JP2646760B2 - Performance data separation device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (a) Industrial Field of the Invention The present invention relates to performance data (right hand data) of a right hand part to be played by a right hand and performance of a left hand part to be played by a left hand in a field of electronic musical instruments and the like. The present invention relates to a performance data separation device that separates performance data in which data (left hand data) is mixed into right hand data and left hand data, and more particularly to a performance data separation device that applies fuzzy inference to performance data separation.

(B) Prior Art Performance data generated when a keyboard instrument is played with both hands is generated in a state where right-hand data, which is performance data of a right-hand part, and right-hand data, which is performance data of a left-hand part, are mixed. Therefore, when this performance data is recorded, 1
The right hand data and the left hand data are recorded on one recording track in a mixed state. However, there is a demand for recording the performance data of the right-hand part and the performance data of the left-hand part on separate recording tracks for the purpose of automatically performing the right-hand part and the left-hand part separately. For this reason,
The method of playing and recording the right hand part and the left hand part separately in time, i.e., playing and recording one part with one hand, then playing the other part with the other hand while listening to it It is conceivable to record by recording. Alternatively, a method is conceivable in which performance data recorded by playing with both hands at the same time is later manually separated by a human into sound and left and right parts and re-edited to separate recording tracks.

(C) Problems to be Solved by the Invention However, the former method of recording the right hand part and the left hand part separately in time is different from the case of performing with both hands because the right hand part and the left hand part are not played simultaneously. The performance becomes unnatural in comparison. Also, after the latter, a method of manually separating the sound into left and right parts one by one manually and re-editing them on separate recording tracks requires a great deal of labor.

An object of the present invention is to provide a performance data separation device capable of easily separating performance data in which both hand parts are mixed into a right hand part and a left hand part in view of such a problem.

(D) Means for Solving the Problems According to the first aspect of the present invention, there is provided input means for inputting performance data, and right-hand data which is performance data to be played with the right hand by fuzzy inference on the performance data input from the input means. A performance data separating device for separating performance data to be played by the left hand into left hand data, wherein the separation means determines whether the input performance data is right hand data or left hand data based on the pitch thereof. First inference means for inferring, and second inference for inferring whether the inputted performance is right-hand data or left-hand data based on the performance data before and after the right-hand data.
The inference means is provided.

The invention according to claim 2 is characterized in that the second inference means operates when the performance data cannot be separated by the first inference means.

 Performance data separation device.

According to a third aspect of the present invention, there is provided a validity judging means for judging the validity of the result of the separation of the performance data by the separating means, and the condition of the fuzzy inference of the separating means is changed according to the judgment result of the validity judging means. It is characterized by making it.

(E) Operation In the performance data separation device of the present invention, for example, MIDI data is input as performance data, and fuzzy inference is made based on the pitch (key code) of the input MIDI data to determine whether the data is right-hand data or left-hand data. (First inference means). Furthermore, the relationship between the data before and after the input MIDI data (for example, whether the previous sound is right-hand data or left-hand data,
Then, based on whether the current sound is a sound close to the previous sound, etc.), a fuzzy inference is performed as to whether this data is right-hand data or left-hand data (second inference means). This separates the input MIDI data into right-hand data / left-hand data.

Further, in the invention of claim 2, the performance data is separated into right-hand data / left-hand data mainly based on the first inference means, and if the separation cannot be performed by the first inference means, the second inference means To separate it. As a result, when the performance data has a clearly high pitch or a low pitch, separation can be quickly performed, and even in the case of an intermediate pitch, reliable separation becomes possible.

According to the third aspect of the present invention, the validity determination unit determines the validity of the separation result of the separation unit. If it is determined that the fuzzy inference is not appropriate, the condition of the fuzzy inference is changed.
This always allows accurate separation.

