JP4530199B2 - Model creation device and model creation program for natural instrument musical sound judgment device - Google Patents

Description

  The present invention relates to a model creation device and a model creation program in a natural musical instrument musical sound judgment device, and more particularly, to create a model for judging whether or not an operation as instructed by a performance instruction means has been performed in the performance of a natural musical instrument. The present invention relates to a model creation device and a model creation program in a musical instrument tone determination device capable of performing the same.

  2. Description of the Related Art Conventionally, an electronic musical instrument having a key pressing instruction means for self-studying the performance of a keyboard instrument is known. For example, in Patent Document 1 below, performance information stored in advance is sequentially read out to a display circuit, and a display lamp provided corresponding to each key of the keyboard is energized according to the performance information to give a key depression instruction. The pitch of the key pressed by the performer is detected by the key scanning circuit, and when the pitch matches the pitch of the performance information indicated by the display lamp, the next performance information is read to the display circuit. An electronic musical instrument having a key instruction means is described.

  The key press instruction according to the following Patent Document 1 can be easily realized with an electronic musical instrument provided with a key scan circuit, but is difficult to realize with a natural musical instrument such as an acoustic piano without a key scan circuit. Although it is conceivable to provide a key scan circuit for a natural musical instrument, the work of retrofitting key switches and display lamps necessary for the key scan circuit for each key is complicated and cannot be performed easily.

  In order to enable performance instructions such as key pressing even for natural musical instruments, the Applicant takes a single note of each pitch from a microphone and creates and stores the power spectrum of each single note as a model beforehand. Sometimes it reads out the model of the next sound to be played according to the performance instruction, compares the power spectrum of the performance sound captured with the microphone, and determines the musical sound of a natural instrument by advancing the performance instruction when both approach the distance within the threshold An apparatus was proposed in Japanese Patent Application No. 2003-134372 (prior application).

This device compares a model created by actually playing a natural instrument with the performance sound at the time of the lesson, so it is not necessary to detect parameters such as the pitch of the performance sound, and it is a very simple calculation, In addition, even if the natural instrument is out of tune, it is possible to detect the match between the model and the performance sound during the lesson without any problem. The power spectrum must be created and stored in advance as a model without error.
Japanese Patent Publication No. 2-705

  In the above-mentioned prior application, it is desirable to detect the power spectrum of the model and the performance sound at the time of lesson as early as possible. Therefore, when the model is created, the attack of the performance sound after key depression is detected, and the power spectrum at the time of attack detection is calculated. Remember as a model. An attack is detected when the volume or amplitude exceeds a threshold that is greater than or equal to the expected noise volume or amplitude, and if a volume or amplitude that exceeds the threshold is simply input, the attack is detected and a model is created. .

  However, there is a possibility that the performer may press the key with various strengths, and it is not preferable to create a model based on the performance sound generated with different key pressing strengths for each key. In the lesson, if the next note to be played is a chord, a single note model that composes the chord is synthesized to create a model. Since they differ greatly, it is not appropriate to synthesize by normalizing the volume of each single tone during model synthesis. From the above, we ask the performer to play each key with the same strength when creating a model, and create a model based on the performance sound generated by it. It is necessary to devise a method for synthesizing the product as it is and normalizing it as a whole after synthesis.

  Also, since attack detection is affected by the volume or amplitude of the input digital signal, the level of the input performance sound needs to be set appropriately. Furthermore, even if you play with the same strength, there is a large difference in the volume generated by the range (pitch), so if you simply guide the maximum volume to the same level, you will have to play with the same strength It is not possible. In addition, if the key is pressed too strongly when creating a model, a high-order harmonic component appears strongly, which makes it difficult to detect the coincidence between the created model and the power spectrum of the performance sound during the lesson. For this reason, it is desirable to have the key pressed at a normal strength or slightly weaker when creating a model.

  Also, when creating a model, the keys of the range to create the model are pressed one after another to create and store a model of each single note, but if you press the next key too early, the reverberation of the previous performance sound However, there is still a problem that an extra frequency component is included and the original model corresponding to the depressed key is not created. Then, even if you press the correct key during the lesson, you will not be able to get a match with the model. You can eliminate this inconvenience by waiting until there is enough reverberation to start creating the next key model, but it is wasteful to spend more time than necessary for each single note. It takes a lot of time to create it. Also, it is not clear at what point in time the next key is pressed to enter the model creation.

  In addition, when the power spectrum of a single tone at each pitch described above is created as a model, the input sound from the microphone includes a piano action sound in addition to the piano sound. The Fourier spectrum (FFT) is used to compare the power spectrum of the model and the performance sound during the lesson. The power spectrum of this action sound is a low frequency of the power spectrum of the model and the performance sound during the lesson. In the club.

  In the case of high-pitched keys, the power spectrum of the action sound is stronger than the vibration sound of the string that is the original piano sound, so when comparing the model and the performance sound at the lesson, the power spectrum of the action sound matches Will be evaluated relatively large. Since this action sound includes the model and the performance sound at the time of the lesson, it is inconvenient that a correct match between the two cannot be obtained.

  To prevent the action sound from being evaluated, the action sound can be removed. However, the action sound is like noise emitted from the action part, and its power spectrum is different every time, so it is difficult to select and remove only the action sound.

  In addition, normalization of the power spectrum is performed as a pre-process for comparing the power spectrum of the model and the performance sound at the time of the lesson, but simply normalizing the power spectrum with the volume will also normalize noise due to the external environment. Normalizing a weakly played sound results in a significant increase in the noise component. Therefore, even if such noise is included in the performance sound at the time of the model or lesson, a correct match between the two cannot be obtained.

  When FFT is performed to create a power spectrum, a DC (direct current) component appears in the power spectrum. Even if such a DC component is included in the power spectrum of the performance sound at the time of the model or lesson, a correct match between the power spectrum of the performance sound at the time of the model and the lesson cannot be obtained.

  Furthermore, when the volume is measured for normalization of the power spectrum when comparing the power spectrum of the performance sound between the model and the lesson, the volume cannot be measured correctly due to the influence of the DC component. Since the DC offset on the analog signal before A / D conversion is the same as the digital signal after A / D conversion, if you try to find the volume from the amplitude of the waveform, the DC offset will correct the volume correctly. It cannot be measured.

  Therefore, in order to correctly compare the power spectrum of the model and the performance sound at the time of the lesson, it is necessary to devise measures not to be affected by noise components and DC components.

  An object of the present invention is to make it possible to create a model for determining whether or not an operation in accordance with an instruction from a performance instruction means has been performed in playing a natural musical instrument without error.

