Talking plush · Acoustics inside fabric
Talking plush OEM — the voice comes out through fabric, and that decides half the electronics
Hard-shell robots get their acoustics designed; plush gets them inherited. A driver behind 20 mm of polyester fibre loses most of its high end, a microphone under the same fibre loses sensitivity and gains the sound of the child’s sleeve, and the cavity the speaker needs is the cavity the stuffing is currently occupying. Every one of those is solvable — but only if the plush pattern and the electronics are drawn by people talking to each other, which is usually not how a toy brand and a board house work.
Who usually comes to us
Plush toy makers adding voice
You have the patterns, the sewing lines and the retail accounts. What you do not have is a firmware team, and the acoustic problem is new to you.
Character brands with a bedtime story
The character already talks in a series or a book; the plush has to sound like the same character at 8pm in a dark room.
Companion and wellness brands
A soft object for adults or older users where the interaction is calm and the voice is quiet — no screen, no app-first onboarding.
Retail private label
A seasonal SKU that must hit a landed cost and a shelf date, where the differentiator is the character and the price, not the silicon.
The acoustic path through fabric
Four decisions, in this order. Get the order wrong and you end up with a product that needs a stronger speaker to overcome a hole that should not be there.
- Where the sound leaves
- A moulded mesh window behind a seam, at chest height or the side of the head, away from where a hand naturally grips. Not the face: the face is where the child looks, and a grille there reads as a device.
- What volume is behind the driver
- A sealed rear chamber of a defined size, or a hollow body designed for it. Fill the rest with fibre, never the chamber. Decide this with the pattern maker, in the same meeting as the hug-feel.
- Where the microphone sits
- Front-facing, high enough to clear a hug, isolated from the shell with a soft mount, and as far as possible from the driver so the toy does not hear itself. Two microphones give direction and let you reject the sleeve noise coming from below.
- What the firmware knows about the room
- A short noise profile on wake, adaptive gain, and an explicit give-up: if recognition confidence is low, the character asks again in persona instead of guessing. Guessing is what makes a toy feel stupid, and it is a firmware choice.
Measure at 1 m and at 3 m, in a carpeted room and in a tiled one, with a television on. The tiled room with the TV on is the only measurement that predicts the complaint.
The pod: washability, service and returns
- 01
One connector, keyed, rated
Power, audio and the sensor bundle through one connector with a stated mating cycle count. Two connectors means two chances to plug it in wrong, and a parent will.
- 02
A closure a parent can work
Velcro, a zip with a child-proof pull, or a screwed plate — chosen against the age band and the market’s small-parts rules. The closure is a safety component, not a sewing detail.
- 03
A toy without the pod
The plush has to keep its shape and its face when the electronics are out, because that is how it spends wash day. If it collapses, the child will not accept it back.
- 04
Field diagnosis over the app
Battery health, wake counts, recognition failure rate, firmware version and the last blocked answer, readable by support without the unit coming back. A returned plush costs more than the whole warranty budget of a good one.
What decides whether a talking plush sounds like a toy or a speaker in a bag
Two millimetres of fabric is a low-pass filter
High frequencies are absorbed by fibre and scatter at the weave. The fix is not more power — it is a hard grille or a moulded mesh window hidden behind a seam, positioned where a child’s hand does not cover it. Both have to be in the pattern, and the pattern is usually already approved.
The cavity is inside the stuffing
A driver needs a volume of air behind it or it produces no low end, and plush is by definition full of the thing that absorbs it. Moulded foam shells, a sealed rear chamber or a designed hollow body change how the toy feels when hugged — which is a product decision, not an audio one.
Microphone under fabric, next to fabric
Two noise sources arrive together: the sensitivity loss through the cloth, and the sound of the cloth itself — a sleeve brushing the toy, a blanket dragging across it, the seams moving when it is squeezed. Directional placement and a rubber-isolated mount do more than noise suppression does, because the suppression has already thrown away the child’s quieter syllables.