(F) Embodiment FIG. 1 shows an embodiment of the present invention. This embodiment is an apparatus for inferring the distribution of right and left fingers in fingering of a piano. FIG. 1 shows the functional configuration of this device. This device reads the piano key code data stored in the MIDI format and infers which music should be played with the left or right hand. The inference result is converted to MIDI data again and stored. Since data such as the key-on intensity of each music is stored in the MIDI data in addition to the key code data, in this embodiment, data extraction 2 for extracting only necessary key code data from the MIDI data 1 is performed.
The extracted key code data is input to the first inference means 3 and the second inference means 4. The first inference means 3 and the second inference means 4 infer with which finger each music of the input key code data should be keyed on based on different fuzzy rules and membership functions. The inference result is input to the validity inference means 5, and it is determined whether the inference result is appropriate in light of input conditions or the like. If the inference result is determined to be valid, the result is input to the data generation means 6 and the key-on intensity data and the like of the MIDI data 1 are synthesized to create new right-hand part and left-hand part-specific MIDI data 7 (right-hand data, left-hand data).
Is reconstructed and stored.

FIGS. 2 and 3 show fuzzy rules and membership functions used in the first inference means 3. FIG. In the first inference means 3, the right hand and the left hand are determined on the basis of the degree of left and right of the keyboard and whether it is right or left of the center (C4). Therefore, four sets of rules 10 to 13 are used as fuzzy rules.

FIGS. 4 and 5 show fuzzy rules and membership functions used in the second inference means 4. FIG. In this inference, the right hand and the left hand are determined based on an algorithm that the key in the vicinity of the key being turned on is often keyed on by the finger of the same hand. For this purpose, four sets of rules 20 to 34 are used.

The validity inference means 5 also performs fuzzy inference using the inference results up to that time (the contents of the left-hand register and the right-hand register) and the current inference result as input conditions. The fuzzy rules and membership functions in this case are as follows: "If the number of sounds turned on with the right hand is less than 4, KC can be turned on with the right hand.""The number of sounds turned on with the left hand is 4 If it is less than, the KC can be turned on with the left hand. "" If the spread between the sound turned on with the right hand and KC is less than one octave, it is easy to turn on with the right hand. "" Sound turned on with the right hand If the spread of KC is wider than one octave, it is difficult to turn on with the right hand. "If the spread of the sound turned on with the left hand and KC is narrower than one octave, it is easy to turn on with the left hand." If the spread between the sound and the KC is wider than one octave, it is difficult to turn on with the left hand. "

Here, KC is a register in which a key code of a sound taken in at the time of inference execution is stored, and indicates the contents thereof.

The above fuzzy rules and membership functions have been simplified for ease of understanding the description of this embodiment. In order to actually distribute each music to the left and right fingers, it is necessary to make inferences about each of the left and right fingers.

FIG. 6 is a block diagram of a control unit of the right / left hand sorting device. This device is a device that receives a micro floppy disk storing piano performance data by the MIDI method and detects fingering of the piano by a fuzzy inference operation by the CPU. This is a stand-alone device that is used separately from automatic performance pianos and electronic musical instruments. For this reason, the device configuration is simplified, and CPU10, ROM12, RAM1
3, a micro-floppy disk drive 14, a display 15, and a hard disk device 16 are connected via a bus 11. A basic operation program and the like are stored in the ROM 12, and a storage area shown in FIG. 7 is set in the RAM 13.

FIG. 7 shows a partial configuration diagram of the RAM 13. Figure (A)
Indicates a storage area for data loaded from the hard disk device 16 at startup. M1 is a control program storage area that defines the operation of the apparatus, and M2 and M3 are a fuzzy rule storage area and a membership function storage area for executing fuzzy inference by the CPU 10. M4 is a fitness coefficient storage area for adjusting the degree to which each fuzzy rule contributes to the inference result, and other left hand threshold Th1, right hand threshold Th
2 is loaded.

