Helping people regulate their voice without being corrected
Noise-canceling headphones removed the feedback people rely on to self-regulate. Mezzo replaces social correction with a private, continuous signal.
Replace with the redesigned Mezzo interface — dark mode default, calibration and PiP states.
Social correction to
private self-regulation
The problem was not volume. It was the feedback channel.
People in shared offices already wanted to be considerate. They had no private, real-time way to know whether they were succeeding.
A private feedback system for shared workspaces
The feedback loop broke when people could no longer hear themselves
Shared offices rely on an unspoken contract: collaborate without disrupting others. But dense seating, frequent desk calls, and noise-canceling headphones make that contract difficult to honor.
With both ears sealed, people cannot reliably hear their own voice. The only remaining signal is social correction: someone telling them they are too loud after disruption has already happened.
The problem was not that people did not care. The feedback loop was broken.
Product frame: Mezzo is not a monitoring tool. It is a private self-regulation aid for people already trying to be considerate in shared workspaces.
How do you give someone private feedback without making it punitive?
The challenge was not technical. It was behavioral: delivering awareness in a way that felt helpful, not judgmental, and private, not surveillant.
No way to know you are too loud until someone tells you
Social correction arrives too late to feel helpful
People in shared offices were already trying to self-regulate. They pulled one ear off, watched for reactions, moved away from others, or avoided calls entirely.
Social correction arrives late, feels embarrassing, and often causes the most considerate people to overcorrect or withdraw.
The people most affected were not necessarily the loudest. They were often the most considerate.
Person speaks
They cannot accurately hear their own level.
Volume carries
The room absorbs the disruption first.
Others react
Awareness comes from social cues.
Correction
Someone has to say something.
Withdrawal
The user feels embarrassed or overcorrects.
Correction comes too late to feel helpful.
This was not a volume problem. It was a feedback channel problem.
The product category shifted when the research made this clear: people did not need to be told they were too loud. They needed to know it themselves, privately, before anyone else had to intervene.
That changed what the product needed to optimize for: private awareness over public enforcement, continuous signal over moment-of-crisis alerts, and structural privacy over promised discretion.
Not a monitoring tool.
A private self-regulation aid.
The key design move was giving users a continuous reference point instead of a correction — awareness that arrived before anyone else had to say anything.
People did not need enforcement. They needed private, real-time awareness of where they stood — before the room reacted.
Each decision made the feedback useful without making it punitive, public, or surveillant
Design for self-regulation, not correction
Problem Social correction arrives after disruption and causes embarrassment, withdrawal, or overcorrection.
Decision Replace social correction with private, continuous feedback.
Tradeoff Avoided explicit “you are too loud” alerts, making the feedback less forceful.
Impact Users could recalibrate earlier without social friction.
Make privacy structural, not promised
Problem A microphone-based product could easily feel like surveillance.
Decision Use real-time audio measurement only: no accounts, no recording, no transcription, no stored history, and no AI interpretation.
Tradeoff The system could not personalize over time or provide historical insights.
Impact Trust was built into what the system could not do.
Design for patterns, not spikes
Problem A simple volume alert would punish normal behavior like laughing, emphasis, or excitement.
Decision Use temporal escalation: brief spikes are forgiven, sustained loudness triggers stronger feedback.
Tradeoff Some genuinely loud moments may go unflagged.
Impact The system avoids training users to fear or ignore the signal.
Prioritize private visibility over seamless integration
Problem Feedback visible during screen sharing would turn private awareness into public judgment.
Decision Use Picture-in-Picture for peripheral, private feedback instead of embedding directly in meeting tools.
Tradeoff PiP preserved privacy but reduced convenience and discoverability.
Impact The MVP validated the concept while revealing platform proximity as the next adoption constraint.
The product’s trust model came from containment
The interface never needed raw audio to provide useful feedback
The audio layer measures real-time signal behavior and emits abstract states. The UI renders only those states.
This kept trust structural: no accounts, no transcripts, no recordings, no stored history, and no AI interpretation.
Trust was not communicated through interface copy. It was designed into what the system was structurally incapable of doing.
Microphone
Audio input stays local.
Audio engine
Measures real-time level.
Abstract state
Converts signal to feedback.
UI feedback
Displays only state.
Raw audio stays contained. Nothing is recorded or stored.
Each decision reinforced the same behavioral goal
Private signal
Enables self-regulation.
Structural privacy
Builds trust through limits.
Temporal logic
Prevents overcorrection.
Separated architecture
Enforces containment.
PiP delivery
Supports real-time use.
Reactive correction becomes continuous awareness.
A lightweight tool for private peripheral awareness
Awareness, not enforcement
Mezzo is a web app that measures real-time voice levels, calibrates around a comfortable conversation range, and displays feedback through Picture-in-Picture during calls.
The system avoids scores, warnings, recordings, and transcripts. It gives users a continuous reference point instead of a public correction.
Calibration. Volume is framed as a social range, not a technical decibel target.
PiP feedback. Peripheral feedback stays available during calls without becoming public in screen share.
The concept worked. The next constraint was proximity.
The MVP was used in real meetings. One participant used Mezzo in a room known to leak sound and reported feeling safe when the signal stayed within range.
The concept was understandable and desirable. But the next product risk became platform proximity: participants wanted the tool closer to Zoom, Teams, taskbar, or browser workflows.
Concept validated
Users understood the system without explanation and found the feedback non-judgmental.
Next constraint named
Platform proximity emerged as the adoption barrier. Participants wanted Mezzo closer to their existing meeting tools.
Participants understood the system without explanation.
Participants described the feedback as neutral rather than punitive.
The privacy model did not trigger concern in testing.
Participants wanted the tool closer to Zoom, Teams, or taskbar workflows.
The MVP validated the concept while surfacing the next risk. The core question became whether Mezzo could move closer to where people already work without losing the structural privacy that made it trustworthy.
Designing for how feedback feels, not just whether it is accurate
A feedback system can be technically correct and still fail if it feels punitive, public, or socially misaligned.
Mezzo worked because it delivered information users already wanted, privately, before anyone else had to say it. The design challenge was not accuracy. It was timing, tone, and structural trust.
That changed how I think about feedback systems in sensitive contexts. The system’s constraints — what it cannot do — matter as much as its capabilities.
Trust is designed into what the system cannot do.
In trust-sensitive systems, structural constraints communicate more reliably than interface copy. Users do not trust promises. They trust what the system is incapable of doing by design.
The next risk is proximity, not desirability
The core question: can Mezzo become ambient infrastructure without becoming visible, intrusive, or surveillant?