Advanced: the voice
Your voice, tuned for how they hear now.
Aging changes hearing before anyone notices. The crisp edges of words fade first, phone lines strip away even more, and by evening a tired mind works twice as hard for half the message. So they ask again. And again.
Proxi speaks in your voice, rebuilt for their ears by a purpose-built speech engine. Not louder. Clearer.
Four registers
And when the moment calls for it, your voice knows how to meet it.
One voice cannot do every job. A voice built only to soothe cannot be heard when something is wrong, and a voice built to carry cannot sit quietly with someone at two in the morning.
Same voice. Four ways of using it, the way you already do.
The psychoacoustics
This is not a volume knob and a warmth filter.
It is forty years of hearing science, aimed at one voice. Yours.
It learns how you would speak to someone hard of hearing.
A neural network trained only on your recordings reshapes your articulation the way you naturally would if you knew they were struggling to catch it. Not slower and louder. Fuller consonants, wider contrasts, a little more room between phrases.
The technique descends from clear-speech research begun at MIT in the 1980s, which found that talkers who deliberately speak to be understood produce large intelligibility gains that come mostly from articulation rather than from slowing down.
It knows every consonant before you hear it.
Because Proxi creates the speech rather than processing a recording, it knows what sound is coming next. It can strengthen exactly the ones aging ears lose first. The s, the t, the edges of words.
Sound by sound, only where it is needed, and only within the range your own voice actually occupies.
It protects the part of speech the brain actually decodes.
Under the words there is a slow rhythm, the rise and fall of energy across syllables. That pattern is what the brain leans on to pull words out of sound, and it is the same signal that makes cochlear implants work at all.
It is also the first thing an ordinary audio pipeline damages. Ours protects it at every stage, the way premium hearing aids devote dedicated processors to guarding it.
It is calibrated to them, not to an average.
At setup we tune to their hearing using your actual voice, on the actual device, at their bedside. Too soft, comfortable, too loud. Their ears, in their room, not a curve from a textbook.
Rooms differ, speakers differ, and hearing changes. The calibration is something we come back to, not a box ticked once.
And every sentence is checked against you before it plays.
All of the above is shaping. Shaping is exactly the thing that could, over time, walk a voice away from the person it came from.
It never stops sounding like you. That is the whole point.
Why a general model with a good prompt is not this.
What it watches for, and when it reaches a person.
Proxi is not emergency support and does not provide medical advice.