FIG. 2B shows a register set in the RAM 13. The RAM 13 has a left-hand register M11, a right-hand register M12, a fuzzy rule pointer i, a consequent part function value register V, a consequent part operation value register M, a conclusion maximum value register MAX, and a center of gravity register G.
In addition, a fitness coefficient correction value register M14 and a membership function correction value register M15 are set. Left hand register M1
1. The right hand register M12 stores the key code of the key that is inferred to be turned on by the left hand and right hand at each timing. The fuzzy rule pointer i designates the number of a fuzzy rule calculated by the CPU 10. This pointer is necessary for the CPU 10 to sequentially process a plurality of fuzzy rules. The antecedent part function value register V and the consequent part operation value register M store the value calculated as the operation value of the fuzzy rule. The antecedent part function value register V is a scalar quantity, but the contents of the consequent part operation value register M are two-dimensionally spread quantities. Conclusion The maximum value register MAX stores the logical sum of the operation values of a plurality of fuzzy rules. The center of gravity of the maximum value register MAX is stored in the center of gravity register G. Also,
The contents of the correction, which are corrected so as to make the degree of conformity and the membership function of the consequent part more appropriate with the fuzzy inference, are stored in the degree-of-fit correction value register M14 and the membership function correction value register M15.

FIG. 8 is a flowchart showing the operation of the CPU 10. This operation starts when the micro floppy disk is inserted into the drive device 14. First, an initialization operation such as resetting a register is performed. At this time, all MIDI data is read from the micro floppy disk into the memory (n1). For n2, this MIDI
Data is fetched, and the fetched key code is stored in the key code register KC (n2 '). Whether this data is a key-off event or a key-on event is determined by n3 and n4. N for key-off event
Proceed to 3 → n5, and if it is a key-on event, proceed to n4 → n10.

In n5, the key code of the key that was turned off is stored in the left-hand register or the right-hand register (whether the key is turned on with the right hand or the key is turned on with the left hand)
Is determined, and if the key code is stored in the left hand register, the MIDI data indicating that the key of this key code is turned off is set to channel 1 (n6), and the data of the left hand register corresponding to this key is cleared ( n7). Then return to n2. On the other hand, if the key code of the key-off event is stored in the right-hand register, MIDI data indicating that the key of this key code has been turned off is set to channel 2 (n
8). After that, the data of the corresponding right-hand register is cleared (n9), and the process returns to n2.

When the operation proceeds to n10 due to a key-on event, first, a fuzzy rule pointer i is set to 0, and fuzzy inference is executed in the following procedure.

The contents of the key code register KC (hereinafter "key code KC"
: The same applies to other registers). Antecedent members of rule (i) when fuzzy rules 10 to 13 and 20 to 23 are rules (0) to (7) when input values are used as input values. The function value of the ship function is obtained and stored in the function value register V (n11:
(See FIG. 9 (A)).

The function value V is multiplied by the fitness factor (i) to obtain a new V (n12: see FIG. 9B).

Next, a so-called truncation operation of the membership function of the consequent part of the rule (i) is performed with the function value V and set in the operation value register M (n13).

The union of the operation value M and the contents of the maximum value MAX is taken and this set is set in the maximum value register MAX (n14: see FIG. 9 (C)).

This operation is repeatedly executed until i becomes 0 → 3 (n15, n16).

When the above operation is completed for the rules (0) to (3), the center of gravity G of the maximum value MAX is calculated (n17). This G
If G is smaller than the left hand threshold Th1, it is determined that the tone is a musical tone to be keyed on with the left finger, and the process proceeds from n18 to n30. . It is a validity inference means together with the operation below n30 and the operation below n36. G is left hand threshold Th
If it is between 1 and the right hand threshold value Th2, it is determined that the right hand / left finger cannot be determined by this inference, and the operation proceeds to n20 or less as the second inference means.

In n20, i is further advanced. In n20 to n25, the same fuzzy inference operation as in n11 to n16 is executed for rules 4 to 7. Calculate the center of gravity G of the maximum value MAX in n26,
If the center of gravity is larger than 0, the process proceeds to n36 as a tone to be played with the right hand, and if G is smaller than 0, the process proceeds to n30 as a tone to be played with the left hand.

Below n30 is the validity inference operation. Steps n30 to n33 are operations performed when it is determined that the musical sound should be played with the left hand. First, it is determined whether the number of key codes stored in the left-hand register is four or less, and whether the key code KC and the key code stored in the left-hand register are all within one octave (n30). , n31).
If both conditions are satisfied, the key code KC and the accompanying MIDI data (key-on intensity data, etc.) are set in the channel 1 track (n32), and the key code KC is set.
Set C in the left hand register (n33). Next, the antecedent membership function of the second inference means is corrected based on KC (n34). This correction is performed so that the function value becomes maximum when the input value is KC. That all data has been edited
The operation of n2 or less is repeatedly executed until the judgment is made in n35.