In order to solve the above-described problems, the present invention is to adjust the level of a single performance sound generated by actually playing a natural instrument at the time of model creation, and to adjust the volume of the performance sound whose level is adjusted. In addition, the attack is detected in comparison with a threshold value set based on the volume of the noise captured in the level adjustment state, and the power spectrum of the performance sound when the attack is detected is calculated and stored as a model of the single sound. There are features.

In addition, the present invention captures a single performance sound generated by actually playing a natural instrument at the time of model creation , stores the single performance model as a model, compares the volume of the captured performance sound with a threshold value, It is characterized in that when the sound volume falls below the threshold continuously for a certain period of time, it is determined that there is no sound and the next single-tone model is created.

In addition, the present invention captures a single performance sound generated by actually playing a natural instrument at the time of model creation, obtains a power spectrum of the captured performance sound, and calculates a fundamental frequency of a single pitch to be modeled. It is characterized in that a power spectrum obtained by cutting a lower frequency component is stored as a model of the single sound .

In addition, the present invention obtains a power spectrum of a sound captured without playing a natural instrument at the time of creating a model, and uses the power value of the spectrum with the maximum power in the power spectrum as a noise reference value, and the power spectrum of a single performance sound the power spectrum power values were cut below the noise reference value is characterized in that stored as the single sound model in.

Furthermore, the present invention takes in a single performance sound generated by actually playing a natural musical instrument when creating a model, obtains a power spectrum of the performance sound, and cuts a component near the lowest frequency in the power spectrum. It is characterized in that the power spectrum is stored as a model of the sound .

Incidentally, silence detection, action sound or noise and DC component removal techniques can be employed in appropriate combination, the invention also can be configured as an apparatus or program.

According to the features of the present invention, it is possible to appropriately adjust the volume of the performance sound at the time of creating the model, to detect well the attack of the single performance sound input after the level adjustment, and if the volume of the performance sound is not appropriate Since a message to that effect can be sent to prompt the player to play with almost constant strength, an error-free model can be created.

  Further, by detecting the silent state and proceeding to the next model creation, it is possible to eliminate the influence of the reverberation of the performance sound in the previous model creation, and it is possible to create the models one after another in peace.

  Furthermore, the power spectrum below the fundamental frequency of the model is removed when creating the model, the power spectrum with a power lower than the maximum noise power is removed from the model power spectrum, or the lowest frequency in the power spectrum of the performance sound By removing nearby components and DC components, it is possible to prevent the effects of action sounds, noise, and DC components from appearing in the model.

  FIG. 1 is a block diagram of a musical tone determination apparatus to which the present invention is applied, and is an example configured as a musical tone determination apparatus for an acoustic piano (hereinafter, simply referred to as “piano” unless otherwise distinguished from an electronic piano). is there.

  1 is a personal computer including a CPU 1, a ROM 2, a RAM 3, a sound source device 4, a speaker 5, an A / D converter 6, an input device (keyboard or mouse) 7, and a display device 8. Can be configured. As the display device 8, a known display means for displaying the processing result of the personal computer, such as a liquid crystal display or a cathode ray tube, can be used. An external storage device 9 is connected to the personal computer, and a microphone 10 is connected via the A / D converter 6. The microphone 10 is provided to capture the sound generated by the piano P, and is preferably disposed in the piano P.

  The CPU 1 includes a volume measuring unit 11, a spectrum creating unit 12, an attack detecting unit 13, a spectrum comparing unit 14, and a key pressing instruction unit 15 as main functions. That is, the sound volume measurement unit 11, the spectrum creation unit 12, the attack detection unit 13, the spectrum comparison unit 14, and the key press instruction unit 15 are configured as software. Of course, these parts can also be configured by hardware.

  The volume measuring unit 11 detects the level (volume) of the sound input from the microphone 10. The spectrum creating unit 12 has an FFT (Fourier transform) function for obtaining a power spectrum from a digital musical tone signal input through the microphone 10 and the A / D converter 6. The power spectrum obtained by the FFT is stored in the RAM 3.

  The attack detection unit 13 detects an attack based on the detected power spectrum. The spectrum comparison unit 14 compares the power spectrum distances created by the spectrum creation unit 12 in the model creation phase and the lesson phase, and determines whether or not the performance has been performed according to the key depression instruction. The model creation phase and lesson phase will be described in detail later.

  The key pressing instruction unit 15 creates display data for displaying on the display device 8 a key pressing instruction according to the performance information. Display data is generated in response to a performance start instruction input from the input device 7 and input to the display device 8.

  The processing in the musical tone determination apparatus includes a model creation phase in which a single tone model of each pitch is created according to a performance sound generated when the performer actually plays the piano P, and a performance sound input according to the lesson performance of the performer. And the model are compared to determine a match, and the lesson phase is advanced according to the determination result. Hereinafter, the performance sound generated in the model creation phase is particularly referred to as “model creation performance sound”.

  In the model creation phase, first, the player presses the keys of each pitch of the acoustic piano P one by one, and the generated model creation performance sound is captured from the microphone 10. The analog signal thus input is converted into a digital signal by the A / D converter 6, and the volume of the digital signal is measured by the volume measuring unit 11. The attack detection unit 13 detects the rising edge (attack) of the sound from the volume measured by the volume measurement unit 11.

  When an attack is detected, the spectrum creation unit 12 creates a power spectrum by performing FFT on the digital signal. The power spectrum created by the spectrum creating unit 12 is stored in the RAM 3 or the external storage device 9 as a single tone model of the pitch of the depressed key.

  In the lesson phase, the performance information of the music stored in the RAM 3 or the external storage device 9 is read, and the key press instruction for the first musical tone of the music is displayed on the display device 8. At the same time, the musical tone model at the beginning of the music is read from the RAM 3 or the external storage device 9. The sound pressed by the piano P is taken from the microphone 10 in accordance with the key pressing instruction displayed on the display device 8. An analog signal input through the microphone 10 is converted into a digital signal by the A / D converter 6, and the volume of the digital signal is measured by the volume measuring unit 11.

  The spectrum creation unit 12 creates a power spectrum by performing FFT on the digital signal if the volume measured by the volume measurement unit 11 is equal to or greater than a threshold value. The attack detection unit 13 detects a sound attack attack from the power spectrum created by the spectrum creation unit 12. If an attack is detected even once, the spectrum comparison unit 14 compares the power spectrum obtained by FFT with the read model, and if they match, the spectrum pressing unit 14 instructs the display device 8 to press the key. Advance to a musical tone.

  In the comparison, if the key depression is a single note, the single note model can be used as it is, but if it is a chord, the single note model may be synthesized to compare the power spectrum. The lesson is advanced by repeating the above sequence of actions.

