简体中文
English
Website review
This review may include AI-generated content and is not individually verified. Domain, certificate and other facts come from the observed data on this page.
What is NSynth Super?
NSynth Super is an experimental physical instrument from Google's Magenta research project, made in collaboration with Google Creative Lab, that lets musicians play sounds generated by the NSynth machine-learning algorithm. It is an open-source prototype rather than a finished commercial product: the page describes an "experience prototype" shared with a small group of musicians to see how they would use it creatively.
The key idea is that NSynth does not blend or mix existing recordings. It uses a deep neural network to learn the acoustic characteristics of sounds and then synthesizes a genuinely new sound from them — the site's example is something that is part flute and part sitar at once.
How it works in practice
- Four source sounds per dial. Each dial is assigned four source sounds, and the musician chooses which sounds to explore between.
- Touchscreen navigation. Dragging a finger across the touchscreen moves through the generated sounds that carry the acoustic qualities of those four sources.
- A large precomputed library. For this experiment, 16 original sounds across 15 pitches were recorded in a studio and fed into NSynth; the resulting 100,000-plus sounds were precomputed and loaded into the prototype.
Who it suits
It is aimed at musicians and sound designers curious about machine-learning timbres, and at developers or researchers interested in open-source musical interfaces. Because it is a prototype, expect an exploratory tool for finding unusual sounds rather than a gig-ready instrument with a polished workflow.
As a next step, listen to the demo film on the project page and note which source-sound pairings interest you — that will tell you quickly whether the four-source, touchscreen model fits how you like to shape sound. For background on the algorithm itself, see Magenta.
How does NSynth Super let me combine four source sounds into a new playable sound?
NSynth Super gives you a touchscreen surface where four source sounds sit at the corners. Dragging your finger across that surface moves you between them, and the instrument plays a newly synthesized sound shaped by the acoustic qualities of all four. The key distinction is that the sounds aren't blended or layered like a conventional sampler or crossfader: the NSynth algorithm generates a new sound that borrows characteristics from the sources, so a position between them can sound part flute and part sitar rather than like two recordings played at once.
What the controls do
- Four dials each hold one of your source sounds, so you choose which four sounds define the space you explore.
- The touchscreen is the performance surface: your finger's position selects where in that four-sound space the current note sits.
- The pitch input (the keyboard-style interface) sets the note you're playing, so the same touchscreen position produces that new sound at different pitches.
The prototype's sounds were prepared in advance: 16 studio-recorded source sounds across 15 pitches were run through NSynth, producing over 100,000 generated sounds loaded into the instrument. That means exploration is fast and immediate, but you're choosing from a fixed prepared set rather than feeding in your own samples live.
A concrete way to try it
Assign four deliberately contrasting sources — say a bowed string, a breathy flute, a metallic percussion hit and a low drone — then play a slow melody while sweeping from the string corner toward the flute corner. Listen for where the attack softens and the sustain changes character; those transition zones are where the synthesized middle ground is most audible and where the instrument differs most from a blend.
Trade-offs to weigh
The appeal is genuinely new timbres and a tactile, exploratory playing style that rewards moving around the surface rather than hitting fixed presets. The limits follow from the same design: you work within the prepared sound set, the four-source frame constrains how far a single gesture can travel, and results depend on picking sources with distinct acoustic profiles — four similar sounds will produce a narrow, less interesting space. For a fuller picture of the algorithm behind it, see Magenta and Google Experiments.
What is the difference between how NSynth generates sound and how traditional synthesizers blend sounds?
NSynth generates a new sound rather than mixing existing ones. According to the project's own description, the algorithm uses a deep neural network to learn the characteristics of sounds and then synthesize an entirely new sound from their acoustic qualities. The page's example is a sound that is "part flute and part sitar all at once" — not a flute layered or crossfaded with a sitar, but a single new timbre built from both.
A traditional synthesizer, by contrast, typically combines or blends its sources: you mix oscillator waveforms, layer samples, or crossfade between patches, and you still hear the contributing parts. NSynth's output is described as a synthesis from learned characteristics, so the source sounds inform the result without being audible as separate layers.
What this means in practice
- If you want recognizable layering — a pad plus a choir, with control over each level — a conventional synth or sampler is the direct tool.
- If you want hybrid timbres — an instrument that sits between two or four sources and doesn't quite exist — NSynth's approach is the point. On NSynth Super, four source sounds are assigned to a dial, and you move across the touchscreen to explore the in-between sounds.
- If you want reproducibility — traditional synthesis is deterministic and easy to recall; a learned model's output depends on the trained model and the source sounds fed in.
The trade-off is control versus novelty. Blending gives you predictable, tweakable components; NSynth gives you sounds you couldn't have dialed in by hand, at the cost of the fine-grained control a conventional synth offers over each layer.
