The one voice they have always known
is still the one that calls them home.

VoiceKind delivers familiar family voices to patients who can no longer communicate — after a stroke, with dementia, or in a disorder of consciousness. Families record from their phone. Care staff play at the bedside. The response is measured.

Penn State HealthClinical co-investigator
USPTOPatent pending — 12 methods
NIH SBIRPhase I application in preparation
Delaware C-CorpFounded 2026
The problem

An estimated 14 million Americans cannot communicate with their families.

63 million family caregivers are trying to reach them. When families are not at the bedside, patients hear monitors, alarms, and strangers — not the one voice they would recognize.

“I know what I want to say.”

The thought is intact. The words will not come. Family visits less, because visiting is painful.

Stroke · aphasia2M+
“Where is my daughter?”

Asked dozens of times a day. Answered, forgotten, asked again. The distress is new every time.

Dementia · Alzheimer’s7.2M
They may be hearing everything.

Brain imaging shows a familiar voice still activates language and memory centers. No way to respond.

Disorders of consciousness117–322K
The evidence

Familiar voice therapy is proven.
The delivery mechanism did not exist.

VoiceKind is built on published, peer-reviewed research. Every claim below is sourced.

Family voice outperforms staff voice

Family-delivered voice stimulation produced approximately five times greater improvement in consciousness scores than staff-delivered stimulation.

Blue Cross Blue Shield Medical Policy Review MED205.014, November 2024
fMRI

Patients in a coma still process a familiar voice

Familiar voice activated language, memory, and prefrontal regions on functional MRI in patients with disorders of consciousness.

FAST Study, NCT00557076 — VA-funded, Northwestern University

Their own name is the strongest signal

Patients oriented to hearing their own name 40 percent of the time, compared with 23 percent for a neutral sound.

BMC Neurology, 2013 — 86 patients, PMID 23506067
40%

Misdiagnosis is common, and standardized assessment catches it

Misdiagnosis among patients with disorders of consciousness consistently approximates 40 percent. Standardized behavioral assessment is more sensitive than clinical consensus.

American Academy of Neurology / ACRM / NIDILRR systematic review guideline
How it works

Built to fit the shift, not interrupt it.

VoiceKind adds roughly ten seconds per patient to rounds that already happen.

A family member records

Thirty seconds on a phone, from anywhere. A greeting, a story, ordinary news. Guidance is built in, tailored to the condition.

The message reaches the bedside

Encrypted transfer. It appears in the care team’s queue within seconds, flagged if the family saved it for a particular day.

Staff play it during rounds

One tap while checking vitals. A pillow speaker keeps it private from a roommate. The nurse is not the audience — the patient is.

The response is recorded

A standardized response score, structured observations, and an optional note. Five seconds. This is the clinical data.

The system learns

Over time, patterns emerge — which voice, which content, which time of day. Recommendations follow the data, and staff always decide.

Coverage

Nine conditions. One platform.

Features are routed by diagnosis. A patient in a disorder of consciousness and a patient recovering from a stroke need almost nothing in common.

Stroke & aphasia Disorders of consciousness Traumatic brain injury Dementia Alzheimer’s disease Parkinson’s disease ALS Cancer care Palliative care
Governance

AI voice cloning, designed for clinical review.

Voice cloning in healthcare raises real questions. We built the answers into the architecture rather than the terms of service.

Every recording is labelled

Staff always see whether a voice is a real recording or AI-generated. There is no ambiguity at the point of care.

Consent is explicit and revocable

A documented multi-step process before any voice is cloned. One tap deletes the model and every recording made from it, permanently.

Patient disclosure is configurable

An audible cue, a spoken statement, or staff-mediated notice. The clinical oversight body selects the mode per patient.

Everything is logged

An append-only audit trail covering consent, generation, playback, and revocation. Exportable for IRB or regulatory review.

Team & advisors

Clinical neurology. Biostatistics. Voice authentication.

Two advisory appointments are in progress and will be announced once confirmed.

Wun Kuen Ng

Founder & Chief Executive Officer
B.A. Barnard College, Columbia University · MFA + MBA, Mills College

Designed the platform architecture across nine conditions, the twelve patent-pending methods, and the AI governance framework.

Dr. Ramin Zand, MD

Strategic Partner · Clinical Co-Investigator
Penn State Health

Vascular neurologist with more than 130 peer-reviewed publications in stroke outcomes and machine learning in neurology. Co-designing the clinical trial.

Biostatistics AdvisorTO BE ANNOUNCED

Clinical trial design & statistical analysis

Advising on trial methodology, outcome measures, and analysis plan for the clinical pilot and NIH SBIR application.

Voice Authentication & AI AdvisorTO BE ANNOUNCED

Privacy-preserving voice architecture

Advising on synthetic speech detection, on-premises inference, and the machine learning architecture behind voice recommendation.

Questions

What people ask us

Is VoiceKind available today?

Not yet. We are preparing a clinical pilot and building toward a production release. If you are a facility interested in participating, or a family who would like to be notified when we launch, please get in touch.

Does this add work for nursing staff?

Roughly ten seconds per patient, during rounds that already happen. The nurse taps play while checking vitals and rates the response afterwards on a single scale. The design constraint was that it had to fit inside existing workflow, not alongside it.

How is patient privacy handled?

All data is encrypted and access-controlled, and every action is recorded in an append-only audit trail. Messages play through a pillow speaker so a roommate does not hear them. We are designing toward on-premises inference so that voice data need not leave the facility at all.

Is this a medical device?

No. VoiceKind is a general wellness product. It does not diagnose, treat, cure, or prevent disease. Clinical research involving human subjects is conducted under institutional review board oversight.

What does the AI actually do?

Two things. It can generate new messages in a family member’s voice from typed text, so a daughter does not have to record every message individually. And it learns which voices and content produce the strongest response for an individual patient, then recommends accordingly. Staff always make the final decision.

Can families use it without a facility?

Not initially. Our first deployments are through skilled nursing facilities and rehabilitation units, where staff play messages at the bedside and record the clinical response. Home use is on the roadmap.

Get in touch

We are looking for partners.

If any of the following describes you, we would like to hear from you.

Healthcare facilities

Nursing homes and rehabilitation centers in the New York area, for a thirty-day clinical pilot. VoiceKind provides the platform, devices, and staff training at no cost.

Discuss a pilot

Clinicians & researchers

We are building an advisory board across neurology, speech-language pathology, and biostatistics, and preparing an NIH SBIR application.

Get in touch

Engineers

We are hiring a CTO or technical lead. React, Node.js, and healthcare experience. Fractional or full-time.

See the role

Investors

We are raising a pre-seed round to fund the production build and the clinical pilot.

Request materials