tape.card audio player

A book you can put on the table.

A four-year-old picks a card off the shelf, puts it on the box, and hears their dad read Where the Wild Things Are. Takes it off, playback stops. Puts it back tomorrow, it carries on where it stopped. No screen, no account, no subscription, no internet.

Commercial versions of this exist and they are good. They are also priced like razor blades: the player is $80–$130 and every story after that is $10–$20, forever, and the catalogue is somebody else’s. The part that does the magic — the chip in the card — costs about twenty cents.

This surface is where that gets designed and built, in the open. Hardware, firmware, the phone-side recording pipeline, and the enclosure. Everything here is either a finished decision with the reasoning attached, or an open question marked as one.

The one fact that shapes everything. An NFC tag cannot hold the audio. The cheapest sticker (NTAG213) has 144 bytes of user memory; the biggest tag anyone sells cheaply (NTAG216) has 888. One minute of speech is about 180,000 bytes. So the card holds a 64-bit pointer and the audio lives on an SD card inside the box. Everything nice about this design falls out of that — including “many books on one card”, which stops being a feature and becomes the length of an array. The full argument.

How it works

phone or laptop record a parent reading it aloud upload the box NFC reader 13.56 MHz SD card audio + manifest microcontroller card id → playlist → decode → audio out buttons back / next / vol audio out wired, or Bluetooth a card one 64-bit id ~20¢, no battery speaker loud, but not too
The card never carries audio. It carries an id; the manifest on the SD card turns that id into a playlist of one book or six. The upload takes whichever of three routes is open — the box’s own page over the LAN, the SD card by hand, or an opt-in encrypted relay for a grandparent in another city.

What it costs

 TonieboxYoto Playertape
The player$129.99$79.99–$109.99~$45, or ~$56 portable*
One more story$19.99$9.99–$14.99$0.14
Ten more stories$199.90$99.90+$1.40
Who reads ita voice actora voice actorwhoever the child misses
Works with no internetyesyesyes
Where the recording livesvendor cloudvendor cloudan SD card in your house
Still works if the company foldspartlypartlyyes

* Real parts, now chosen — the bill of materials has every line and what it beat. (Last round this said “~$35”, guessed before choosing anything; the honest figure is higher.) Add ~$14 for a deck of cards and a hundred stickers, which is 14¢ a story.

Where things are

Design record this is the meat

The product architecture, the landscape survey, and every decision with its reason. Read this first — four of the obvious assumptions about this project turn out to be wrong, and it is cheaper to find out here than at the soldering iron.

Hardware chosen & linked

Every part with a checked link to buy it, what it beat and why, the schematic and the power chain, and a live power budget. The check is not “does it return 200” — it is whether the page still says the part’s name.

Record your first story start here

Fifteen minutes, most of it spent reading a book out loud. Written for a parent with a phone, not an engineer with a bench — including how to read aloud for a recording, which matters more than any of the electronics.

Assembly structure only

Eight stages, each ending at a test you can fail, ordered so that a fault is always in the one thing you just changed. The steps wait on a box actually existing.

Firmware specified

What runs on the box: the card-presence state machine, the manifest index, the ingest server, and the audio pipeline behind a swappable output stage.

Studio works today

Record a book in the browser. Runs here for trying out, and ships inside the box so that the recording and the upload share an origin — which they have to.

Cards works today

What is actually on a card, byte by byte, and a writer that puts it there over Web NFC. Every card’s URL lands here, so a card found under the sofa can say what it is.

Simulator works today

The whole box, in a tab. Tap a card, hear it play, press the buttons. The point is to get the interaction wrong here, where being wrong is free.

Where this actually is

 State
Product & site architecturesettled18 decisions with their reasons, three of them corrections found before anything was bought
ComponentschosenEvery line linked and the link verified — not that it returns 200, that the page still says the part’s name
Schematic & pin mapdoneGenerated from code, so the drawing cannot disagree with the firmware
Power budgetmodelled~15 h of stories a charge, ~6 days between them
The web halfworkingstudio, card writer, simulator — all live, all tested
Enclosure3 routes, none chosenWaits on one measurement
Firmwarenot one lineFully specified, entirely unwritten. This is the long pole and it is honest to say so
Anything physicalnothing builtNo part has been bought, no joint soldered, no card read
The next move is a $145 order and a ruler. Wave one is everything that runs off USB; it answers the one question the enclosure, the panel layout and the connectors all depend on — how far this actually reads, through the material you meant to use, with a sticker on a real playing card. The two-wave split, and the five things most likely to go wrong.
The cards are a deck of playing cards. A poker deck, an NFC sticker on the back, the id written on in biro — and the card lives in the book it belongs to, which is a better filing system than anything a product designer would have reached. Cards are consumable at 14¢; the recordings are the asset, and they never move.