. However, in the case of performing chord detection using only the base sound and chord constituent sound as information in the chord detection section as described above, the accuracy of the chord detection result is still insufficient.

  For example, there is a genre of music that is preferred for each user. If chord detection can be performed in consideration of chord progression and chord theory that are generally used in the genre, a highly accurate chord detection result can be expected.

  The present invention has been made for the above-mentioned problem, and does not depend only on the bass sound or the chord constituent sound, but also detects the correct chord in consideration of the genre of the song and the tendency of the song preferred by the user. It is an object of the present invention to provide a code detection device and a code detection program that can be configured to be configured.

In order to solve the above problems and achieve the object, the present invention comprises means for detecting the power of each musical tone of a musical sound signal, means for setting a plurality of chord detection sections in the acoustic signal, and the power From the above, means for detecting the power of each scale sound in the predetermined bass sound detection range in the portion corresponding to the first beat of each chord detection section, and the sound having the maximum power among the scale sounds as the base sound Means for determining, extracting a predetermined number of pitch names in descending order of power from the scale sounds of the entire chord detection section, detecting chord candidates each having the predetermined number of pitch names as root sounds, and all the chord candidates A likelihood is calculated from the average power level of the chord constituent sound, and a chord candidate having the maximum likelihood is determined as a chord, and a genre of a genre of a song that is a chord detection target, In determining the serial bass, with power of the fundamental tone and the harmonics of each chromatic note is based sound up sound, in that so as to weight differently to the power of the fundamental and harmonics in response to said genre There is a first feature.

  The present invention also provides means for detecting the power of each musical tone of a musical sound signal, means for setting a plurality of chord detection sections in the acoustic signal, and the first beat of each chord detection section from the power. Means for detecting the power of each scale sound in the planned bass sound detection range in the portion corresponding to the above, means for determining the sound having the highest power among the scale sounds as the base sound, and the scale of the entire chord detection section Means for extracting a predetermined number of pitch names from the sound in descending order of power, detecting chord candidates each having the predetermined number of pitch names as a root sound, and an average power level of all chord constituent sounds for the chord candidates Then, a likelihood is calculated, and a code candidate having the maximum likelihood is determined as a code and a tension level input means. The means for detecting the code candidate has a tension level. Configuration sound number of codes Flip the invention is secondly characterized in the point that is configured to narrow down the code candidate.

  According to a third aspect of the present invention, the means for calculating the likelihood further includes means for correcting the power corresponding to the tension level based on the pitch from the root tone for the detected constituent sound of the code candidate. There are features.

  In the present invention having the first feature, when detecting the bass sound based on the power of the fundamental tone and the harmonic overtone, the power of the fundamental tone and the harmonic overtone can be weighted according to the genre. Therefore, for example, the bass drum sound power can be detected low by reducing the weighting of the fundamental tone for a rock-type song with a high bass drum volume without a harmonic structure. It is possible to avoid erroneous detection of a drum sound as a bass sound.

  In the present invention having the second and third features, chord candidates can be narrowed down by limiting the number of chord constituent sounds according to the tension level. In addition, since the power can be corrected by manipulating the likelihood according to the tension level, a code corresponding to each tension level can be easily detected. Since the tension level tends to vary depending on the genre of the music, specifying the tension level is convenient for detecting a chord from a music of a specific genre or a user's favorite music.

  Hereinafter, the present invention will be described in detail with reference to the drawings. FIG. 2 is a block diagram showing a hardware configuration of a personal computer as a code detection apparatus according to an embodiment of the present invention. The hardware configuration of the personal computer is well known, and when the code detection program according to the present invention is loaded, it operates according to the function as the code detection device.

  In FIG. 2, the personal computer 1 has a CPU 2, a ROM 3, a RAM 4, a display device (for example, a liquid crystal display) 5, and an external storage device (hard disk device) 6, which are connected via a system bus 7. Further, a keyboard 9, a sound system 10, and a CD-ROM drive 11 are connected to the system bus 7 via an input / output interface 8. Each part of the personal computer 1 inputs and outputs signals and data with each other through the system bus 7. Note that the input device for giving an instruction to the personal computer 1 by the user is not limited to the keyboard 9 but also includes a pointing device such as a mouse.

  The ROM 3 is a storage area in which the BIOS (Bios) of the personal computer 1 is stored. The RAM 4 is used not only as a storage area for programs read from the hard disk 7 but also as a temporary storage area for work areas, coefficients, parameters, and the like.