On the other hand, n36 to n39 are operations when it is determined that the musical tone should be hit with the right hand. First, it is determined whether the number of key codes stored in the right hand register is 4 or less, and whether the key code KC and the key code stored in the right hand register are all within one octave (n36). , n37). If both conditions are satisfied, the key code KC and its associated MIDI data are set on channel 2 (n38), and the key code KC is set on the right hand register (n39). Then, proceed to n34.

If n31, n32 and n37, n38 determine that this inference is not valid (if the number of key codes stored in the left-hand (right-hand) register exceeds four, or tap with the left-hand (right-hand) If the key code to be spread is beyond the range of one octave, the process proceeds to n40. In n40, the fitness coefficient is corrected so that the inference result of the fuzzy inference becomes valid, and the first inference means (n10 or less) ) And execute again.

In the above embodiment, the fuzzy rules and the membership functions have been simplified for easy explanation and understanding. However, in order to infer fingering of the piano (on the keyboard) and actually distribute the fingering to the left and right, a continuous musical tone is used. Reasoning by connections (eg, “stretching from middle finger to thumb to play the rising scale of the right hand”), functional limitations of the finger (eg, “It is difficult to play keys close by legato while trilling .)) And techniques (for example, "If you turn on the right beating sound with the right little finger without erasing the sound of the little finger of the right hand, replace the finger (change the finger while keeping the key that is on)." It is necessary to construct a complicated fuzzy rule in consideration of the above) and make more inferences.

(G) Effects of the Invention As described above, according to the first aspect of the present invention, the input performance data is converted by the first inference means and the second inference means into right hand data played by the right hand and left hand data played by the left hand. And separated. As a result, the performance data in which the right hand data and the left hand data are mixed can be separated and reconstructed into the right hand data track and the left hand data track. Also, in order to separate performance data using fuzzy inference,
Does not require complicated algorithms for separation.

According to the second aspect of the present invention, the performance of the separation operation can be made more efficient by not using the second inference means for the performance data that can be separated by the first inference means. For those that cannot be separated by inference means,
By separating the right-hand data and the left-hand data using the inference means of the above, the accuracy of the separation can be ensured.

Furthermore, according to the third aspect of the invention, the validity of the separation result is determined by the validity determination means, so that incomplete separation by fuzzy inference can be compensated for and the separation accuracy can be increased.

[Brief description of the drawings]

FIG. 1 is a functional block diagram of a right / left hand sorting device for a piano according to an embodiment of the present invention. FIGS. 2 and 3 are fuzzy rules and membership of first inference executed by the left / right hand sorting device. FIGS. 4 and 5 are diagrams showing functions, and FIGS. 4 and 5 are diagrams showing fuzzy rules and membership functions of the second inference. 6 is a block diagram of a control unit of the right / left hand sorting device, FIG. 7 is a diagram showing a memory configuration of the control unit, FIG. 8 is a flowchart showing an operation of the control unit, and FIG. )
FIGS. 4C to 4C are diagrams for explaining a fuzzy inference procedure of the control unit. 3 ... first inference means, 4 ... second inference means, 5 ... validity inference means.

Claims (3)

(57) [Claims] An input means for inputting performance data, and the performance data input from the input means are separated by fuzzy inference into right hand data which is performance data to be played with the right hand and left hand data which is performance data to be played with the left hand. A first inference means for inferring whether the input performance data is right-hand data or left-hand data based on the pitch thereof; Second inference means for inferring whether the performance data is right-hand data or left-hand data based on the relationship with performance data before and after the right-hand data. 2. The performance data separating apparatus according to claim 1, wherein said second inference means operates when the performance data cannot be separated by said first inference means. 3. An apparatus according to claim 1, further comprising a validity judging means for judging a validity of the performance data separation result by said separating means, and changing a fuzzy inference condition of said separating means according to the judgment result of said validity judging means. The performance data separation device according to claim 1, wherein: