A workaround already exists.
The I-Corps presentation records whispering, typing, Notes, notebooks, and calendars as existing alternatives. A wearable has to earn its place alongside them.
MADE WITH
Arduino IDEMicrocontrollers3D printing
PadKey
Private input, close to the body.
A wearable input exploration shaped through customer discovery, sensor tests, and physical prototypes.
↗ Rev Ithaca + NSF I-CorpsSee the prototype ↗
- Role
- Founder · design & prototyping
- Timeline
- 2026 · project archive
- Category
- Hardware · Wearable input
01 · Overview
Voice input isn’t always private.
PadKey explores voice input for moments when speaking openly to a computer does not fit the room.
The actual prototype
The enclosure
and what is inside it.
A microphone, a small electronics assembly, and a printed form made the question tangible.
Watch the recorded demo ↘Voice interfaces assume that speaking aloud is convenient. In a shared room, a private prompt or a computer command can feel out of place. PadKey started with that gap between what an interface can do and what a person is comfortable doing around other people.
I worked through customer discovery, sensing experiments, physical forms, and a microphone-based demonstration. Rev Ithaca gave the project a hands-on prototyping setting; NSF I-Corps helped narrow the customer and the business question.
The question
What would make private input worth wearing?
The signal matters, but so do setup, deliberate activation, feedback, correction, and comfort.
My roleFounder · design · prototyping
ProgramsRev Ithaca + NSF I-Corps
WorkHardware + interaction
StageExperimental prototype
About the project record Original material & scope
The photographs and two videos are from the earlier portfolio. A dated sensor bench note, interview synthesis, and the I-Corps final presentation add context. New diagrams explain those records retrospectively. A captured sensor signal or phrase demonstration is not evidence of silent-speech recognition, everyday reliability, or manufacturing readiness.
02 · Discovery
When speaking aloud doesn’t fit.
The discovery record points toward a specific situation: people who use AI and computing tools but avoid saying sensitive input aloud around others.
The same difficulty is not the same market.
Weak or intermittent voice and AAC needs also appeared in discovery. The final presentation treats those as a separate clinical and buyer path, not one broad customer persona.
Agency needs a clear off state.
The recorded direction calls for deliberate activation and confirmation. Privacy is an interaction requirement, not simply a description of a small microphone.
NSF I-Corps · documented decision
A broader question led to a narrower direction.
The final presentation keeps private voice input for AI-heavy professionals in shared spaces. It parks generic productivity, clinical positioning as the same market, and perception control as parallel product directions.
03 · Approach
Work between the person and the prototype.
The work moved between what someone might need and what a physical prototype could actually capture.
Before the signal
Clarify the situation and the current workaround before choosing a sensor.
At the bench
Make the input legible enough to distinguish a useful response from noise or movement.
Around the body
Use physical forms to expose placement, size, handling, and comfort questions.
04 · Bench evidence
Two inputs. A signal on screen.
The 10 June 2026 bench note records a piezo disc and an electret microphone producing readable changes in Arduino Serial Monitor.
Bench note / 10 June 2026
Two physical inputs.
A visible response.
The useful proof point was acquisition: both the vibration path and acoustic path produced changes that could be inspected.
- Board
- Arduino Nano 33 BLE
- Inputs
- Piezo A3 · electret mic A1
- Mic sample window
- 50 ms
- Serial rate
- 115200 baud
- Not established
- Speech decoding or command accuracy
05 · Physical iterations
From wearable studies to a printed enclosure.
Early forms left room for different placements. Later prototypes brought the microphone path and electronics into a compact printed enclosure.
Physical progression →
From placement studies, to a focused electronics assembly, to an enclosure that could be held and demonstrated.
A render helped explore an appearance. The photographs show what was built.
The visualisation is kept distinct from the physical prototype record.
06 · Technical tradeoffs
Which signal should the design depend on?
Different sensing approaches change placement, setup, feedback, and the kinds of errors the rest of the experience has to handle.
| Path explored | Question it opened | What the record supports |
|---|---|---|
| EMG | Could muscle activity become an input path? | Listed in the original exploration narrative. No recognition benchmark is published here. |
| Piezo vibration | Could contact vibration produce an inspectable signal? | Dated bench observation and a candid experiment video. |
| Microphones | Could acoustic pickup support a practical demonstration? | Analog/digital comparisons, an enclosed assembly, and a recorded phrase demo. |
| Bone conduction | What would body contact and placement change? | A photographed early iteration. Comfort and performance remain test questions. |
The prototype direction
Use microphones for the working demonstration. Keep the other experiments as learning records rather than imply that every path reached the same level of proof.
07 · Prototype demonstrations
Watch the prototype.
These are the original recordings, preserved from the earlier portfolio.
What to watch for
The recording shows how the enclosed prototype was held during a short spoken test. It documents the demonstration; it is not a matched-environment accuracy test.
08 · Reflection
The prototype narrowed the next question.
Building made the uncertainties smaller and more concrete. Customer discovery helped decide which ones were worth pursuing.
Swipe to read all six takeaways →
A useful workaround is the real competitor.
01Customer discovery
A visible signal gives the next experiment a starting point.
02Bench evidence
Placement changes the interaction before software does.
03Physical design
A small device still needs deliberate activation and feedback.
04Interaction design
Different access needs deserve their own discovery path.
05Scope
A working demo earns another test, not a production claim.
06Validation
The next experiment →
Compare microphone behavior under matched phrases and environments, then observe setup, correction, false activation, comfort, and whether the device replaces an existing workaround.
Sources and project notes
Recovered project material
The desktop-style portfolio’s PadKey narrative, sixteen project images, the captioned demonstration, and the piezo experiment. Images are original photographs except the separately labelled AI appearance render.
Program and discovery records
Rev Ithaca hardware accelerator and NSF I-Corps, confirmed by the project owner. The I-Corps final presentation dated 25 August 2026 supplies the narrowing decision reproduced here. The customer-discovery compendium informs the themes; individual interview notes and personal details are not published.
Bench record
PadKey Sensor Bench Note, 10 June 2026: Arduino Nano 33 BLE; piezo A3; electret mic A1; 50 ms sample window; 115200 baud; readable changes in both inputs. This is acquisition evidence, not a speech-classification result.
New explanatory material
The homepage and header artwork are generated conceptual abstractions of silent speech: notation and waveform energy diminish into a flat zero line. They do not depict a new product. Original prototype photographs and videos remain separately labelled.
The process and signal-chain diagrams were drawn for this case study from the archived records. The page documents the hardware exploration; it does not claim the current companion software has been tested end to end in this portfolio pass.