A useful next step: pick two sounds you'd normally layer, then ask whether you want to hear both (blend) or hear something that is neither (synthesize). For the latter, the project page at NSynth Super explains how the source sounds and touchscreen navigation work, and the broader research is documented under Magenta.
Can I build my own NSynth Super, and what open-source materials are available?
Yes—NSynth Super was released as an open-source experimental instrument, so building your own is the intended path. Google's project page presents it as an open-source experimental instrument and notes that the pictured unit was an experience prototype shared with a small group of musicians, not a retail product. The build materials (hardware design files, software, and instructions) are published through the project's open-source repositories rather than sold as a kit.
What the project provides
- Hardware design: The physical interface is documented as open hardware, covering the enclosure, control layout (four dials plus a touchscreen), and electronics. You source and fabricate parts yourself.
- Software: The code that runs the instrument and connects it to precomputed NSynth sounds is open source, so you can inspect and modify it.
- Sound content: For the prototype, 16 source sounds across 15 pitches were recorded and run through NSynth to precompute over 100,000 sounds, which were then loaded onto the device. The instrument works by assigning four source sounds to each dial and letting you drag across the touchscreen to explore sounds that combine the acoustic qualities of those four sources—synthesis of new sounds, not simple blending.
Practical considerations before you start
- This is a build project, not a purchase. Expect to order components, fabricate or 3D-print an enclosure, and flash software. Skills in electronics assembly and basic Linux/command-line work help considerably.
- Sound generation is precomputed. The prototype plays back a precomputed library rather than running NSynth live, so you'll be working within the supplied sound set unless you generate your own.
- Documentation ages. Open-source hardware projects often depend on specific parts and software versions; check that components are still available and that the code runs on a current setup before committing to a bill of materials.
A sensible first step
Read the build documentation and repository on the project site first, then price the bill of materials and confirm part availability. If you mainly want to play with NSynth sounds rather than build hardware, the underlying algorithm is separately available through Google's Magenta project (Magenta) and the original NSynth release, which may be a lower-effort route to experimenting with the sound model before taking on the full instrument build.
How was the NSynth Super experience prototype tested with musicians, and what did they learn?
The prototype was shared with a small community of musicians so the team could observe how they might use it in their creative process. According to the project page, the goal was to understand real creative use rather than to run a formal benchmark: musicians were given an open-source experimental instrument that plays more than 100,000 precomputed NSynth sounds, built from 16 source sounds recorded across 15 pitches, and each dial was assigned four source sounds to explore between on the touchscreen. The page does not report specific findings, quotes or measured results from that testing, so any claim about what musicians concluded would go beyond what is documented.
What the setup implies for testing
The design points to a particular kind of evaluation. Because the sounds were precomputed and loaded into the prototype, the sessions could focus on playability and discovery—how quickly a musician finds a usable sound, whether the four-source dial layout makes sense in performance, and whether the results feel like new instruments rather than blends. The small, invited group suggests qualitative feedback and iterative refinement, not statistical validation.
A practical way to judge it
If you are considering building or adapting the open-source instrument, treat the prototype as a starting point for your own trial. Pick four source sounds with clear contrasts (for example, a bowed string, a flute, a plucked instrument and a voice), then test whether you can reach a playable sound within a minute and whether it holds up in a short improvisation. That tells you more about fit for your workflow than the prototype's own testing can.
For context on the underlying algorithm and related tools, see Magenta and Google Experiments.
What kinds of source sounds and pitches were used to create the 100,000+ precomputed sounds?
The precomputed library was built from 16 original source sounds, each recorded across 15 pitches in a studio. Those recordings were fed into the NSynth algorithm ahead of time, producing more than 100,000 new sounds that were then loaded onto the NSynth Super prototype. The page describes the sources only in those terms—it does not name the individual instruments or list the specific pitch range.
What this means in practice
- Diversity came from the 16 sources, not from live input. A player isn't feeding new audio into the model; they're navigating a fixed map of sounds derived from that original set.
- The 15 pitches per source give the library its pitch coverage. That's why the output count is so large: 16 sources × 15 pitches, multiplied by the many interpolated combinations the algorithm generates between source sounds.
- Each dial held 4 source sounds. Dragging across the touchscreen moves between new sounds that carry the acoustic qualities of those four sources—so the character of a given corner of the interface depends entirely on which four recordings were assigned to it.
A useful way to think about it: if you're evaluating whether such an instrument suits your work, the question isn't "how many sounds are there" but "which 16 timbres are in the set, and do they sit close enough together to blend musically?" A flute-to-sitar blend is the page's own example of the kind of hybrid it aims for.
For background on the underlying algorithm and related tools, see Magenta and Google Experiments.
google.com
User reviews (0)