A new concept synthesizer "Google NSynth Super" to make completely new sounds using machine learning

By using computer-based machine learning technology, it is possible to create a completely different sound from conventional synthesizers "NSynth SuperWe are advancing the development of. This synthesizer makes it possible for computers to understand the unique characteristics of various sounds by machine learning and to create completely new sounds that multiply the characteristics of "flute" and "snare drum" for example.

NSynth Super
https://nsynthsuper.withgoogle.com/

Making music using new sounds generated with machine learning
https://www.blog.google/topics/machine-learning/making-music-using-new-sounds-generated-machine-learning/

NSynth Super is a synthesizer about 20 cm square in size. It is a machine that is used as a "sound module" without a keyboard for playing, and data is stored to create sound inside. A display with a touch sensor is equipped on the top surface, and it is possible to operate with your fingers. The names of different musical instruments are written at the four corners of the display, and it is a new concept synthesizer that synthesizes the components of these four instruments with the balance specified by the finger, thereby creating a new tone.


In the following movies, demonstration performance introduces what kind of sound can actually be made.

Making music with NSynth Super - YouTube


I heard a strange tone like a guitar like "Pong, Pong" and a piano, and the tone changes variously. As shown by the caption "Make music with new sounds produced by machine learning", this sound is produced entirely by NSynth Super from the ground up with four instruments based on the sound with completely new technology.


Without a keyboard, NSynth Super generates sounds in real time based on the performance information input from the external keyboard via the MIDI cable.


In the movie we started rhythm machine here. Performing with the sound of NSynth Super will be performed according to this rhythm track.


NSynth Super makes machine learning of 16 kinds of sounds by computer, and data of each feature is converted into data. When playing, it is possible to select and synthesize four instruments from among them. However, rather than simply mixing the four sounds with the mixer, it is feeling like "decompose the sound into elements and join them at the molecular level to create a new sound" to create a completely new sound in real time And that.


Each tone color can be assigned with a large dial at the corners of the main body.


And use the fingers to adjust the balance of the four instruments.


It is also possible to change the generated sound like an analog synth.


The small dial under the display has four parameters "Attack", "Decay", "Sustain" and "Release" that characterize the tone "ADSRIt is to change.


It is also possible to operate "POS" which adjusts the part to be used with the sound waveform.


Unlike the change due to the filter of the analog synth, the sound generator which each maker is sending has also become a demonstration play that can listen to different sound changes.


NSynth Super is, Google research projects that produce music and art in machine learning "MagentaWhat is being developed in the middle of. Douglas Eck who leads the development team talks about its aim.

Making music using new sounds generated with machine learning - YouTube


A small instrument shop in the city


This person who is playing the synthesizer is Mr. Eck, who leads Magenta and is developing NSynth Super.


Magenta is a project to study how machine learning can be applied in the field of art.


Machine learning is defined as "research field that gives computer the ability to learn without explicit programming", father of machine learningArthur SamuelThe words are introduced.


Magenta creates "sound" using a neural network that reproduces the human brain.


The sound made here is not the existing sound, but "sound" at the level of "creating a new instrument".


The team first worked on a synthesizer algorithm using a neural network "NSynthIt was development of.


NSynth lets you learn the core elements of various sounds, so it is possible to combine multiple sounds.


The tone produced in this way can be what can not be realized in the ordinary world.


The sound that is the source of learning is very simple ... ....


In order to process and learn it, extremely advanced technology is required. However, for many musicians the technology is too high.


In order to bridge the gap, an attempt was made to drop the technology into an easy-to-use state.


Production of prototypes and internal design are done ......




It was born, "NSynth SuperThat's why.


Making sounds that have never been before will enhance the imagination of musicians.


For example, with the sound of a flute ......


I will multiply the sound of the snare drum. Because the flute's sound and the snare's sound are quite different in nature, it seems impossible to coalesce if you think normally.


The tone produced is like a little distorted electric guitar. The sound of a completely new musical instrument "Fujua" also has a sparkling feature of the snare drum while leaving the atmosphere like a flute.


It can be said that the history of music evolution is integrated with the history of the evolution of equipment. Various stringed instruments, percussion instruments, keyboard instruments have been invented, and in recent years electronic devices have created a completely new way of making sounds, and with that the new tones have been born one after another. And now, "NSynth Super" which produces sounds based on completely new concepts may be trying to make a page of the history of new instruments again.