The Synthesizer That Talks Back: How an Open-Source Tool Is Making Hardware Synths Accessible

On the floor of the Brooklyn Synth Expo today, a familiar ritual will play out: a musician reaches for a knob on a synthesizer, turns it, and hears the instrument answer back. Not with a different tone, but with words. "Cutoff frequency, 62 percent." For blind and low-vision musicians, that spoken label is the difference between working by memory and working with confidence.

The tool behind it is called MIDItoSpeech, and it comes out of NYU's Ability Project. This week it crossed a threshold that matters: eight of the most popular hardware synthesizers on the market now have official device profiles, created in collaboration with blind musicians who use the gear. The profiles were built by Sequential, Oberheim, and Novation, with NYU professor Luke Dubois and UK-based blind musician and producer Jason Dasent leading the effort. Dasent and Dubois are demonstrating the work at the expo.

The problem with knobs you cannot see

Hardware synthesizers are physical instruments. They are built around knobs, switches, and tiny LED screens, and even the knob-per-function designs that players love hide a problem: from touch alone, one knob feels much like another, and many modern synths bury critical settings in menus or make a single control do different things depending on the mode. A sighted player reads the panel and glances at the screen. A blind player works from memory, from trial and error, or from a separate computer editor that adds a second interface to learn.

Software was supposed to fix this. Desktop music applications have had accessibility APIs for years, and screen readers like VoiceOver on the Mac can navigate them. But hardware synths transmit nothing their own interface can announce. The intelligence was already traveling out of the instrument, though, in a form musicians have taken for granted for decades: MIDI.

An old protocol becomes a screen reader

Every time you turn a knob on most synthesizers, the instrument broadcasts what happened as a MIDI message. Continuous Controller (CC) and Non-Registered Parameter Number (NRPN) messages carry the control number and its new value. They are how synths talk to recording software. They just never talked to the player.

MIDItoSpeech listens to those messages and translates them. A community-built JSON device profile maps each control number on a specific synth to a speakable label: which knob it is, what value it reports, what unit it means. The result is a free web application, built on the Web MIDI API, that narrates every knob turn, switch flip, and patch change. NYU also ships it as a Max/MSP patcher with downloadable executables for Mac and Windows, and on a Mac the speech can flow through VoiceOver.

Writing a profile is detailed work. Each control needs a label, a value range, and rules for how the number should be read: as a raw value, a percentage, a frequency in hertz, a bipolar range, or an enumerated list of named settings. The project is open source under the MIT license, and the profile repository is community-driven, so musicians can build profiles for legacy synths the manufacturers never got around to.

The intelligence was already traveling out of the instrument, in a form musicians took for granted for decades: MIDI. It just never talked to the player.

Designed with, not for

The origin story matters as much as the code. Sequential and Oberheim CEO David Gibbons, writing on the Oberheim site this week, describes meeting Dasent in August 2025. Dasent had bought an Oberheim TEO-5 and was running into accessibility walls, and he made a case for the everyday friction blind musicians hit even on the friendliest hardware layouts. Weeks later, an unsolicited email arrived from Dubois, who had already built an early version for the Sequential Prophet-6. Gibbons introduced the two, and Dasent began visiting NYU to refine the tool.

That sequence is the whole point. The feature was not designed in a lab and delivered to blind musicians as a favor. A working musician who is blind identified the gap, a university researcher had a prototype aimed at it, and the companies that make the instruments brought the missing piece: accurate, complete profiles for their own products, written by the support and engineering people who know the control maps best. Gus Callahan, who leads support at Sequential, and Julio Ortiz, its lead software developer, built the Sequential and Oberheim profiles. Paul Mansell and Dan Mitchell of the Novation team in the UK did the same for Novation.

Eight profiles, and an open door

The launch lineup covers eight instruments: the Sequential Fourm, Take 5, Prophet-6, and Prophet Rev2; the Oberheim TEO-5; and the Novation BassStation II, Summit, and Peak. That list includes some of the best-selling analog and hybrid synths of the last decade, which means the tool is usable by a real working population from day one rather than living as a demo for one flagship model.

Gibbons makes a broader claim in the announcement: most synthesizers on the market could follow this approach almost overnight. The reason is that the hard part is not engineering. The MIDI messages already exist. The web app already exists. What remains is documentation, and a manufacturer that knows its own control map can write a profile with far less effort than building accessibility from scratch. His call to action runs both directions: players should write to the makers of the synths they own and say they want profiles, and manufacturers should create them, because the barrier is low.

There is a second-order benefit that the NYU team has thought about from the start. Because profiles are just labeled JSON files, they can be translated. The same device file that reads out "cutoff frequency" in English could read it in Spanish, Mandarin, or any other language, turning accessibility work into localization work at the same time. Community contributors can also keep older and discontinued instruments alive, which matters in a scene where a 1980s synth is often a working instrument, not a museum piece.

The deeper lesson here is a design one. Accessibility is usually framed as something added to a product: a mode, an option, a special edition. MIDItoSpeech went the other way. It took a protocol every synth already speaks, added a layer of description, and made the instrument announce itself. Nothing about the hardware changed. Everything about who can use it did.