Instagram story viewer> @nemus_project> Posts
NEMUS(@nemus_project)'s profile avatar HD Profile

NEMUS

@nemus_project

218
followers
160
following
• Numerical Restoration of Historical Musical Instruments
• ERC-funded project @erc_research
• physical modelling sound synthesis
POSTS STORIES REELS TAGGED
Always a pleasure hosting delegates from foreign institutions and browsing the magnificent Tagliavini collection at San Colombano. Our guests from the National Technical University of Athens were mesmerised by the beauty of the instruments hosted there, and had a go with our augmented keyboard 🎧 Many thanks to the museum staff for letting us into the museum before business hours. #harpsichord #earlymusic #computermusic #museumcollection taken in San Colombano - Collezione Tagliavini by @nemus_project
0
2 days ago
Download
A digital twin of Gadi’s own electroacoustic sound. Played live. From his own violin. For a recent UN performance, Gadi worked with Datamind Audio — his old Berklee classmate Ben Cantil, now doing his PhD at Edinburgh — to train an AI model on his own material. What the tool gave back was a model of Gadi’s electroacoustic output, which he could then drive, in performance, from his acoustic violin. Not a sampler. Not a preset. A neural rendering of a specific artist’s sonic fingerprint, controlled by that artist’s own gesture. It’s not quite there yet: about 300 milliseconds of latency, enough to break the timing on anything you’d actually want to play. But the shape of where this ends up is legible — physical modelling on one side of the room, learned models on the other, gesture in the middle joining them up. And, as Gadi points out, the interesting bit isn’t just the technology. Datamind is experimenting with compensation models that pay artists when their trained voice gets used. Whatever else neural audio ends up being, the “who owns the sound you learned from” question is at least being asked out loud. Seventh of seven. That’s the series. Full conversation on YouTube — link in bio. With Michele Ducceschi, Craig Webb (@physicalaudiouk) , Stefan Bilbao, @gadisassoon #NEMUS #physicalmodelling #AImusic #digitaltwin #computermusic by @nemus_project
0
5 days ago
Download
You can build a physical model of a 1547 harpsichord. The hard part is what someone sits at to play it. Gesture is what NESS mostly didn't have to think about — it was offline, a render pipeline, nobody was playing anything in real time. NEMUS lives on the opposite side of that line. A physical model that responds to touch has to be attached to something you touch, and if the model itself lives in the physics of the original — Trasuntino's action, its two-stop registers, the resistance of quill on string — then the interface has to give the model that same information: not just note-on and note-off, not just velocity, but continuous position, effort, release. MIDI can't. MPE gets closer. Neither is really enough. Which is why the NEMUS team keeps insisting on getting early-music players into the design conversation from the beginning. Not to test the finished thing. To tell you, halfway through, that the gesture doesn't feel like anything. "I want them to come in early during the design process, so they can really tell us whether the gesture or control we afford them feels accurate or not." Sixth of seven. Full conversation on YouTube — link in bio. With Michele Ducceschi, Craig Webb (@physicalaudiouk), Stefan Bilbao, @gadisassoon #NEMUS #physicalmodelling #MPE #musicalinstruments #computermusic by @nemus_project
0
10 days ago
Download
Ten years ago, doing this in real time on a laptop was science fiction. Physical modelling lives and dies on how much compute a non-linear system needs at 48,000 samples a second — and non-linearities are what stop synthesis from sounding predictable. For most of the NESS era the answer was: too much. Simulations ran overnight. Composers wrote for renders they wouldn't hear until the next morning. Then around 2018 Michele Ducceschi and Stefan Bilbao — with Craig Webb and others — worked out a family of methods for the exact class of non-linear systems that had been the worst offenders. Turned out that maths didn't need what everyone had assumed it needed. The knock-on effect is the Physical Audio plug-ins you can now buy — Modus, Preparation — running as real-time VSTs on ordinary machines. Things that were a GPU-cluster problem in the NESS lab are now a plug-in slot on your DAW. Same equations, radically less compute. Which, incidentally, is what made NEMUS possible in the first place. Fifth of seven. Full conversation on YouTube — link in bio. With Michele Ducceschi, Craig Webb (@physicalaudiouk) , Stefan Bilbao, @gadisassoon #NEMUS #physicalmodelling #realtimeaudio #VST #computermusic by @nemus_project