  FIG. 2 is a flowchart showing the processing of the CPU 1 in the model creation phase. When the model creation is instructed, a screen prompting the user to select a song or a lesson (skill) level is first displayed on the display device 8 to ask the player to select a song or to select a lesson level. (S1). According to this selection, a model-created sound range is set (S2).

  When a musical piece is selected, the performance information of the musical piece is scanned, and for example, as shown in FIG. 3, a musical range including at least the pitch included therein is automatically set as a musical range for model creation. The range of the model creation may be automatically set according to the level of the selected music, or the pitch that appears in the selected music is shown to the performer, and the range that includes at least those pitches May be instructed by the performer to set the model-created range.

  When the lesson level is selected, the range included in the music of the selected lesson level is set as the model-created range. For example, as shown in FIG. 4, a wide sound range is set according to the introductory, beginner, intermediate, and advanced levels. When the model-created range is set, the first tone of the model-created pitch, for example, the lowest tone included in the model-created range is set (S3).

  Next, the key corresponding to the pitch of the model created pitch is displayed on the keyboard displayed on the display device 8, and the player is instructed to press the key (S4). At this time, the sound source 4 may be instructed to generate a sound of the pitch that has been instructed to press the key. When the performer presses the designated key, the model-created performance sound emitted from the piano P is captured by the microphone 10. If the sound of the pitch that was instructed to be pressed in S4 is generated from the speaker 5, it is possible to make the player aware of the mistake in pressing the key compared with the model-created performance sound emitted from the piano P.

  The analog signal input through the microphone 10 is converted into a digital signal by the A / D converter 6, and the volume is measured by the volume measuring unit 11. Volume measurement continues until the key release instruction. If the volume measured by the volume measuring unit 11 exceeds the threshold, it is regarded as an attack (S5). This attack detection process will be described in detail later. The spectrum creation unit 12 creates a power spectrum by performing FFT on the digitel signal, and the volume measurement unit 11 calculates the volume (S6). The power spectrum and sound volume when the attack is detected are temporarily stored in the RAM 3 as a buffer for later use.

  When the power spectrum is created and the volume is calculated, an instruction to release the key is displayed on the display device 8 (S7). The key release instruction may be a sound or voice instruction from the speaker 5. After that, the process proceeds to the next process, but the piano P needs to be silent before the next process so that it is not affected by the reverberation of the previous pitch. Wait until the sound volume calculated by the measurement unit 11 is equal to or lower than the silence detection threshold (S8). If the volume falls below this threshold, it is determined that the sound is substantially silent. In the case of a piano, there is considerable reverberation, so silence detection is important to create a model without error.

  If it can be determined that the silent state has been reached, it is determined whether the maximum value of the volume or the maximum value of the amplitude during the period from the key pressing instruction is within the allowable range (S9). If it is determined that the volume or amplitude is not within the allowable range and that it is too high, for example, a message “Please play smaller” is displayed, and it is determined that it is too low. In such a case, a message such as “Please play larger” is displayed, and the process returns to S4 to instruct the performer to press the key again. This is because the power spectrum when pressed extremely strongly or weakly is not suitable as a model. The maximum volume or amplitude of the period from the key press instruction to the previous period is checked instead of the volume and amplitude at the time of attack detection. The volume and amplitude at the time of attack detection are not peak values, and the volume and amplitude are usually This is because the peak value is obtained after the attack is detected.

  If it can be determined that the maximum volume or maximum amplitude during the period from the key press instruction is within the allowable range, the temporarily stored power spectrum and volume are pressed. Is stored in the RAM 11 as a sound model (S10).

  A single-tone model is stored by S1 to S10, and a single-tone model within a set range can be stored by repeating the above processing sequentially for successive pitches. Instructs the user to press the same key again so that the model of a single note of each pitch can be stored correctly even if noise is generated due to the environment, etc., and the resulting model matches the previously stored model If it does, it tries to leave it as a regular model.

  S11 to S17 described below is this process. After storing the model of the first key depression in the RAM 3, the player is instructed to depress the same key again (S11). This key pressing instruction may be performed by display on the display device 8, sound generation by the sound source 4, or both, as in the first time. Next, the model stored in S10 is read (S12).

  The model-created performance sound by the player's key depression is captured by the microphone 10. After the analog signal input through the microphone 10 is converted into a digital signal by the A / D converter 6, the volume is detected by the volume measuring unit 11 (S13). When a sound volume of a certain level or more is detected, the spectrum creating unit 12 creates a power spectrum by performing FFT on the digital signal, and this is used as a second model (S14). By creating a power spectrum only after a sound volume of a certain level or more is detected, useless processing can be eliminated, and the load on the CPU when this apparatus is configured by a personal computer or the like can be reduced.

  After the second model is created, it is checked whether an attack has already been detected (S15). At first, since no attack is detected, the process proceeds to the next step to detect the attack (S16). Attack detection is to detect the rise of sound, and can be executed by, for example, a method of detecting by volume change or a method of detecting from the power increase of each spectrum of the power spectrum.

  If no attack is detected in S16, the process returns to S13 and starts again from the detection of the volume. If an attack can be detected, the process proceeds to a comparison process (S17) between the model read in S12 (first model) and the model created in S14 (second model). In the attack detection (S16), once it is detected during the same musical tone, it is not necessary to detect it after the second time. Usually, the attack is detected only at the moment when the sound rises, and the power spectrum of the musical sound at the time of the attack does not necessarily match the first model. In the comparison process, it is preferable to normalize the volume so that the volume is the same in order to eliminate the difference in volume between the first and second key presses.

  In the comparison process (S17), for example, the sum of the insufficient power spectrum that is insufficient for the second model compared to the first model is calculated as a distance, and this is compared with a predetermined threshold, and the distance exceeds the threshold Can be executed by a method in which it is determined that they do not match, and when the distance is equal to or less than a threshold value, they are determined to match. This method is effective for instruments such as a piano that have a lot of reverberation. At this time, even when the distance is equal to or smaller than the threshold value, it is possible to determine that there is a discrepancy when a characteristic for each constituent sound is not found in the power spectrum of the tone of the key that has been pressed.

  If it is determined that there is a mismatch in S17, the process returns to S4 and starts again from the first model creation. The model created by this re-processing is overwritten on the previously stored model.

  If it is determined that they match in S17, the model stored in RAM 3 in S10 is left as a correct model. Next, it is determined whether the sound of the created model is the final sound of model creation (S18). If it is determined that it is not the final sound (highest sound) in the range set in S2, the model-created performance sound is advanced by one (set in the RAM 3), for example, the sound of the next higher pitch is set, and the process returns to S4 ( S19) The same processing is repeated for the sound. If it is determined in S18 that the sound is the final sound, all the created pitch models are stored in the RAM 3 or the external storage device 9 (S20), and the model creation is terminated.

