How it works

From a first sign-in to a voice of their own

How an owner’s voice and persona are made, checked and handed back to them for review, and where Ovance’s own console fits.

Onboarding at a glance

  1. 1 Owner to App Signs in to their own app, never to Ovance
  2. 2 App to API Create the account and the person
  3. 2 API to Data A key of its own, rows nobody else can read
  4. 3 App to API Questionnaire answers, word for word
  5. 4 Owner to Voice The voice sample, straight to the voice engine
  6. 4 Voice to Speech Was the consent heard? Cut the best stretches
  7. 6 API to Model Everything the owner wrote and said
  8. 6 Model to API A persona, every quote traced to its source
  9. 7 API to App Signed event: the persona is ready to review
  10. 8 API to Voice A number, contacts and rules: ready for calls
  11. 8 Console to Voice Administrators watch over the live system

Scroll to see how it flows

Step 1:Two doors, one engine

  • Your app
  • Your sign-in
  • Owner
  • App
  • Console

Owners sign in to the app they already use. Ovance’s own console is a separate door, for Ovance’s administrators only.

  • Owners never see Ovance. The app keeps its own sign-in and its own brand. Its server speaks to Ovance on the owner’s behalf.
  • Same engine behind both. Every onboarding step below works the same way whichever door it is reached through.

Step 2:An account of its own

  • Platform API
  • App
  • API
  • Data

The first sign-in creates an account and the person inside it. From that moment the account’s data is sealed off from every other.

  • Platform, account, person. An app is a platform; each customer is an account; the voice belongs to one person in it.
  • Its own key. A key is created for the account alone. Everything personal is encrypted under it.
  • Safe to retry. A repeated request never creates a second account.
  • Database. The database itself refuses to show one account’s rows to another.
  • Key vault. One key per account. Destroying it is how an account is deleted for good, backups included.

Step 3:What they would say

  • About ten minutes
  • Owner
  • App
  • API
  • Model

A written conversation that learns how the owner greets people, where their boundaries are, and what they do when they don’t know.

  • A conversation, not a form. Every answer can earn a follow-up. It stops when the ground is covered, not when a list runs out.
  • Kept word for word. Answers are stored exactly as written. They are evidence, and the persona must quote from them.
  • Never asked twice. Anything already answered aloud is credited, so the owner is not asked for it again.

Step 4:The voice sample

  • About two minutes
  • Owner
  • Voice
  • Speech
  • Data

About two minutes of speech. The audio goes straight from the owner’s browser to the voice engine, never through the app.

  1. 1 Consent

    The owner reads a consent statement aloud, ending in a sentence made up for that one reading. The recording is kept as the proof.

  2. 2 Read and talk

    A paragraph written to sound like talking, then two short stretches of free talk. The page checks the room is quiet enough.

  3. 3 Check

    The consent is transcribed as it is read. Nothing continues until the statement and its sentence are heard on the recording.

  4. 4 Pick the best

    The cleanest stretches of whole sentences are cut out, with room noise gated away. The best becomes the voice; the rest stay as alternates.

  5. 5 The voice

    A voice is made from that stretch alone. Only the owner’s own voice is ever cloned, and they can withdraw consent at any time.

Step 5:A spoken interview, if wanted

  • Optional
  • Owner
  • Voice
  • Model

Optional. An AI interviewer talks with the owner to hear how they really sound: their stories, their laugh, their way of closing a call.

  • Typed or spoken. The same interview can be typed or held aloud.
  • It asks about what they wrote. Questions build on the questionnaire, which draws out real speech instead of performed speech.
  • Never holds onboarding up. With today’s voice it is not required. It sits under "More you can do".

Step 6:Building the persona

  • Grounded in their own words
  • API
  • Model
  • Data

What the owner wrote and said becomes a description of how they talk and decide. Nothing in it may be invented.

  1. 1 Evidence

    The questionnaire and any interview, exactly as given. The originals are kept untouched beside the persona.

  2. 2 Extract

    A language model fills in the persona, quoting the owner wherever it claims to use their words, and tagging each quote with where it came from.

  3. 3 Ground

    No model here: every quote must actually appear in the evidence, and the answer to "are you an AI?" must never deny it.

  4. 4 Verify

    A second, different model reads the evidence and the draft and decides what stands. The model that wrote it may not judge itself.

  5. 5 Stamp

    The finished persona records which model made it, from which evidence, and when. The app is told it is ready.

Step 7:The owner has the last word

  • Nothing hidden
  • Owner
  • App
  • API

Before it ever speaks for them, the owner sees what it learned and can keep, change or reject each point.

  • Review what it learned. Each thing it believes about the owner is shown with the words it came from.
  • Never say. A list of things the double must never say aloud. Once saved, even the owner’s app cannot read them back.
  • The result, as it will be used. The owner can read the persona exactly as the double will follow it.

Step 8:Ready for calls

  • A number and its rules
  • Owner
  • App
  • API
  • Voice

A number, the people who may reach it, and a rule for each. Then a test call, to hear it for themselves.

  • A shield number. A second number, chosen by area code, that the double answers.
  • Contacts and rules. For each person: answer, take a message, pass it on, or politely decline.
  • Try it. A call from the browser, straight to the voice engine, before any real caller rings.

The console, and what stands behind onboarding

Everything onboarding touches is lit.

Cyan edge below: used in onboarding

Sign-in

  • Owners Sign in to their app. Ovance never holds their password.

Data

  • Database Accounts, evidence, personas, consent, review items.
  • Key vault One key per account; destroying it is the deletion receipt.
  • File storage Voice samples and recordings. Biometric data, reachable only through the API.

Models

  • Interviewer Leads the questionnaire and the interview, and judges when enough is covered.
  • Persona builder and verifier One model writes; a different one decides what stands.
  • Voice The cloned voice. The consent text names everyone who holds it.

Our services

  • Platform API Accounts, onboarding, personas, numbers, contacts.
  • Voice engine Recording, the spoken interview, test calls.
  • Console For Ovance’s administrators.

Promises we keep

  • Consent first No voice is made without recorded consent, and it can be withdrawn.
  • Deleted means deleted Closing an account destroys its key, so its data can’t be read again, and its voice is removed from every provider that holds it.
  • It won’t make things up Every line of the persona traces back to the owner’s own words.