1
2 months ago
Download
NESS was a five-year offline project — massive simulations you'd kick off at six in the evening and pick up at six the next morning, aimed at composers who wanted to bend the physics of impossible instruments. NEMUS, the five-year project that followed, moves the same maths into real time and points it at something else: historical keyboard instruments too fragile to touch, in the Tagliavini Collection @genusbononiae at Museo San Colombano in Bologna. The centrepiece is a 1547 Trasuntino harpsichord — the oldest Italian two-by-eight-foot harpsichord in existence — which cannot be played. NEMUS built a physical model of it, running on a keyboard that sits in the museum's permanent display next to the silent original, and anyone can walk up and play. From "no realtime, ever" to "realtime is the whole point." From a server humming through the night to a museum instrument you touch with your hands. The scope shifted. The maths didn't. @erc_research funded both projects. Fourth of seven. Full conversation on YouTube — link in bio. With Michele Ducceschi, Craig Webb (@physicalaudiouk) , Stefan Bilbao, @gadisassoon #NEMUS #physicalmodelling #NESSproject #historicalinstruments by @nemus_project
0
2 months ago
Download
A room and a string are the same equation. That's not a metaphor. The wave equation on a string, extended into three dimensions, is the wave equation in a room — same maths, higher dimension. Which means the model NESS uses to describe a plucked string can also describe an entire hall, and any instrument struck inside that hall stops being synthesiser plus reverb and becomes one physical model with three parts to it — acoustic field, membrane, mallet. Synthesis and spatialisation collapse into the same thing. Most acoustic simulation you've heard of doesn't work like this. It uses ray tracing — borrowed from computer graphics — which pretends sound propagates in straight lines that bounce off walls until they die: the sound-design version of room acoustics. It's fine for stadiums and quietly wrong for anything smaller: waves diffract, bend around corners, and behave differently at different frequencies. Rays don't. The alternative — actually solving the wave equation over the whole 3D space — is honest and very expensive. In the NESS lab it meant a box of £3,000 Tesla GPUs and simulations Stefan would kick off at six in the evening and pick up at six the next morning. What comes back isn't a reverb tail. It's the entire acoustic field. You can put a microphone anywhere in the room after the fact, encode it to Ambisonics, whatever you want. And then the same equations describe the string that made the sound in the first place. Third of seven. Full conversation on YouTube — link in bio. With Michele Ducceschi, Craig Webb (@physicalaudiouk), Stefan Bilbao, @gadisassoon. by @nemus_project
0
2 months ago
Download
EP 02 · Learning through failure Gadi sits at a screen for hours, growing more excited by the minute as a GPU array grinds through his render. Something spectacular must be coming — nothing has ever taken this long. When the file finally opens, it contains one click. Then silence. Perfectly still DC offset, all the way to the end. The setup was this: Gadi had been poking at the edges of the NESS tools, seeing what fell out when he deliberately broke the physics. On that particular night, he'd stripped out the parameter that stops a finger, once it strikes the fingerboard, from bouncing back forever. He thought it was inertia. Stefan said restoring force. Turned out to be a loss. The simulated finger just kept moving, and moving, and moving. And here's the line that makes the whole story — "the longer it took, the more excited I got." That was Stefan's "Bell Labs" deal, working exactly as designed. He'd wanted NESS to feel like the 1960s on purpose: renders you had to sit through, no preview, no undo, real stakes on whether the eight hours you just spent produced anything at all. If you were paying attention, you learned. If you weren't, you didn't. Gadi calls this his humbling. It's also, in a strange way, the fastest education in physical modelling anyone in the room got. Second of seven. Full conversation on YouTube — link in bio. With Michele Ducceschi, Craig Webb(@physicalaudiouk) , Stefan Bilbao, @gadisassoon by @nemus_project
1
3 months ago
Download