  The display device 5 is controlled by a display control unit (not shown) that performs necessary image processing in accordance with an instruction from the CPU 2 and displays an image processing result.

  The CD-ROM drive 11 is a reading unit that reads data from a CD-ROM and stores it in the hard disk device 6, and is a drive device that reads into the personal computer 1 a code detection program and an acoustic signal to be code-detected. Programs and data (including acoustic signals) read by the CD-ROM drive 11 are stored in the hard disk device 6, and main programs are stored in the RAM 4.

  FIG. 3 is a flowchart showing main processing by the code detection program. In FIG. 3, in step S1, an acoustic signal of a chord detection target music is read. The acoustic signal can be read by ripping from a CD-ROM set in the CD-ROM drive 11 or designating a musical sound waveform pre-stored in the WAV file. In step S2, a code detection range in the read acoustic signal is determined. The entire read sound signal (that is, the entire music piece) may be used as the chord detection range, or for example, the user may designate an arbitrary measure in the intro part or the chorus part of the music piece using the input means. In order for the user to specify the code detection range, the waveform of the acoustic signal is displayed on the display device 5 so that the user can specify the range using the keyboard 9 or a pointing device.

  In step S3, the time signature is set. The time signature is detected based on an instruction input by the user. For example, the time signature is set by, for example, reproducing the music based on the sound signal read in step S1, listening to the reproduced sound, the user determining the time signature, and displaying the determined time signature on the display device 5, for example. Enter from the time signature input screen.

  In step S4, the position of the beat is detected. The beat is detected based on tapping (hit a schedule key on the keyboard 9) input by the user. That is, the music is reproduced according to the acoustic signal, and the user determines the position of the beat while listening to the reproduced music, and hits the keyboard 9 at that position to input the beat. The tapping start position is from the first beat of the measure. That is, the user determines the first beat according to the reproduced music, and starts tapping from that timing. When the tapping interval is stabilized and falls within a predetermined fluctuation range, the beat number (beats per minute) is determined.

  In step S5, the bar line position is determined based on the time signature and the number of beats. Since the first tapping is the first beat, for example, in the case of 4/4 time, start with the tapping start and apply the numbers 1 to 4 in order for each tapping. A tapping position indicated by 4, that is, a measure including four beats is determined. The bar line is drawn at the first beat position among the four beats.

  Thus, when one bar is determined, bar lines are drawn for the acoustic signals before and after the bar size, and one bar is included in the chord detection range set in step S2. Is determined. The chord detection section is a section for obtaining the power spectrum of the scale sound by performing FFT calculation of the acoustic signal in order to detect the chord.

  Note that the method for determining the code detection section is not limited to the one that is determined based on the time signature and the number of beats input by the user. For example, in addition to inputting the beat interval by time signature or tapping, the sound signal is subjected to FFT calculation for each predetermined frame to obtain the degree of change in power of each scale sound for each frame, and the average beat interval is obtained from this change degree. It is also possible to obtain a tempo candidate and select a tempo candidate that is close to the tapping tempo as the beat position. As a technique for detecting a bar line of a song by FFT calculation of an acoustic signal, the one disclosed in the specification of Japanese Patent Application No. 2006-216362 (Japanese Patent Laid-Open No. 2008-40284) filed earlier by the present applicant is applied. can do. In short, in this embodiment, it is only necessary to specify a section in units of one measure as the code detection section. However, the present invention is not limited to one bar as a code detection section, and as described in Patent Documents 1 and 2, the bar may be further divided so that the width of one bar is set as a plurality of code detection sections. .

  As described above, if a code detection section is set in the code detection range, code detection is performed in order from the first code detection section in step S6.

  Next, the code detection process will be described. FIG. 4 is a flowchart showing the main part of the code detection process. In FIG. 4, in step S60, the bass sound of the chord detection section is detected from the sound signal of the chord detection section (for one bar). The base sound is obtained based on the power of the scale sound corresponding to the first beat of the chord detection section. The power of each scale sound is detected for each frame from the obtained power spectrum by performing an FFT operation at predetermined time intervals (frames) from the input acoustic signal. In step S61, code candidates in the code detection section are detected.

  In step S62, the likelihood of the code candidate detected in step S61 (appropriateness as a code candidate) is calculated according to Equation 1 described later. In step S63, the code name is determined with the code candidate having the highest calculated likelihood as the detection code.

