Tuck · Voice cloning
Lullabies in a voice your baby loves
TL;DR. Tuck composes new AI lullabies in a cloned family voice — yours, your partner's, or a grandparent who lives far away. The clone needs 90 seconds of consented recording. The model is scoped to your account, never used for anything else, and can be deleted any time. Available on Tuck Pro+.
Why a cloned voice instead of a generic lullaby
Babies recognize the voices of their family from the womb. By the third trimester they respond differently to mom's voice, to dad's voice, and to the voices of older siblings around them. That recognition doesn't fade after birth — if anything, it sharpens.
Most baby monitors give you a small library of canned lullabies recorded by a stranger. They sound nice. But they don't carry the emotional weight of your voice — or the voice of a grandparent who lives a thousand miles away and only sees baby twice a year.
Tuck's voice cloning is purpose-built for that: a short consented recording, a private voice model, and a lullaby composer that uses the voice to sing baby to sleep night after night, never repeating the same melody twice.
How voice cloning works in Tuck
Choose who to clone
Open Tuck Pro+ on the parent device and tap Voice cloning → New voice. The person being cloned needs to be present and consenting — Tuck records a short consent statement as part of enrollment.
Record a 90-second sample
Hand the iPhone to the person being cloned. They read a short script aloud — about 90 seconds of varied speech. Tuck guides them through it: a few sentences, a few sung syllables, a few quiet hums.
Tuck builds a private voice model
The sample is sent encrypted to Mureka, our AI music partner. They build a voice model scoped to your Tuck account. The model itself does not leave your account — only your Tuck app can request lullabies from it.
Compose lullabies on demand
From the parent device, tap Sing a lullaby and pick the voice. Tuck composes a fresh lullaby in that voice — different every night, calmly looped on the nursery iPhone, gently faded when baby settles.
Delete the voice any time
From Settings → Voices, you can delete a voice. Tuck deletes the model from our servers and from Mureka within 24 hours, and the original recording is purged immediately.
Listen to four sample lullabies
Before we shipped voice cloning, we ran four blind listening tests to confirm the lullaby quality was actually shippable. These four samples — Emma, Liam, Sofia, and Aarav — are the canonical reference for what a Tuck lullaby sounds like. Each is generated from a different prompt and different vocal timbre.
- Emma — gentle, acoustic guitar, lyrical
- Liam — dreamy, ambient pads, vowel-heavy
- Sofia — jazz nursery, brushed drums, hummed melody
- Aarav — hush, minimal pads, near-whispered
We'll add an in-page audio player as part of the launch. The four MP3s and lyrics are in the Tuck repository as the founding reference.
Privacy and consent — the version we'd want for our own families
Voice cloning is the most ethically loaded feature in Tuck. Three design choices came out of arguing about it for weeks:
- Consent is recorded, not assumed. Every voice clone — including your own — requires reading the consent statement aloud as part of enrollment. The audio is stored alongside the voice model so we can prove consent was given for any specific voice.
- No cloning of absent or deceased people. Tuck only supports live recording. Old voicemails, video clips, and audio from anyone not present don't enter the system. We've decided this is a line we won't cross even on user request.
- Voice models are deletable, immediately. Settings → Voices → Delete removes the model from our servers and from Mureka within 24 hours. The original recording is purged at model training, well before then.
Frequently asked questions
Whose voice can I clone?
Anyone who consents and is present to record the sample. Common choices: yourself, your partner, a grandparent who lives far away, or an older sibling. Voice cloning of someone who is not present, or anyone who hasn't recorded the consent statement, is not supported by design.
Do I need consent if I'm cloning my own voice?
Tuck still asks you to record the consent statement when cloning your own voice. It's the same flow whether you're cloning yourself or a family member — we don't shortcut it because consent flows are easier to get right when they're consistent.
How long is the recording sample?
About 90 seconds of varied speech: a few sentences spoken naturally, a few sung syllables (any tune, no skill required), and a few seconds of quiet humming. The varied content gives the voice model enough range to compose lullabies that don't sound robotic.
What if the person being cloned can't sing?
That's the point. The model only needs to hear their normal voice and a hint of their pitch range — Tuck does the actual composition. Most people who say 'I can't sing' produce wonderful cloned lullabies because the model fills in the musical piece.
Can I clone a voice from an old recording instead of doing it live?
No. Tuck specifically requires a live, consented recording. Cloning from old voicemails, video clips, or audio of someone who isn't present is not supported. This rules out cloning a deceased family member, which is a hard line we've drawn for ethical reasons.
Where does the voice model live?
The voice model is stored on Mureka's servers and scoped to your Tuck account via a per-account API key. The original recording is held only briefly (under 24 hours) for model training, then deleted. The model can only be queried by your Tuck app — it isn't exposed to other Mureka customers or to Tuck staff.
What's a lullaby actually like? Is it weird?
It's surprisingly tender. Tuck's lullabies are calm, ambient, repetitive — closer to a music-box loop than a pop song. The cloned voice carries the melody softly. We have four sample lullabies on the Tuck homepage you can listen to before signing up; the AI quality is what convinced us to ship this product.
Can my baby hear that it's their family member?
Babies recognize the voices of their parents and family from the womb — the recognition is real, even when the lullaby itself is novel. Anecdotally, families report that babies settle faster to a parent's cloned voice than to a stranger's lullaby. We're not making medical claims here, just sharing what testers have told us.
Is there a way to clone a voice without sending it to a third party?
Not today. Voice cloning quality requires models that don't yet run efficiently on iPhone hardware. As Apple Intelligence and on-device generative audio mature, we plan to migrate voice cloning to fully on-device — that's our roadmap, not a feature that exists yet.
How is this different from ElevenLabs or Apple Personal Voice?
ElevenLabs is a general-purpose voice-cloning API for developers, not a consumer product. Apple Personal Voice clones your voice for accessibility (synthesizing your speech if you lose it) and is locked to your Apple ID. Tuck's voice cloning is purpose-built for parent-to-baby lullabies: short cloning sample, scoped to your Tuck account, lullaby composition baked in, consent flow required for everyone.
Try Tuck
Voice cloning is a Tuck Pro+ feature ($11.99/month or $119/year). The base monitor is free forever — you only need Pro+ when you want unlimited AI lullabies and a cloned family voice.