JPH03219292A - Automatic player - Google Patents

Abstract

PURPOSE:To reproduce performance data in the best state by inferring the best driving data by a neural network according to performance data. CONSTITUTION:A ROM 13 is stored with an operation control program, the preset synapse weight of the neural network 15, etc. A register for storing various data generated during performance is set in a RAM 4, which is further stored with synapse weight that a user learns. Further, the neural network 15 infers a solenoid driving voltage and key-on timing according to performance data read out of a floppy disk. The neural network 15 is a hierarchical type neural network and its input layer is learnt so that at least touch data is received and the best driving data is outputted. Consequently, the best driving data of a driving means can be obtained according to the performance data, and the intention of the performance data can be reflected well.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to an automatic performance device that plays music by vibrating a sounding body based on performance data to generate musical tones.

(b) Prior Art Currently, automatic performance devices such as automatic performance pianos are in practical use. This type of automatic performance device has a sounding body that is the same as a natural instrument (
Because it automatically plays natural instruments, it can produce realistic and warm sounds that are not found in electronic instruments. For example, a player piano is equipped with a solenoid in the hammer action section of each key, and automatically plays a song by driving a predetermined hammer at a predetermined strength based on performance data stored on a floppy disk, etc. It is a device for playing music. The performance data includes keystroke strength data, and the player piano controls the solenoid drive voltage and timing (the time difference between the start of driving and the actual keystroke) based on this data (Figure 2). reference). Note that in the same figure, the start-up current is the first large current flowing regardless of the keystroke strength.

(C) Problems to be Solved by the Invention However, the keystroke strength and the solenoid drive current do not correspond linearly. Furthermore, since the sounding body is not electronic, the task of dealing with this problem is extremely troublesome.

That is, it is determined empirically through trial and error while comparing the musical tones produced by actual keystrokes and the musical tones produced by driving a solenoid based on the keystroke strength data. For this reason, it takes a lot of time to make a decision, and since the decided values are stored as a table, there is a drawback that the memory capacity is large. Furthermore, the timing has conventionally been determined empirically by engineers.

An object of the present invention is to provide an automatic performance device that solves the above problems by determining the intensity and timing using a neural network.

(Means for Solving Problem d1) This invention comprises a sounding body, a sounding means for vibrating the sounding body to generate musical tones, a driving means for driving the sounding means, and a storage means for storing performance data. and a neural network comprising an input layer that receives performance data read from the storage means, and an output layer that outputs drive data for the drive means when the input layer receives the performance data. shall be.

(e1 Function of the Invention In the automatic performance device of the present invention, when the neural network receives performance data, it outputs the drive data of the drive means (solenoid, etc.).The neural network outputs the optimal drive data when the performance data is input. This makes it possible to obtain optimal drive data for the drive means based on the performance data.
The intent of the performance data can be well reflected.

(f) Embodiment FIG. 1 is a block diagram of a control section of an automatic performance apparatus according to an embodiment of the present invention. This automatic performance device is a piano body 1
It is a device built into the floppy disk 16.
It has a function of driving the solenoid 3 to perform automatic performance based on the performance data stored in the piano 1, and also reading the performance data with the touch sensor 5 when the piano 1 is played by the player and storing it on the floppy disk 16. are doing. Therefore, solenoid 3. touch sensor 5
is provided for each key (88 keys).

The operation of this entire device is controlled by CPUI1. A ROM 13. is connected to the CPU II via a lotus 12.
RAM 14. Neural network 15 Floppy disk device 16. An operation panel 17 and an interface 18 are connected.

The solenoid drive device 4 and touch sensor 5 of the piano body 1 are connected to the interface 18 .

The solenoid drive device 3 drives the solenoid 3 based on commands from the CPUI. The solenoid 3 is attached to the hammer action part or the bottom of the key, and when driven, the hammer strikes the string in the same way as when a key is pressed.

The ROM 13 stores an operation control program, preset synapse weights of the neural network 5, and the like. RAM4 is set with registers for storing various data generated during performance, and
The synaptic weights learned by the user are stored.

Further, the neural network 15 infers the solenoid drive voltage and key-on timing (how many milliseconds before the solenoid is turned on) based on the performance data (touch data) read from the floppy disk. The neural network 5 is a hierarchical neural network, and the input layer is configured to accept at least touch data (keystroke strength data). The output solenoid drive current may be one value or may be a current waveform until the end of keystroke. By outputting a current waveform, a so-called touch feeling can be created by changing the waveform, and the tone can be varied. This neural network 5 may have any hardware configuration as long as it is capable of hierarchical inference. If the inference can be executed at high speed (on the order of a few seconds), a Neumann type microprocessor may be used.

The operation panel is provided with mode selection switches such as learning mode, record mode and play mode, and a selection switch for the synapse weight to be used.

The teacher data may be entered using a keyboard in the learning mode. The information may also be input using a MIDI interface.

In addition, by converting the actually played piano sound into data and inputting it as feedback, the solenoid drive current can be adjusted automatically.
It is also possible to learn the relationship between the musical tones (touch data) that are produced.

Although the above embodiment describes an automatic piano performance device, any musical instrument can be automatically played (one that makes a sound by striking, one that makes a sound by playing, one that makes a sound by blowing). , operation amount (drive data) corresponding to the instrument
All you have to do is infer.

(g) Effects of the Invention As described above, according to the automatic performance device of the present invention, the optimal drive data is inferred by the neural network based on the performance data, so that the recorded performance data can be optimally reproduced. I can do it.

Furthermore, by updating the learning of the neural network, it is also possible to respond to changes in the sounding body over time.

[Brief explanation of drawings]

FIG. 1 is a block diagram of a control section of an automatic performance device according to an embodiment of the present invention, and FIG. 2 is a diagram showing an example of a solenoid drive current of the automatic piano performance device. 3-Solenoid, 4-Solenoid drive device, 15-Neural network, 16-Frombie disk device.

Claims (1)

[Claims] (1) A sounding body, a sounding means for vibrating the sounding body to generate musical tones, a driving means for driving the sounding means, a storage means for storing performance data, and a memory means for storing performance data read from the storage means. An automatic performance device comprising: an input layer that receives performance data; and a neural network that outputs drive data for the driving means when the input layer receives the performance data.