  In the chord detection program of this embodiment, detection options can be set to improve the accuracy of bass sound detection and chord candidate detection. The detection of the bass sound and the chord candidate with the detection option added will be described later. First, the base sound detection procedure and the chord candidate detection procedure without the detection option will be described.

  The base sound is determined based on the power of the scale sound of the acoustic signal corresponding to the first beat of the chord detection section. Of the power spectrum corresponding to the frequency in the range of 50 cents above and below the fundamental frequency of each scale sound (100 cents is a semitone), the maximum value can be the power of the scale sound. Of the power spectrum calculated for all scale sounds (C1 to A6), the power of the scale sound in the bass sound detection range (for example, C2 to B3) is used to detect the bass sound.

  The power of the scale sound is an average of the i-th scale sound between frame numbers fs and fe, where Li (t) is the level of the i-th (lowest to i-th) scale sound at frame time t. The power Lavgi (fs, fe) can be calculated by Equation 1.

  The average power calculated by Equation 1 is calculated for all scale sounds i (0 ≦ i ≦ 69). The power calculated here is stored in a predetermined storage unit. Then, the scale sound having the maximum average power in the bass sound detection range is detected as the bass sound.

  If the bass sound of the scale tone in the portion corresponding to the first beat is detected in the chord detection section (one measure), the process proceeds to chord candidate detection. For detection of chord candidates, the power of the scale sound in the chord detection section (one measure) is used. The average power of each scale tone in a chord detection range (for example, C3 to A6) designated in advance is integrated for every 12 pitch names (C, C #, D, D #,..., B), Calculate the power for each note name. Then, a maximum of six pitch names having the largest power for each pitch name are extracted. Although the maximum number of extracted pitch names is six, only five may be detected. This is because, in order to prevent a code candidate from being erroneously detected in a chord detection section close to silence, a code candidate whose average power is equal to or less than a threshold is not set as a code candidate.

  If the detected bass sound is not included in the extracted maximum six sounds, the chord name is determined from the maximum seven sounds including the bass sound.

  In order to determine a chord candidate, first, any one of a maximum of seven detected sounds is assumed as a chord root tone, and the root tone and a predetermined chord type are located at a predetermined pitch from the root. For all notes, calculate the total power and divide by the number of chords. In this way, the chord candidate including the constituent sounds including the temporary root sound and the likelihood of the chord candidate are calculated.

  For example, when four sounds (do, mi, seo, and shi) are extracted as sounds with high power, first, it is assumed that “do” is a root sound, and root sounds for each preset chord type are assumed. The powers of the scale sounds in the pitch (interval) relation (see FIG. 5) of the constituent sounds based on the above are summed.

  FIG. 5 is an example of a table showing chord types and pitches of constituent sounds for each chord type (pitch based on the root tone). With reference to FIG. 5, the pitch name for detecting the power is determined. For example, in the major (Maj), when “do” is a root sound, a sound (mi) having a pitch of four semitones from the root sound and a pitch of seven semitones from the root sound (sound) ) And are selected. Then, the likelihood is calculated for these three sounds. The likelihood is a value obtained by dividing "the total average power of all constituent sounds" by "the number of chord constituent sounds".

  For example, regarding the major chord (Maj), when the root note is “do”, the average power of the note names “do”, “mi” and “so” of the major chord in the chord detection section is calculated. , Summed for each octave in the range of C3 to A6, divided by the number of octaves and averaged, obtained for each component sound, and the total was calculated as "the sum of the average power of all component sounds" And In the case of a major chord, the number of chord constituent sounds is “3”.

  Here, when the constituent sound calculated from the temporary root sound is not included in the extracted maximum seven sounds, the power of the sound may be “zero”.

  Thus, the likelihood is calculated for all code candidates. Calculations are performed for all code types shown in FIG. For example, if there are four extracted note names, there are four types of provisional root sounds, so the number obtained by multiplying 15 types of chords from Maj to 13 and the number of constituent sounds “4” ( The likelihood is calculated for the candidate code 60).

  If the likelihood is obtained for each code candidate, the code candidate having the maximum likelihood among the likelihoods is determined as a detection code. The code name is completed by combining the temporary route and the code type set for the confirmed code candidate. In general, a major chord is not added with a chord type, and only a root sound, for example, “C” is described.

