Conversational AIVoice CloningEdge AIEmbedded

iAvatars Companion Kiosk — Voice AI That Speaks in a Loved One's Voice

iAvatars' product, built with Aeonix as its technology partner: a kiosk that talks with families in the cloned voice of someone they love. Pilot batch on the way.

As presented at TEDxFOT · featured on Times Now, Zee News and Hindustan Times

iAvatars Companion Kiosk — Voice AI That Speaks in a Loved One's Voice

~3 s

From question to spoken answer (English)

Languages: English and Hindi

Pilot

Batch on the way

Countries with patents pending

Project Gallery

iAvatars Companion Kiosk — Voice AI That Speaks in a Loved One's Voice - Image 1

iAvatars Companion Kiosk — Voice AI That Speaks in a Loved One's Voice - Image 2

iAvatars Companion Kiosk — Voice AI That Speaks in a Loved One's Voice - Image 3

iAvatars Companion Kiosk — Voice AI That Speaks in a Loved One's Voice - Image 4

See It In Action

Client countryUSA

ClientiAvatars LLC, USA

ProductThe iAvatars companion kiosk is iAvatars' product

Our roleTechnology partner: hardware, voice AI and software

DeploymentHomes and senior living

TechnologiesConversational AI · Voice cloning · Edge AI · Embedded systems

StatusPilot batch on the way

The Challenge

Families who live far apart, and people in senior living, want more than a video call: a companion that remembers, talks back, and sounds like someone they love.

iAvatars needed a device that anyone could simply talk to, with no technical skills, that answered quickly and in their own language, and that kept a loved one's voice and memories private to the family.

The Solution

As iAvatars' technology partner, we built the device end to end: the kiosk hardware, the wake word that listens for the person's name, speech recognition, the AI that answers from the family's own memories, and a reply in the cloned voice with a lip-synced avatar on screen.

A simple admin panel lets the family pick the avatar and switch its language between English and Hindi, and iAvatars' team manages avatars and memories remotely.

System Architecture

Step 1Say the nameHands-free: a wake word listens for the person's name.
Step 2Answers from family memoriesSpeech is recognised, and curated family answers come first.
Step 3Replies in their voiceA cloned voice answers, with a lip-synced avatar, in about 3 seconds.

Under the hood

Device

Kiosk hardware built end to end by Aeonix for iAvatars
NVIDIA Jetson edge computer
Round screen with a lip-synced avatar

Voice AI

Wake word that listens for the person's name
Speech recognition in English and Hindi
Large language model answering from curated family memories
Voice cloning, so replies come in a family member's voice

Software

Admin panel: choose the avatar and set its language
Remote management of avatars and memories by iAvatars' team
Reminders and alarms
Python application stack

Hard problems we solved

The problemA conversation needs a fast spoken replyHow we solved itThe whole chain, from wake word and speech recognition to the AI answer and the cloned voice, returns a spoken answer in 2.6–3.4 s in English, measured on real hardware.

The problemA loved one's voice must stay in characterHow we solved itCurated family answers come first, and the AI answers from the family's own memories, so the avatar speaks as the person the family knows.

The problemOne avatar, two languagesHow we solved itEach avatar's language is set in the admin panel, so the same avatar can talk in English or Hindi.