Ask the internet and you’ll mostly get filter banks tuned to sound like a guitar — Karplus–Strong, some clever delay lines, a plausible pluck. Useful, expressive, culturally huge. But it starts from the sound and works backwards. The tradition NESS and NEMUS come from is stricter. As Stefan Bilbao puts, the shortcut is: think of computer-graphics physics engines, but for sound. You write down the equations that describe how a string, a bore, a plate actually vibrates — and the sound is whatever falls out of solving them. The gold standard isn’t a target waveform. It’s a set of equations. That distinction matters. It’s why the same techniques end up in engineering, computational fluid dynamics and geophysics — the same maths, different interpretation. @gadisassoon, coming in as a composer rather than a researcher, admits he only really understood the difference once he’d spent two years with the tools. Which is honest, and also the whole point of the series. With Michele Ducceschi, Craig Webb (@physicalaudiouk), Stefan Bilbao, and Gadi Sassoon. First of seven. Full conversation on YouTube — link in bio. by @nemus_project
1
3 months ago
Download
A DECADE OF PHYSICAL MODELLING RESEARCH Seven clips. One per topic. From this week. On 9 June 2024, in @gadisassoon 's Milan studio, we sat down to unpack a decade of physical modelling research. Almost two hours of conversation came out — too much for a single sitting. So we're serving it in courses. Seven clips, one per theme: 01 · Defining physical modelling 02 · Learning through failure 03 · Raytrace 04 · From NESS to NEMUS 05 · Computational advances 06 · Gesture 07 · Gadi's violin With Michele Ducceschi, Craig Webb (@physicalaudiouk), Stefan Bilbao, and Gadi Sassoon. The full conversation is already on YouTube — link in bio. by @nemus_project
2
3 months ago
Download
The 1547 Alessandro Trasuntino — the oldest surviving Italian two-register harpsichord — sits unplayable in Bologna’s Collezione Tagliavini. The collection’s management opted against restoration, because doing so would erase the harpsichord’s documentary evidence, including its unmodified soundboard. So the NEMUS project built a playable copy from measurements of the original, and sampled every note. 🔨 Acoustic copy hand-built by Roberto Livi (Pesaro) 🎙 Recorded by @simonecoen at @turangalilarecording Studio 🎵 Meantone tuning, A = 415 Hz 📀 Two versions — Ottava Corta (historical short-octave) & Esteso (fully chromatic) 🆓 Free · Kontakt 7+ Link in bio → nemusproject.eu by @nemus_project
2
3 months ago
Download
Sound alone won’t do it. You can record a 1547 harpsichord and stream the samples through any keyboard on the market. What comes out is the 1547. What you feel under your fingers is a modern plastic controller — and your ears know. The tactile McGurk effect: touch contaminates listening, and no sample library survives it. So the NEMUS digital copy isn’t a sample player. It’s a second keyboard, mechanically identical to the acoustic copy Roberto Livi built — same key weight, same jack rail, same escapement, same 1547 action under the fingers. The sound comes from optical sensing. Each jack carries a printed greyscale gradient; a QRE1113 reflectance sensor reads its motion — position, velocity, continuous, not just note-on/note-off. Nothing in the historical action is drilled, glued, or replaced. Pull the electronics out and the instrument is exactly what Livi built. The signal drives a sample library and a physical model in parallel. What you hear is the 1547. What you feel is the 1547. And the original stays untouched. The digital copy is installed at the Tagliavini Collection, Museo San Colombano, Bologna — one object in the room visitors are invited to play. One idea. Three bodies. This is the third. Full case study → nemusproject.eu/case-studies/trasuntino-restoration #NEMUS #Trasuntino #Harpsichord taken in San Colombano - Collezione Tagliavini by @nemus_project
1
3 months ago
Download
The 1547 stays silent. Its twin doesn't. Between 2024 and early 2026, master harpsichord builder Roberto Livi rebuilt the Trasuntino by hand in his Pesaro workshop — geometry, string plan, voicing, all followed from the original itself. But the original could not be opened. Its soundboard — the acoustic heart of any harpsichord — had to be mapped from behind: through a small hole in the case, with a magnetic gauge, point by point. What came back on paper contradicted centuries of craft instinct. The board is thicker than any modern builder would voluntarily leave it. Livi trusted the evidence. The copy was built to match, even where the hands wanted to plane it thinner. The finished copy stays at Livi's workshop, kept available to musicians and artists who want to hear — and play — a 1547 harpsichord today. Meet the acoustic copy. The digital one comes next. Full case study → nemusproject.eu/case-studies/trasuntino-restoration by @nemus_project
1
3 months ago
Download

Download All

Save the photos and videos currently loaded on this page.

0Photos
0Videos
Ready to download 0 / 0

Keep this tab open until your download is ready.
back to up