  As described above, when the code name is determined for one code detection section, the code is similarly detected for the next code detection section (bar). If a code name is detected for the specified code detection range, the code name is pasted on the code track provided for each code detection section and displayed on the display device 5. You may display as a lead sheet which described the detection code name. Alternatively, the detection chord may be played using a built-in rhythm pattern that can be preset in the chord detection program.

  FIG. 1 is a functional block diagram showing the main part of the code detection program. In FIG. 1, an acoustic signal input unit 12 is means for inputting an acoustic signal from a music CD, a waveform file, or the like to the acoustic signal storage unit 13. The chord detection range setting unit 14 accepts designation of the chord detection range input by the user, and sets the range to the acoustic signal stored in the acoustic signal storage unit 13. The chord detection section setting unit 15 determines a bar line based on the time signature and beat input by the user, and sets a chord detection section. The method for setting the code detection section is not limited to this, as described above.

  The power calculation unit 16 calculates the power spectrum of the scale sound of the acoustic signal by FFT calculation. The calculated power spectrum is stored in the scale sound power storage unit 17. The base sound detection unit 18 detects a sound having the highest power among the powers of the acoustic signal for the portion corresponding to the first beat of the chord detection period as the base sound.

  The chord candidate detection unit 19 obtains the average power of each scale note in the chord detection section, extracts a predetermined number of pitch names having the largest power for each pitch name, and routes each of these extracted pitch names. It has a function of extracting a chord candidate assumed to be a sound according to the pitch of the pitch table 19a. As described with reference to FIG. 5, the example of the pitch table 19a indicates a predetermined number of chord types and pitches from the constituent sound routes.

  The likelihood calculating unit 20 has a function of dividing, for each chord candidate, the total average power of all constituent sounds including the root sound in the chord detection section by the number of chord constituent sounds. The code name determining unit 21 determines a code candidate having the maximum likelihood calculated by the likelihood calculating unit 20 as a detected code name, and outputs the code candidate to the code name display unit 21. The code name display unit 21 causes the display device 5 to display the code name pasted on the code track.

  Next, detection options that can improve code detection accuracy will be described. As a first detection option, a bass sound detection function using a bass overtone structure by specifying a genre of music can be added. In the above-described bass sound detection without options, the sound having the highest average level in the bass sound detection range is used as the base sound. However, this method may erroneously detect the sound of a bass drum or the like as a bass sound. In particular, there is a risk of erroneous detection in a rock-type song with a high bass drum volume. However, generally, the sound of a bass drum is different from a bass sound having a specific harmonic structure depending on the tone color in that it does not have a harmonic structure. Therefore, in this embodiment, as an option, a bass drum sound having no overtone structure is not erroneously detected as a bass sound.

  As described above, the bass sound is detected based on the power Lavgi (fs, fe) of the scale sound in the portion corresponding to the first beat of the chord detection section calculated by Equation 1. The power Lavgi (fs, fe) is corrected as follows. That is, the average power Lavgi (fs, fe) of the i-th scale tone is expressed as Lavg (i) with i as a variable, and not only the fundamental tone but also the second harmonic, the third harmonic, ..., the powers of the eighth harmonic overtones Lavg (i) to Lavg (i + 36) are added to obtain the average power of the i-th scale. In this case, a coefficient is multiplied for each fundamental tone and overtone.

  That is, the average power Lavg (i) of the i-th scale sound is calculated by the following formula 2.

  Lavg (i) = {(K1 · Lavg (i)) + (K2 · Lavg (i + 12)) + (K3 · Lavg (i + 19)) + (K4 · Lavg (i + 24)) + (K5 · Lavg (i + 28)) + (K6 · Lavg (i + 31)) + (K7 · Lavg (i + 34)) + (K8 · Lavg (i + 36))} ÷ (K1 + K2 + K3 + K4 + K5 + K6 + K7 + K8) ...... Equation 2

  The coefficients (genre coefficients) K1 to K8 are numerical values that can be preset according to the genre of the music. As can be understood from Equation 2, after multiplying each power of the scale tone from the fundamental tone to the eighth harmonic, which is the pitch of the i-th tone, by the genre coefficient, the sum is added to each other, and the sum is further obtained. The power is calculated by dividing by the total of the genre coefficients K1 to K8.

  For example, the coefficients K1 to K8 can be set to the following values. FIG. 6 is an example of genre coefficients K1 to K8. In this example, for the jazz, the fundamental tone coefficient K1 is made extremely large to emphasize the fundamental tone power. The coefficient K is reduced for harmonics other than the fundamental tone. On the other hand, for the rock, the fundamental tone coefficient K1 is extremely smaller than that of jazz, and the coefficients K2 to K8 of the second overtone are set to the same values as in the case of jazz.

  When the user designates jazz as the genre, the bass sound is detected with emphasis on the fundamental tone. This jazz designation should be made for songs where the bass drum is not played very strongly or drumless songs. In jazz music, wood bass (contrabass) is often used, and the sound of this wood bass is strongly attenuated and overtones are not detected strongly. Therefore, in this respect as well, setting a coefficient that emphasizes the fundamental tone is meaningful.

  The selection of rock as a genre is in the case of rock-type songs where the bass drum is played strongly. Since the power of harmonics is about the same as or about one third of the power of the fundamental tone, the sound that has only the fundamental tone, such as the bass drum, is detected with a relatively small power relative to the harmonics. , As a result is eliminated.

  In order to realize this detection option, in the chord detection program, a display for allowing the user to specify a genre is performed before the start of the bass sound detection in step S60. That is, a genre designation screen on which the user can input an instruction can be displayed on the display device 5. For example, it is preferable to display a genre (jazz, rock), have a switch function, and accept a genre designation by clicking on the display.

  The bass sound detection unit 18 is provided with a coefficient table describing the genre-specific coefficients K1 to K8. The bass sound detection unit 18 calculates the average power Lavgi (fs, fe) of the i-th scale tone from the frame fs to fe according to Equation 1, and then calculates the coefficients K1 to K8 from the coefficient table according to the specified genre. Reading and correction are performed according to Equation 2.

  Note that the values of the coefficients K1 to K8 shown in FIG. 6 are not limited to this, and how many overtones are added can be modified. Further, when the overtone exceeds the range of the scale tone obtained by the FFT calculation, the power is added as “0” in the calculation of Expression 2.

  The average power of each scale sound corrected in this way is determined as the base sound with the highest power in the bass sound detection range, but the calculated average power of each scale sound is below a predetermined threshold value. Is preferably excluded from bass sound candidates in advance. Therefore, for all sounds in the bass sound detection range, if the corrected average power is less than the threshold, no bass sound is detected.

  In the second detection option, the code candidate can be corrected by the tension level. As described above, when the code name is determined, the code candidate having the maximum likelihood is determined as the code name of the code detection section. In the second detection option, chord candidates are narrowed down by the tension level at the stage of detecting chord candidates using the pitch table of FIG. As a result, it is possible to narrow down code candidates according to the characteristics of the song, such as the song or genre of the song that the user likes. For example, a plurality of tension levels such as three chords, four chords, and five chords or more can be designated, and chord candidates for each tension level are detected. There are four types of tension levels, for example, 1 to 3 and N. FIG. 7 is a diagram illustrating an example of narrowing chords for each tension level and each tension level. At the tension level “0”, only the three chords are extracted or the three chords are extracted. At the tension level “1”, only four chords are extracted, or the four chords are extracted. At the tension level “2”, only five chords or more or five chords or more are extracted. At the tension level “N”, the detected code is output as it is as a code candidate without being narrowed down.

  The chord candidate detection unit 19 selects a chord type of a specified tension level from the pitch table of FIG. 5 and detects chord candidates having a sound corresponding to the pitch of the chord type as a constituent sound. In order to realize the second detection option, in the code detection program, before starting the code candidate detection in the step S61, a display for designating the tension level to the user is performed. That is, a display that allows the user to specify the tension level is performed on the display device 5. For example, the tension level (1 to 3, N) is displayed on the display device 5, the display unit is provided with a switch function, and the tension level can be received by clicking the display. The code candidate detection unit 19 is configured to output a code candidate corresponding to the tension level when the designation of the tension level is received.

  Further, the likelihood calculation can be corrected according to the designated tension level. That is, when calculating the likelihood using the table of FIG. 5, the average power of each component sound is multiplied by a coefficient corresponding to the pitch from the root sound when the average power of the component sounds is calculated. Then add up.

  FIG. 8 is a diagram showing the power correction coefficient for each tension level. In FIG. 8, when the tension level is “0”, the coefficient applied to the sound power of 7th (Seventh) to 13th (Sirence) is reduced. When the tension level is “1”, only the power of the seventh sound is multiplied by a coefficient larger than “1” to reduce the power of the sound from the nineth to the third. When the tension level is “2”, the power of all sounds from the seventh to the third is multiplied by a coefficient larger than “1”. When the tension level is “N”, the power of the sound from the seventh to the third is multiplied by a coefficient “1” so that the calculated power is not corrected.

  In this way, the tension power corresponding to the designated tension level is relatively increased, so that a desired code is easily generated and only the designated tension level is not detected.

It may determine the code name from power strong sound for each code detection section, but in order to increase more the detection accuracy can be further utilized chord progression. A chord progression is a concatenation of a plurality of chords and represents a chord flow. For example, when the key is C major, the chord progression of “Dm7 → G7 → C” is known as the most common chord progression. As described above, since chord progression includes frequently used chord progressions and unusual chord progressions, detection of chords according to frequently used chord progressions can be expected to improve detection accuracy.

In order to improve detection accuracy, when using chord progression, first, chord candidates are detected and the likelihood is calculated for all chord detection sections in the chord detection range. Compensation options and the likelihood of the above tension level, should go ahead of this process. When code candidates are detected for all chord detection sections, the chord names of the chord candidates in the previous chord detection section are used in order from the second chord detection section toward the end of the song. A correction is made to the likelihood of the code candidate in the detection section.

  Prepare a chord progression database for correction. If the chord progression in the chord progression database is compared with the chord progression in the chord progression database and the chord progression in the chord progression database is compared with the chord progression in the chord progression database, FIG. Increases the likelihood of the current code candidate calculated using If the chord progression is prohibited, the likelihood is corrected to be lowered.

  FIG. 9 is a diagram illustrating an example of chord progression and a coefficient by which likelihood is multiplied by chord progression. In FIG. 9, a large coefficient “1.3” is set for general chord progression, and a small coefficient “1.2, 1.1, etc.” is set for less common chord progression. For chord progressions that are not normally used, a coefficient of “0.9” smaller than “1.0” is set.

  In FIG. 9, the Roman numerals constituting the chord name are the pitches from the tonic of the key (key) of the portion where the chord is located in the song or song. Likelihood can be corrected only for the key by having the user input the key. However, in general, the tonality of a song often changes from part to part even within one song. Therefore, regardless of the key, the chord of the previous chord detection section and the chord root of the chord candidate of the target chord detection section It is only necessary to determine whether the chord progression matches based on the chord type only.

  In FIG. 9, the major chord (Maj) code (1, 4, etc.) written as a triad may include not only the major chord but also the M7 chord. The minor (m) code (2m, 3m, etc.) may include not only the minor triad (3 chords) but also the m7 code.

  The chord progression shown in FIG. 9 is for the major key (major key), but a chord progression database for the minor key (minor key) can be created in the same manner.

  FIG. 10 is a flowchart of a main part of the code detection process including selection of a detection option. In step S70, a process for causing the user to input a genre as a detection option is performed. For example, a character “Please input the genre of the song” is displayed on the display device 5 and a selection button for designating jazz or rock is also displayed.

  In step S71, processing for requesting the user to specify a tension level as a detection option is performed. For example, a character “Please input tension level” is displayed on the display device 5 and a selection button for designating any one of tension levels “1” to “3” and “N” is also displayed.

  In step S72, it is determined whether the genre of the song selected by the user is jazz or rock, and the genre coefficients K1 to K8 are read from FIG. In step S73, the average power of each tone of the first beat in the chord detection section is calculated using Equation 1. In step S74, the average power of each tone of the first beat in the chord detection section is corrected using the genre coefficients K1 to K8 read in step S73. In step S75, a sound having the maximum average power of each tone of the first beat in the corrected chord detection section is extracted as a base sound.

  In step S76, the tension level designated by the user is read. In step S77, chord candidates are extracted only for chords having the number of constituent sounds corresponding to the tension level. This step is omitted if the chord is not limited to a chord having a number corresponding to the tension level. In step S78, the power correction coefficient is read according to the tension level. In step S79, the likelihood is calculated for the code candidate by incorporating the power correction coefficient using FIG.

  In step S80, the code name is determined using the code candidate having the maximum likelihood as the detection code.

  Likelihood correction by chord progression is performed after extraction of chord candidates and calculation of likelihood for all chord detection intervals. That is, after the processing of steps S73 to S79 is executed for all the bars, it is determined for each bar whether or not the connection between the chord candidate detected in that bar and the chord candidate of the immediately preceding bar is a predetermined chord progression. Then, the likelihood is corrected depending on whether the chord progression is general or unordinary chord progression, and finally the chord name having the highest likelihood is selected from the corrected chord candidates as the detection code.

  In the above embodiment, the case where the code candidate having the maximum likelihood is used as the code name of the code detection section has been described. However, the present invention is not limited to this. For example, a plurality of code candidates are displayed on the display device 5 in descending order of likelihood. The user may be allowed to select. At this time, MIDI information based on the chord candidate may be transferred to the sound system 10 to generate a chord, and as a result of hearing it, one of a plurality of chord candidates may be selected as a detection code.

  In addition, the screen for displaying the chord detection result can be provided with a playback switch for simultaneously performing a performance by an acoustic signal (original waveform) of a song to be chord detected and a chord performance by MIDI of the detection chord. In response to the user operating the playback switch, a song and a chord based on the detection code are played, so that the user can confirm whether the detection code is correct by listening to both. If an incorrect detection code is played along with the song, it will sound uncomfortable, and the user can find an error in the detection code. At this time, in order to easily confirm the detected chord, it is preferable that the volume balance of the performance sound of the song and the chord performance sound based on the detected chord name can be individually adjusted.

  When an error in the detection code is found, the user can correct the code name. For example, the second candidate can be selected instead of the first candidate for the code. Further, for example, the code name pasted on the code track is displayed on the display screen 5, and the display portion can be specified and corrected by the user. It is preferable that code name candidates for correction are displayed on the screen so that the user can select them as correction codes. Data correction on the screen can be performed by employing a well-known method.

  Further, when the user corrects the detected code name, a chord progression formed by concatenating the corrected chord and the chord in the preceding chord detection section is a predetermined chord progression database (see FIG. 9). If not included, the chord progression including the chord after correction is registered in the chord progression database. Thereby, the chord progression can be easily detected thereafter. In other words, the chord progressions that are often used in the genre of the song that the user likes are accumulated as a database, so the effect of the detection option using chord progression becomes more prominent as the chord progression added by modification increases. Come.

  Furthermore, when the chord name is corrected by the user and the chord progression is newly registered, chord detection may be performed again in the chord detection section after the chord detection section. When the user makes a correction, the chord progression due to the connection between the chord in the chord detection section that has been corrected and the chord in the chord detection section immediately after the chord changes. This is also automatically performed for the code detection section. Therefore, it is possible to shorten the time required for checking and correcting the entire code name.

  As described above, according to the present embodiment, when a chord is detected from an acoustic signal, a chord related to a user's favorite song can be detected quickly and accurately by designating the genre and tension level of the song. can do. In addition, an appropriate chord can be found by referring to a frequently used chord progression according to the genre of the song. Furthermore, the code detection accuracy can be further improved by correcting the determined code and including the correction result in the database.

In the present embodiment, the example is shown in which the user inputs the time signature and the beat position and the unit of the chord detection section is a measure. However, the present invention is not limited to this, and as in the programs shown in Patent Documents 1 and 2, the time signature and beat position are automatically detected from the acoustic signal, and a plurality of chord detection sections are set within one measure. It can also be applied to code detection programs. In short, when calculating the power of the scale sound for the acoustic signal, detecting the bass sound based on the detection result, and further detecting the chord candidate and calculating the likelihood, the genre, tension level, or Furthermore , it is important to be able to narrow down the detection code candidates by the chord progression database.

  In this embodiment, the power spectrum is calculated by FFT calculation to obtain the power of the scale sound. However, the power of the scale sound may be obtained through a band pass filter corresponding to each scale sound. .

It is a block diagram which shows the principal part function of the code | cord | chord detection program which does not include a detection option. It is a figure which shows the hardware constitutions of the personal computer used for execution of the code | cord | chord detection program which concerns on one Embodiment of this invention. It is a flowchart which shows the principal part of the code | cord | chord detection program which concerns on one Embodiment of this invention. 4 is a detailed flowchart of code detection processing shown in FIG. 3. It is a figure which shows the example of the table which shows the pitch from the root sound of the structure sound for every chord type and each chord type. It is a figure which shows the example of the coefficient used for every overtone according to a genre. It is a figure which shows the example of narrowing down the chord for every tension level and each tension level. It is a figure which shows the power correction coefficient for every tension level. It is a figure which shows an example of the coefficient of the likelihood which multiplies for every chord progression and chord progression. It is a principal part flowchart of a code detection process including selection of a detection option.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Personal computer, 5 ... Display apparatus, 9 ... Keyboard, 11 ... CD-ROM drive, 12 ... Acoustic signal input part, 13 ... Acoustic signal storage part, 15 ... Code detection area setting part, 16 ... Power calculation part, 18 ... Bass sound detection unit 19. Chord candidate detection unit 20. Likelihood calculation unit 21. Chord name determination unit

Claims (6)

In a code detection program for causing a computer to function as a sound signal code detection device by being read and executed by a computer,
Reading an acoustic signal of a song from an acoustic signal source;
Detecting the power of each tone of the acoustic signal;
Setting a plurality of chord detection sections in the acoustic signal;
Detecting from the power the power of each tone of the planned bass detection range in the portion corresponding to the first beat of each chord detection section;
Determining a sound having the maximum power among the scale sounds as a bass sound;
Extracting a predetermined number of pitch names in descending order of power from the scale sounds of the entire chord detection section, and detecting chord candidates each having the predetermined number of pitch names as a root sound;
Calculating likelihood from the average power magnitude of all chord constituent sounds for the chord candidates;
Determining a code candidate having the maximum likelihood as a code;
And accepting the input of the genre of the song that is the code detection target,
The step of determining the base sound further includes the step of using the sound having the maximum power of the fundamental tone and harmonics of each scale tone as the base tone, and the step of separately weighting the power of the fundamental tone and harmonics according to the genre. A code detection program comprising: In a code detection program for causing a computer to function as a sound signal code detection device by being read and executed by a computer,
Reading an acoustic signal of a song from an acoustic signal source;
Detecting the power of each tone of the acoustic signal;
Setting a plurality of chord detection sections in the acoustic signal;
Detecting from the power the power of each tone of the planned bass detection range in the portion corresponding to the first beat of each chord detection section;
Determining a sound having the maximum power among the scale sounds as a bass sound;
Extracting a predetermined number of pitch names in descending order of power from the scale sounds of the entire chord detection section, and detecting chord candidates each having the predetermined number of pitch names as a root sound;
Calculating likelihood from the average power magnitude of all chord constituent sounds for the chord candidates;
Determining a code candidate having the maximum likelihood as a code;
The step of accepting the input of the tension level,
The code detection program characterized in that the step of detecting the chord candidates further includes a step of narrowing chord candidates down to chords having constituent numbers corresponding to tension levels.   The chord according to claim 2, wherein the step of calculating the likelihood further includes a step of performing power correction in accordance with a tension level based on a pitch from a root tone with respect to a constituent tone of a detected chord candidate. Detection program. Based on the power of each tone of the musical signal of the song, the base sound is determined as the sound having the maximum power of each scale sound in the planned base sound detection range in the portion corresponding to the first beat of each chord detection section Sound detection means;
A chord candidate detection means for extracting a predetermined number of pitch names in order of increasing power from the scale sounds of the entire chord detection section, and detecting chord candidates each having the predetermined number of pitch names as a root sound;
A likelihood calculating means for calculating a likelihood from an average power level of all chord constituent sounds for the code candidate, and configured to determine a code candidate having the maximum likelihood as a code. In the code detection device,
Comprising genre input means for designating the genre of a song consisting of an acoustic signal that is a code detection target;
The bass sound detecting means is configured to use the sound having the highest fundamental and harmonic power of each scale as a bass sound, and separately weight the fundamental and harmonic power according to the genre. A code detection device characterized by the above. Based on the power of each tone of the musical signal of the song, the base sound is determined as the sound having the maximum power of each scale sound in the planned base sound detection range in the portion corresponding to the first beat of each chord detection section Sound detection means;
A chord candidate detection means for extracting a predetermined number of pitch names in order of increasing power from the scale sounds of the entire chord detection section, and detecting chord candidates each having the predetermined number of pitch names as a root sound;
A likelihood calculating means for calculating a likelihood from an average power level of all chord constituent sounds for the code candidate, and configured to determine a code candidate having the maximum likelihood as a code. In the code detection device,
A tension level input means for designating the tension level;
The chord detection apparatus further comprises a chord candidate detection means for narrowing chord candidates to a chord having a number of constituent sounds corresponding to a tension level. The likelihood calculation means is configured to calculate the likelihood with the power corrected according to the tension level by the pitch from the root sound for the constituent sound of the detected chord candidate. 5. The code detection device according to 5 .

Images