It has to be washable, which means it has to come apart
A parent will eventually put it in a machine whatever the label says. The electronics pod has to be removable with one hand, the connector has to survive a hundred cycles, and the toy has to still be a toy with the pod out — because that is the state it will live in for a day. Design the pod as a module and the washing question stops being a support problem.
The antenna is inside a bag of water and thread
Plush absorbs 2.4 GHz, and stuffing density varies from unit to unit because a person stuffs it. Keep the antenna away from the battery, the speaker magnet and the metallic trim, and verify range on ten units rather than one — the spread is the number that matters, not the best sample.
Directions to start from
AI Talking Plush
A plush your child actually talks to
- Parent app
- Content guardrails
- Offline wake word
- No camera
Ages 3–10
NFC Story Plush
Tap a card, hear a story
- NFC cards & charms
- Tap to play
- Moving ears
- Touch sensing
Ages 3–8
Two ways to develop with us
White-label plush with your voice
An existing plush shell and electronics, your persona, your voice pack and your packaging. No tooling on our side; your artwork and label are the brand.
Best for: A first connected SKU with a shelf date to hit.
Your pattern, our electronics
You bring the character and the pattern; we design the acoustic window, the cavity, the pod and the board to fit inside it, then run the tolerance work with your sewing line.
Best for: The soft goods are the brand and must not be redesigned by a hardware team.
Full development
Pattern, electronics, mechanism, firmware, app and cloud from a blank sheet, with the acoustic prototype built before the tooling is cut.
Best for: A form factor that does not exist yet — a puppet that moves, a plush that behaves.
Platforms
Bring the AI stack your market needs
Our firmware is decoupled from any single provider. Depending on where you sell, your data-residency rules and your budget, we connect the device to the right LLM, speech and IoT cloud. These are the kinds of services we have integrated.
For China
- Baidu AI Cloud Solution sales partner Cloud AI and voice services
- Mainstream Chinese LLMs Several to choose from, per project
- Speech recognition & synthesis Major speech engines
- IoT provisioning & branded app Device setup and app under your brand
- Audio / video storage & delivery Audio / video cloud for content
For global markets
- Leading overseas LLMs Chosen per market and language
- Realtime voice dialogue Realtime voice models
- Neural speech synthesis Speech recognition and natural voices
- Device connectivity & fleet management Connect and manage devices at scale
- EU / US data hosting Data kept in the region you sell in
Questions buyers ask
Can the electronics be removed so the plush can be washed?
Yes, and on a children’s plush we treat it as a requirement rather than an option. A removable pod with a keyed connector, a closure a parent can open without tools, and a label that states what may go in the machine. It costs a little BOM and a little assembly time, and it removes the most common early support call.
Does the plush need a camera for the toy to feel alive?
No, and we do not put one in plush. Presence comes from timing and sound — a pause before answering, a small sound when picked up, ears that move at the right moment. A camera in a children’s plush is a liability with a small upside, and the plush lines we build have no camera and no interface to add one.
How loud is too loud for a toy held against the ear?
Quieter than most demos. A plush gets pressed against a face, so the design target is a ceiling that cannot be exceeded by button-mashing or by a firmware bug, implemented as a limiter in the audio path rather than a maximum value in the UI. We set it with the client against the requirements of the markets they sell into.
Can you hit a target landed cost?
Tell us the number and the volume and we will tell you which of the three configurations fits it, honestly, including when none does. Cost in this category is decided by driver size, battery capacity and whether the shell is a new mould — not by the board, which is the part buyers assume is expensive.
Who writes the personality and the answers?
You own the persona — it is the brand. We build the container: the prompt architecture, the guardrail layer, the refusal style, the voice (cloning supported), and a private knowledge base so the plush answers from your material rather than the open internet. If you have no content team, we can draft a persona for you to approve and rewrite.
Tell us what you want to build
The category, your market, a quantity and whether an IP is involved is plenty for a first answer. Samples available; proven products are available as OEM, custom ones as ODM. Development and sample arrangements are agreed together once we understand the project.
Start a conversation