  Next, the attack detection process (S5) will be described. Attack detection is performed based on a threshold corresponding to noise or an input signal level. Since it is desirable to detect the power spectrum of the model and the performance sound at the time of lesson as early as possible, the attack of the performance sound after the key depression is detected at the time of model creation, and the power spectrum at the time of the attack detection is stored as a model.

  At the time of attack detection, first, the volume measuring unit 11 measures the volume of noise (background noise) without making an excessive sound while the piano P is not being played. This measurement time is not particularly problematic, but may be, for example, about 5 seconds, and if it is too long, the possibility of noise entering increases, so it may be about 1 second.

The sound volume measuring unit 11 measures the sound volume by the following formula every time N output samples from the A / D converter 3 are accumulated. Note that S _i represents the i-th output (amplitude).

When the sampling frequency is 11025 Hz, the value of N is preferably about 128. In this case, the volume is repeatedly calculated about every 12 msec. For example, the maximum value for 5 seconds of the volume _PW obtained by the above equation is stored as the maximum value P _WNOISE of the noise volume.

When the maximum value _{P WNOISE} noise volume is stored, the value obtained by adding the volume _{P WATTACK} regarded as an attack sound volume threshold _{P WTH} Attack detection thereto. That is, it is determined that the piano P has been played when a certain level of sound is input above the noise level. It shows the relationship between _{P WNOISE, P WATTACK} and _{P WTH} in FIG.

  Since the attack detection described above is affected by the volume or amplitude of the input digital signal, the input level of the microphone 1 needs to be set appropriately.

  FIG. 6 is a block diagram showing a detailed configuration of the microphone 2 and the A / D converter 3 (FIG. 1). The output of the microphone 2 is input to the microphone volume 201 and then input to the A / D converter 3 via the microphone amplifier 202. By adjusting the microphone volume 202, the input signal level of the A / D converter 3 can be changed according to the sensitivity of the microphone and the volume of the input sound. For example, when the musical tone determination apparatus is mounted on a personal computer, the mixer control of the recording device corresponds to this.

  When the performance sound of the piano P is captured by the microphone 2 and input to the A / D converter 3, the display device 8 is used when creating the model so that the digital signal after A / D conversion is not buried or clipped by noise. Display the peak meter.

  For example, as shown in FIG. 7, the peak meter displays the maximum value of the amplitude of the digital signal after A / D conversion by changing the lighting condition (color change), and the maximum value is When 0, no display is turned on. Further, when the maximum value becomes the maximum value of the amplitude determined by the number of quantization bits of the digital signal or about half of the maximum value, the display is turned on.

  If the microphone volume 201 is adjusted so that about 70 to 80% of the peak meter set as above lights up when the piano P is played strongly, an appropriate input that will not be buried or clipped by noise The signal level can be set. It should be noted that the key played during this adjustment is more suitable for the mid range and the low range than for the high range. This is because an appropriate volume can be obtained in the middle range and the low range. The fixed time interval for updating the maximum value of the amplitude is suitably about 10 to 50 msec. This is because if the speed is too fast, the eyes cannot catch up with the movement of the peak meter, and if the speed is too slow, the instantaneous volume change will not be displayed.

The volume P _WATTACK regarded as an attack is changed in accordance with the shake level of the displayed peak meter in order to properly set the microphone input level. There is no problem if about 70 to 80% of the peak meter can be swung by adjusting the microphone volume 201. However, since the sensitivity of the microphone 2 is low, even if the microphone volume 201 is maximized, the peak meter does not swing excessively. In the case of the same threshold as when about 70 to 80% is swung, the attack is not detected even if played strongly, or the attack is not detected unless played very strongly. Also, if you play it too strongly with a musical instrument such as a piano, high harmonics will be strong, so if you play strongly and create a model, you will not be able to get a match unless you play it as well during lessons.

Since the peak meter displays the average value of the amplitude of the digital waveform after A / D conversion as the volume P _W , if the peak meter can swing only half, the volume P _WATTACK regarded as an attack can be reduced to about half. Good. In other words, when it is experimentally known that the volume P _{WATTACK, which} is regarded as an attack when normal (the peak meter swings about 70 to 80% when played strongly), is 100th, it is just half that peak meter If it can only be _{shaken, the} volume P _WATTACK regarded as an attack may be reduced to the 50th position.

Next, the setting of the maximum volume allowable range used in the volume (amplitude) check process (S9) will be described. FIG. 8 shows the characteristics of pitch vs. maximum volume when a piano is played with almost the same strength from 88 keys from the lowest note pitch (note number) 21 to the highest note number 108. The maximum volume value is the maximum value of the volume P _W calculated by the above formula when each key is played. As can be seen from the figure, there is a large difference in volume value depending on the range (pitch) even when playing with the same strength. In particular, in the range close to the highest sound, only a very small volume maximum value can be obtained. Since there is a large difference in the maximum volume value depending on the sound range in this way, it is not possible to play with the same strength simply by guiding the maximum volume value to the same level.

  If the characteristics of the pitch vs. volume maximum value in FIG. 8 are known in advance, the allowable range of the volume maximum value may be changed according to this characteristic, but this characteristic varies depending on the individual piano and depends on the characteristics of the microphone. However, since the sound volume differs for each sound range, a sound volume maximum value allowable range common to all devices cannot be set in advance.

  This can be solved by setting the maximum volume allowable range in each device by the following method. There are individual differences depending on the piano, and the frequency characteristics differ depending on the microphone. However, as long as a general microphone is used in an acoustic piano, the volume per highest sound is the smallest and the volume per mid-range is definitely high. Therefore, the highest sound and one midrange sound (such as the middle sound) are played with a normal touch suitable for model creation or a slightly weak touch, and the maximum volume at that time is stored.

A range in which the two maximum sound volume values have a certain width is defined as a maximum sound volume allowable range. That is, if the maximum volume value of the highest sound is P _WH and the maximum volume value of the middle range is P _WM , the maximum volume value of the input sound is changed from P _WH × R to P _WM / R ′ (R, R ′ is 1.5). Check if it is in the range of (about), and if it is not, press the key again with a message such as "Please play smaller" or "Please play larger" To get. R and R ′ may be the same value or different values.

  Also, if it is difficult to play with normal strength suitable for model creation, and it is difficult to play with the same level of strength, it is quite possible to play one note in the highest and middle range. Measure the maximum volume by playing strongly, analogize the volume of the volume that is a fraction of it with normal strength, and set the maximum volume allowable range with this volume as the reference value You may do it. Since there is a limit to the strong sound due to the structure of the piano, data without individual differences can be input by this method.

  It is also possible to have several notes in the range to be modeled play, create a characteristic of pitch vs. maximum volume from the maximum volume value obtained, and determine the allowable range of maximum volume from this characteristic. it can. In this case as well, if it is difficult to play with standard strength, measure the maximum volume level by playing each key as hard as possible. It can be a reference value.

