Some fold over at Google's Magenta project has unveiled a hardware interface for an algorithm-based synthesizer that uses a deep neural network to generate completely new sounds.
It works by dragging a finger around the NSynth Super's colourful touchscreen to explore the unique sounds offered up by the machine learning algorithm. The NSynth Super interface offers musicians a way to go hands-on with the Magenta project's NSynth – or Neural Synthesizer – algorithm, which uses a neural network to learn sound characteristics of different sources and generate completely new sounds from what it's learned. The project stresses that these new creations are not merely sonic blends, but it is entirely new sounds that would be difficult to produce using hand-tuned synths.
The NSynth algorithm identifies and extracts 16 features from each sound source input, a sound's core attributes, or what makes a sound unique. "These features are then interpolated linearly to create new embeddings (mathematical representations of each sound)," explained the project. "These new embeddings are then decoded into new sounds, which have the acoustic qualities of both inputs." So, for example, a newly generated sound might have both flute- and sitar-like qualities, but the new sound will be neither one nor the other. It will be unique.
The NSynth Super is the first hardware interface that was developed in collaboration with the Google Creative Lab. The interface can be played with any MIDI source, such as a DAW, sequencer or keyboard, and the algorithm can generate over 100,000 new sounds by drawing from different sources.
In an experiment, 16 sources of sound, across 15 different pitches were recorded in a studio and then fed into the NSynth algorithm. Each dial on the Super interface was assigned four source sounds. The characteristics of selected sources sound could then be manipulated using the touchscreen display up top to make novel sounds.
The team created a few working prototypes of the NSynth Super and put them in the hands of musicians, but the hardware won't be released as a commercial product. You can see the potential for music creation in the video below.