  Note that the quantity used to determine the key-pressing strength of each key is not necessarily the maximum volume value, and may be the maximum value of the amplitude of the digital waveform after A / D conversion. It is also possible to use the average value instead of the maximum value.

  Next, the key release instruction (S7) and the silence detection process (S8) will be described. FIG. 9 shows an example of a key pressing instruction in which a keyboard is displayed on the display device 8 at the time of model creation, and the key pressing is instructed by changing the color of only the key to be pressed. Simultaneously with the key pressing instruction, the sound source 4 may be instructed to generate sound from the speaker 5.

  When the key is pressed in accordance with the key pressing instruction, a power spectrum is created, and the volume is calculated, the key release is instructed (S7). The key release instruction is the opposite of the key press instruction, and the key color displayed on the display device 8 is restored or displayed in a different color from the key press instruction, and is silent as will be described below. This is done by returning the key color when it is determined that it has become. If sound is produced from the speaker 5, the sound source 4 is instructed to stop the sound.

Since the reverberant sound remains even after the key is released, it waits for the reverberant sound to disappear before the key depression for the next sound model creation or the second key depression (S8). The volume of the reverberant sound can be obtained by the above formula for obtaining the volume P _W.

FIG. 10 shows a change in volume obtained when the lowest note (note number 21) is pressed and released immediately on an acoustic piano, and is obtained by actual measurement. As shown in the figure, the volume of the reverberant sound after the key release decreases significantly immediately after the attack, but then gradually attenuates while going up and down, so the volume simply falls below the threshold P _WTH once. If it is determined that there is no sound, the volume may increase again thereafter. So, even if equal to or less than a threshold value P _WTH counts the elapsed time as well as keep track of the volume, to clear the elapsed time Once exceeded again volume threshold P _WTH. Then, it is determined that there is no silence _until the elapsed time equal to or less than the threshold value P _WTH is continuously 1 second, for example.

The threshold value P _WTH used here can be determined as follows. Prior to model creation, measure noise volume. This can be done by having the performer quiet so as not to make a sound, taking in the sound for about 1 to 5 seconds from the microphone 10 and calculating the volume by the above formula. The maximum value of the sound volume thus measured is set as the maximum value P _WNOISE of the noise sound volume. Threshold _{P WTH} is a plus volume _{P WSILENT} regarded as silence to the maximum value _{P WNOISE} noise volume.

It can be determined that the volume _PWSILENT regarded as silence is closer to silence as the volume is smaller. However, since the attenuation of the sound decreases as the time elapses, the volume decreases until the volume falls below the threshold _PWTH. Therefore, it should be set so that the sound is judged to be silent when the volume is reduced to a level that does not appear in the power spectrum.

The volume _PWSILENT regarded as silence may be changed according to the amplitude of the digital signal after A / D conversion. For example, when the peak of amplitude of a digital signal after A / D conversion is displayed with a peak meter as shown in FIG. 7, it is normal (when the peak meter swings 70-80% when played hard) when the volume P _WSILENT regarded as silence is that just good at 50-position of the value has been found experimentally, the volume P _WSILENT regarded as silent also at the 25-position of the half when the peak meter is not shake only half position.

  FIG. 11 (a) shows the change in volume obtained when the note number 48 (a sound of 1 octave lower than the center position) is pressed with a normal strength on an acoustic piano, and is obtained by actual measurement. Is. FIG. 11B is an enlarged view of a part thereof. As shown in the figure, the volume decreases significantly immediately after the attack, but then gradually decreases while going up and down, and attenuates to the extent that it cannot be distinguished from noise in about 10 seconds. The lower the sound, the longer this decay takes. However, as soon as the key is pressed for the next sound model creation or the second key press, the sound volume is attenuated as quickly as possible so that it can be judged as silence. It is better to detect and guide to that effect.

  12 to 15 show power spectra obtained immediately after the attack, about 1.4 seconds after the attack, about 4.1 seconds after the attack, and about 5.8 seconds after the attack, respectively. FIG. 16 shows a power spectrum obtained when the next model creation sound (note number 50) is depressed slightly weakly.

By comparing FIGS. 12 to 15 and FIG. 16, it can be determined how much the volume is attenuated in FIG. 11 and the next model creation is not affected. In the case of the sound range in this example, a relatively large power spectrum is obtained even if played softly, so there is almost no influence on the next model creation even after 1.4 seconds or 4.1 seconds after the attack. In other words, the volume P _WSILENT regarded as silence may be about ₂₅₀ to 750, but it is preferable to _{make the} volume smaller and move to the next model creation as close to silence as possible.

On the other hand, FIG. 17 shows the change in volume obtained when the highest note (note number 108) is pressed with normal strength, and it can be seen that only a maximum volume of about 300 is produced. In this case, it is too large if the volume _PWSILENT regarded as silence is set to about ₂₅₀ to 750. From FIG. 10 and FIG. 16, it can be seen that about 10 to 75 is appropriate if the volume P _WSILENT regarded as silence is the same in the whole sound range. If it is larger than that, the maximum volume may be lower than P _WNOISE + P _WSILENT if the key is _pressed slightly weakly in the range where the volume does not come out like the highest note (note number 108), and it is smaller than that. This is because a situation in which it cannot be determined that there is no sound occurs when a little larger noise is input.

  As indicated by black circles in the key pressing instruction in FIG. 9, the mark display S is always displayed on the key such as the center “do” of the keyboard displayed on the display device 8 when the model is created, and the center of the actual piano keyboard is displayed. The correspondence between the two keys can be easily understood by sticking the sticker S with the same mark on the “do” key. Further, a logo 102 such as a manufacturer name is usually attached to the upper part of the keyboard of an actual piano. A mark corresponding to a predetermined position (position “B” in FIG. 9) of the logo 102 is displayed on the display device. 8 may be displayed. Thereby, the burden at the time of model creation can be reduced.

  FIG. 18 is a flowchart showing the processing of the CPU 1 in the lesson phase. When the lesson start is instructed, first, the model is read from the RAM 3 or the external storage device 9 into the work area (S21). The model to be read is a model of all pitches included in the range set in the model creation phase. Here, the model created last time or the model used last time may be automatically read, or the player may select and read from the models created so far.

  Next, a music piece to be performed by the performer is selected (S22). At this time, lesson song data stored in advance in the RAM 3 or the external storage device 9 may be searched to create a list of songs, and the player may select a song to be taught by himself / herself from the list. Then, a music piece suitable for the player's level may be automatically selected and presented on the device side. When the music to be lesson is determined, the performance information of this music is read from the RAM 3 or the external storage device 9 (S23).

FIG. 19 is a diagram showing a format example of performance information used for displaying a key pressing instruction on the display device 8 and performing accompaniment in a lesson. The performance information includes at least event data and timing data for instructing reading timing of the event data. The event data includes note-on data and note-off data including a note number (pitch). The timing data is set, for example, as time information from the end of one event to the next event occurrence. Event data and timing data are stored according to address progression as shown.
This performance information can be stored in the external storage device 9.

  Next, it is checked whether or not all the single tone models included in the selected music are included in the model read in S21 (S24). Here, if it is determined that the single tone models of all pitches of the selected music are not included, the process returns to S22 to instruct the selection of another music, or the process returns to S21 to read another model. Instruct.

  If it is determined in S24 that a single tone model of all the pitches of the selected music is included, the pitch of the first musical tone of this music is set as a variable on the RAM 3 (S25). At the same time, a key pressing instruction for causing the performer to press the key of this pitch is displayed on the display device 8 (S26), and further, this pitch corresponds to the pitch from the model read in S21. A single tone model is read (S27).

  The performer presses the key of the piano P according to the key pressing instruction displayed on the display device 8, and the performance sound generated at that time is captured by the microphone 10. After the analog signal input through the microphone 10 is converted into a digital signal by the A / D converter 6, the volume is detected by the volume measuring unit 11 (S28). Since the process of S28-S32 is the same as the process of S13-S17 of FIG. 2, detailed description is abbreviate | omitted. If it is determined that there is a mismatch in S32, the process returns to S28 and starts again from the detection of the volume.

  If it is determined in S32 that they match, first, attack detection is reset (S33), and it is determined whether or not it is the last musical tone of the selected music (S34). If it is determined that it is not the last musical tone of the music, the pitch of the next musical tone is set in the RAM 3 (S35), and the process returns to S26. The above processing is repeatedly executed until the last musical tone of the music, and when the end of the music is reached, it is asked whether to take a lesson again (S36) or whether another music is to be lessoned (S37). Move to the corresponding process.

  FIG. 20 is a flowchart showing an example of the comparison process (S17) between the first model and the second model, or the comparison process (S32) between the model and the power spectrum of the performance sound at the time of the lesson.

  First, a model for comparison with the performance sound of this time is created from the model read out in S12 or S27 (S41). In the case of a single performance sound, the single sound model may be used as it is, and in the case of a chord performance sound, a single sound model is synthesized. In order to eliminate the difference in volume between the model and the current performance sound, the power spectrum of the model and the current performance sound is normalized by the volume so that the volume at the time of model creation and the current performance sound is the same. In the case of a chord performance sound, normalization is performed by the sum of the volumes at the time of model creation of the sound composing the chord. In addition, you may normalize not with a sound volume but with the peak value of a waveform.

  When the model is created, the sum of the insufficient power spectrum that is insufficient for the current performance sound compared to the model is calculated as a distance (S42). This method is effective for instruments such as a piano that have a lot of reverberation.

  Next, a threshold for determining that the model and the current performance sound match is set (S43), and the distance calculated in S42 is compared with this threshold (S44). When the distance is greater than or equal to the threshold value, it is determined that they do not match, and when the distance is less than the threshold value, they are determined to match. At this time, even if the distance is smaller than the threshold, it may be determined that there is a mismatch if the characteristics of each component sound are not found in the power spectrum of the performance sound of this time (S45). If it is determined that there is a mismatch in S44 or S45, the process returns to S4 (S28) and starts again from the key pressing instruction (volume detection). If it is determined in S45 that they match, the process proceeds to S18 (S33).

  FIG. 21 is a diagram showing an example of display of a key press instruction by a piano roll, that is, a scroll method at the time of a lesson. A keyboard figure K is displayed at the bottom of the display screen 101 of the display device 8, and a key press instruction mark M is displayed above it. The vertical direction of the display screen 101 is the time axis, and the key pressing instruction mark M is displayed at a position corresponding to the key to be pressed, with a length corresponding to the musical performance length. The key press instruction mark M closest to the keyboard figure is a tone to be pressed next, and indicates a tone to be pressed in order from the side closest to the keyboard figure K.

  If it is determined in S32 (FIG. 18) that the distance between the power spectrum calculated from the musical sound model and the performance sound of the piano P to be depressed next is equal to or smaller than the threshold value, the piano roll is scrolled downward. Then, the operation stops when the key depression instruction mark M of the next musical tone to be depressed is lowered to the position closest to the keyboard figure K. Since the scroll of the piano roll is stopped until the musical tone instructed to be depressed is depressed, the performer can depress the next musical tone at his / her own pace. The performance information read in S23 (FIG. 18) can include not only musical sounds that prompt the player to press the key but also performance information of accompaniment. In this case, the accompaniment sound up to the timing of the sound to be pressed next is output from the speaker 5 through the sound source 4.

  The key to be currently pressed is indicated by a mark m. The length of the mark m indicates how long the key must be depressed, so that the player can know in advance when to release the key. As indicated by the bar line BL, in this example, key pressing instructions for two bars are simultaneously displayed on one screen, but the number of bars to be displayed simultaneously and the range of performance information to be displayed are arbitrary.

  Next, a method for reducing action sounds, noise components, and DC components included in a model in model creation will be described. In the following, a natural musical instrument is described as a piano (acoustic piano), but other natural musical instruments can be considered in the same manner.

  First, a method for reducing the action sound included in the model will be described. If the piano P is played when the sound is taken in from the microphone 2, the action sound of the piano P is always taken in. FIG. 22 is a diagram illustrating an example of a spectrum of sound captured from the microphone 2.

  The action sound has a different power spectrum each time and is like noise. The action sound appears mainly as a power spectrum with a strong component in the bass part, but the power spectrum of the original sound of the piano (string vibration sound) is considerably larger than the keys of the treble part, even if the action sound enters. It has little effect on the comparison between the performance sound of the key and the model.

  However, in the case of high-pitched keys, the power spectrum of the string vibration sound is small, so the effect of the power spectrum of the action sound is large, and the power spectrum of the action sound when comparing the performance sound of the pressed key with the model Will be greatly evaluated. Therefore, a correct comparison result between the performance sound of the pressed key and the model cannot be obtained. If the power spectrum of the action sound is the same every time, it can be subtracted and removed, but the power spectrum of the action sound is like noise and is different every time, so it is impossible to simply subtract and remove it. is there.

Therefore, when creating the model, the power spectrum of the fundamental frequency f ₀ or less of the pitch tone to be created is cut. However, since it is conceivable that deviation tuning piano P, the frequency to be actually cut the fundamental frequency f ₀ frequency slightly lower than, be, for example, semitone lower frequency preferred. If the pitch is lower than the frequency component contained in the action sound, the power spectrum of the action sound cannot be removed. Since the power spectrum of the action sound is relatively small, there is no problem in comparing the performance sound of the key pressed during the lesson with the model.

  Next, a method for reducing noise components included in the model will be described. When sound is taken from the microphone 2 when creating a model, noise is usually mixed. The noise includes ambient sound, noise entering the cable of the microphone 2, noise entering the analog circuit portion of the microphone amplifier and the A / D converter 3, and the like. There is no problem if the volume of the sound of the piano P is high and the noise is so low that it is not conspicuous. However, if the performance sound with a weak key press is used as a model, the volume of the model is low. If normalization is performed with the volume when comparing the performance sounds of the pressed keys, the noise component will be greatly increased.

  On the other hand, the performance sound of the key pressed during the lesson also contains noise, but if the volume of the performance sound is high, the power spectrum of the noise component will be relatively low, and the noise component will be normalized by the volume. The spectrum of remains small. For this reason, it becomes impossible to obtain a match between the performance sound of the model and the key pressed during the lesson.

  In order to eliminate this inconvenience, the sound is captured from the microphone 2 for several seconds without the player playing the piano P, and the input sound is subjected to FFT while being overlapped at a constant time interval. Measure and save the maximum power value at. This power maximum value is used as a reference value for determining noise. That is, a component that seems to be noise is removed by cutting a power spectrum having a power lower than the maximum power value from the model power spectrum. FIG. 23 shows examples of power spectra before noise removal (a) and after noise removal (b). Note that the value of the volume of the model is left as it is, and normalization is performed based on the value of the volume when comparing the power spectrum of the model and the performance sound during the lesson.

  Next, a method for reducing the DC component included in the model will be described. When the performance sound of the pressed key is FFTed and converted to a power spectrum, the DC component appears as the power spectrum of the lowest frequency. In some cases, the power spectrum of the DC component is stronger than the power spectrum of the performance sound of the key that is actually pressed. In such a case, when comparing the performance sound of the key pressed during the lesson with the model, the DC component is greatly evaluated, resulting in inconvenience. Therefore, considering the side lobe of the FFT, the lowest frequency in the obtained power spectrum and the adjacent power spectrum are forcibly made zero, and the portion considered to be a DC component is removed.

  Next, a method for removing DC when measuring the volume will be described. The power spectrum is normalized by the volume when comparing the performance sound of the model and the key that was pressed during the lesson. If the analog signal before A / D conversion has a DC offset, it is after A / D conversion. A DC offset is also added to the digital signal, and the measurement of the volume cannot obtain the correct volume because the average of the absolute values of the amplitudes of the waveforms per unit time is used as the volume. Therefore, the average value of the amplitude of the sound is obtained from the waveform, for example, one second before the moment when the volume is measured. Since this average value represents the DC of the waveform, the DC can be removed by subtracting this average value from the waveform of the volume measurement. FIG. 24 shows examples of waveforms before (a) DC removal and after (b) DC removal, and shows how the deviation of the center of the waveform amplitude from the 0 level disappears.

  The present invention can be implemented in various embodiments. For example, the created model is to be used continuously in subsequent lessons, so it needs to be created carefully and definitely, which is a heavy burden on the modeler. In order to reduce this burden, in S19 (FIG. 2), instead of raising the pitch according to the scale to the pitch directly above, the sound is generated so that the white key and the black key are separately pressed separately. It is good to raise the height. This is because if the pitch is increased in order according to the scale, the white key and the black key may be played alternately, and an error may occur in playing a key that does not correspond to the key instructed to be pressed.

  In addition, if all keys in the set model creation range are displayed in a certain color and the key is pressed by changing the color, etc., it is easy to see how many single-tone models must be created. .

  According to the flowchart of FIG. 2, when the first model and the second model do not match, the first model is created again, and according to this, only the first model is stored. However, when a model is created sequentially such as the first, second, third, etc. for a single note at the same pitch, multiple matches, for example, when two matches are obtained, are stored. Then, it is possible to finish the creation of a single tone model of the pitch and adopt the matched model as a regular model.

  Further, the techniques of attack detection, silence detection, action sound, noise, and DC component removal described above can be appropriately combined and employed.

It is a block diagram of the musical tone determination apparatus to which the present invention is applied. It is a flowchart which shows the process of a model creation phase. It is a figure which shows the example of the sound range setting of model preparation. It is a figure which shows the other example of the sound range setting of model creation. It is explanatory drawing of an attack detection process. It is a block diagram which shows the detailed structure of the part of a microphone and an A / D converter. It is a figure which shows the example of a volume display. FIG. 6 is a characteristic diagram of pitch versus volume maximum value. It is a figure which shows the example of the key pressing instruction | indication at the time of model creation. It is a figure which shows the volume change after the key release of the key of the lowest sound (note number 21). It is a figure which shows the volume change after the key release of the key of the lowest sound (note number 48). It is a figure which shows the power spectrum immediately after an attack. It is a figure which shows the power spectrum about 1.4 seconds after an attack. It is a figure which shows the power spectrum about 4.1 seconds after an attack. It is a figure which shows the power spectrum about 5.8 second after an attack. Shows the power spectrum obtained when the model creation sound (note number 50) is pressed. It is a figure which shows the volume change obtained when the highest note (note number 108) was pressed. It is a flowchart which shows the process of a lesson phase. It is a figure which shows the example of a format of performance information. It is a flowchart which shows the example of a comparison process. It is a figure which shows the example of the display of the key press instruction | indication by the scroll system at the time of a lesson. It is a figure which shows the example of the spectrum of the sound taken in from the microphone. It is a figure which shows the example of the power spectrum before noise removal and after noise removal. It is a figure which shows the example of the waveform before DC removal and after DC removal.

Explanation of symbols

1 ... CPU, 2 ... ROM, 3 ... RAM, 4 ... sound source, 5 ... speaker, 6 ... A / D converter, 7 ... input device, ... Display device 9 ... External storage device 10 ... Microphone 11 ... Volume measuring unit 12 ... Spectrum creation unit 13 ... Attack detection unit 14 ... Spectrum comparison unit DESCRIPTION OF SYMBOLS 15 ... Key press instruction | indication part, 101 ... Display screen, 102 ... Logo, 201 ... Microphone volume, 202 ... Microphone amplifier, K ... Keyboard figure, M ... Key press Instruction mark, m ... current key press mark, P ... acoustic piano, BL ... bar line

Claims (18)

The power spectrum of a single tone of each pitch is created and stored in advance as a model, the model of the sound to be played next is read according to the performance instruction, and the power of the performance sound generated by actually playing the natural instrument. In the model creation device in the musical instrument determination device for natural musical instruments that compares the spectra and advances the performance instruction when both approach a distance within the threshold,
A performance sound capturing means for level-acquiring and capturing a single performance sound generated by actually playing a natural instrument when creating a model,
Attack detection means for detecting an attack by comparing the performance sound captured by the performance sound capture means with a threshold value set based on the volume of the performance sound adjusted in level and the noise volume captured in the level adjustment state When,
Power spectrum calculation means for calculating the power spectrum of the performance sound when an attack is detected,
A model creation apparatus in a musical instrument determination device for a natural musical instrument, wherein the power spectrum calculated by the power spectrum calculation means is stored as a model of the single sound.   2. The model creating apparatus for a musical instrument determination device for a natural musical instrument according to claim 1, further comprising display means for displaying a volume of the performance sound whose level is adjusted when the model is created.   A judgment means is provided for judging whether or not the volume of the performance sound at the time of model creation is within an allowable range according to the characteristics of the performance-adjusted sound volume and the musical instrument pitch versus volume maximum value. 3. The model creation device for a musical instrument determination device for a natural musical instrument according to claim 1 or 2, wherein a message to that effect is transmitted when the value is not within the allowable range. Comparing means for comparing the volume of the performance sound captured by the performance sound capturing means with a threshold value, and silence state determining means for determining a silence state when the volume of the performance sound continuously falls below the threshold value for a predetermined time. the power spectrum calculated by the power spectrum calculating means stores as the monophonic models, the to the silent state determination unit claims 1, characterized in that moves the model creation of the next single tone when it is determined that the silence 4. A model creation device in the musical instrument sound determination device according to any one of 3 above . Before Symbol comparison means, tones of natural musical instrument according to claim 4, characterized in that it comprises a threshold setting means for setting a threshold value depending on the volume of the noise captured are level adjusted by the performance sound acquisition means A model creation device in a judgment device. A low frequency component removing means for cutting a low frequency component lower than a fundamental frequency of a pitch of a single tone to be modeled in the power spectrum calculated by the power spectrum calculating means, the low frequency component removing means 6. The model creation device for a musical sound determination device for a natural musical instrument according to claim 1, wherein a power spectrum in which a low frequency component is cut is stored as a model of the single tone. Power spectrum calculation means to obtain the power spectrum of the sound captured without playing the natural instrument at the time of model creation, and the power value of the spectrum with the maximum power among the power spectrum of the sound captured without playing the natural instrument as a noise reference A noise component removing unit that cuts a power spectrum whose power value is equal to or less than a noise reference value in the power spectrum of a single performance sound, and stores the power spectrum cut by the noise component removing unit as a model of the single tone The model creation device in the musical instrument tone determination device according to any one of claims 1 to 6 . DC component removing means for cutting a component near the lowest frequency in the power spectrum calculated by the power spectrum calculating means, and storing the power spectrum cut by the DC component removing means as a model of the single sound. The model creation apparatus in the musical sound determination apparatus of the natural musical instrument according to any one of claims 1 to 7 .   Calculated when comparing the power spectrum of the model and the performance sound with the volume calculation means that calculates the volume average based on the performance sound obtained by subtracting the amplitude average value and obtaining the average amplitude value of the performance sound waveform at the time of model creation 9. The model creation device for a musical instrument determination device for a natural musical instrument according to claim 1, further comprising normalizing means for normalizing the power spectrum of the model with the volume of the played sound. The power spectrum of a single tone of each pitch is created and stored in advance as a model, the model of the sound to be played next is read according to the performance instruction, and the power of the performance sound generated by actually playing the natural instrument. In a program for creating a model in a musical instrument determination device for natural musical instruments that compares spectra and advances performance instructions when both approach a distance within a threshold,
On the computer,
To adjust the level of a single performance sound generated by actually playing a natural instrument when creating a model,
A procedure for detecting an attack by comparing the captured performance sound with a threshold value corresponding to the volume of the performance sound whose level is adjusted and the volume of the noise captured in the level adjustment state;
A program for creating a model in a musical instrument determination device for a natural musical instrument for executing a procedure of calculating a power spectrum of a performance sound when an attack is detected and storing it as a model of the single sound.   11. The program for creating a model in a musical instrument determination device for a natural musical instrument according to claim 10, further comprising a step of displaying a volume of a performance sound whose level has been adjusted at the time of model creation.   The procedure for determining whether the volume of the performance sound at the time of model creation is within the allowable range according to the characteristics of the level-adjusted performance sound and the musical instrument pitch vs. maximum volume, and the performance sound volume is acceptable 12. The program for creating a model in a musical instrument determination device for a natural musical instrument according to claim 10 or 11, further comprising a procedure for sending a message to that effect when not within the range. Including a procedure for comparing the volume of the captured performance sound with a threshold value, and a procedure for determining a silent state when the volume of the performance sound continuously falls below the threshold value for a certain period of time. 13. The program for creating a model in a musical instrument determination device for a natural musical instrument according to claim 10 , wherein the program is stored as a model, and when it is determined that there is no sound, the model for the next single tone is created . 14. The musical sound determination device for a natural musical instrument according to claim 13, wherein the procedure for comparing the volume of the performance sound with a threshold value includes a procedure for setting the threshold value according to the volume of noise that has been acquired by adjusting the level. A model creation program. A method comprising cutting a power spectrum of a low frequency component lower than a fundamental frequency of a pitch of a single tone to be modeled, and storing the power spectrum from which the low frequency component has been cut as a model of the single tone. Item 15. A program for creating a model in the musical sound determination device for a natural musical instrument according to any one of Items 10 to 14 . The procedure for obtaining the power spectrum of a sound captured without playing a natural instrument at the time of model creation and the power value of the spectrum with the maximum power in the power spectrum of a sound captured without playing a natural instrument is used as a noise reference value. The power spectrum having a power value equal to or less than the noise reference value in the spectrum of the performance sound is cut as noise, and the power spectrum from which the noise has been cut is stored as a model of the single sound. A program for creating a model in the musical sound determination device for a natural musical instrument according to any one of claims 15 to 15 . 17. A power spectrum obtained by cutting a component near the lowest frequency in the power spectrum of a performance sound, and storing a power spectrum obtained by cutting a component near the lowest frequency as a model of the single sound. The program for model creation in the musical instrument musical sound judgment apparatus in any one of .   When calculating the average amplitude of a performance sound waveform for a certain period when creating a model, calculating the volume of the performance sound based on the performance sound obtained by subtracting the average amplitude value, and comparing the power spectrum of the model and the performance sound. A program for creating a model in a musical instrument determination device for a natural musical instrument according to any one of 10 to 17, including a procedure for normalizing the power spectrum of the model with the volume of the calculated performance sound.

Images