The smartest thing inside this smart speaker is still a person
In an era when algorithms increasingly mediate the most intimate human exchanges, one father in the maker community built a device that does the opposite of what modern technology promises: rather than replacing himself with artificial intelligence, Sean Clark used a $24 microcontroller to ensure his son's questions reach him directly. Dadlexa, a homemade smart speaker, is less an invention than a philosophical statement — that the most sophisticated response a child can receive is still a human voice that knows and loves them.
- As AI assistants colonize children's bedrooms, one parent grew uneasy with the idea of an algorithm standing between him and his son's curiosity.
- Clark built Dadlexa from a $24 ESP32-S3 board, wiring together a microphone, speaker, and LED ring into something that looks like a smart speaker but behaves like a telephone to dad.
- When his son says 'Hi ESP,' the device records the question, routes it through Telegram, and waits — not for a server farm, but for a father to pick up his phone and speak back.
- Local music playback via microSD and a web-based upload interface give the device independence for simpler tasks, so Clark is only summoned when his son actually needs him.
- The project is landing as a quiet counter-signal in the maker community — proof that the same tools used to automate human connection can, with intention, be used to protect it.
Sean Clark's son speaks to what sounds like an ordinary smart speaker, and an answer comes back — warm, immediate, unmistakably his father's voice. What the boy doesn't know is that his dad is somewhere nearby with his phone, listening and responding in real time. This is Dadlexa, and its most radical feature is what it leaves out.
Built around a $24 Waveshare ESP32-S3 development board, the device uses a local speech recognition system to listen for the wake phrase 'Hi ESP.' When triggered, an LED ring illuminates, recording begins, and the captured audio is sent through Telegram to Clark's phone. He replies with a voice message; the speaker plays it back. No cloud servers, no training data, no AI intermediary — just a direct line between a child's question and a parent's answer.
Clark wrote the firmware himself using Espressif's ESP-IDF framework, and the experience is seamless enough to feel like any commercial smart speaker. For simpler moments, the device also handles local voice commands that trigger MP3 playback from a microSD card — 'play bedtime' works without involving Clark at all. A small web interface lets him upload new audio files and assign custom trigger phrases whenever he likes.
What distinguishes Dadlexa is not its technical complexity, which is modest, but its deliberate refusal. In a landscape built around replacing human presence with automated convenience, Clark used modern hardware to accomplish the opposite — ensuring that the intelligence inside his son's smart speaker is still, unmistakably, a person who is paying attention.
Sean Clark's son asks a question to what sounds like a smart speaker. The device listens, processes the voice, and sends it somewhere. Then an answer comes back through the speaker—crisp, immediate, unmistakably human. What the boy doesn't know is that his dad is on the other end of that conversation, sitting somewhere with his phone, speaking directly into the device that sits in his son's room.
This is Dadlexa, a homemade smart speaker that does something radical in an age of algorithmic mediation: it removes the algorithm entirely. Instead of machine learning, instead of cloud processing, instead of an AI assistant standing between parent and child, there is just a direct line. Clark built it from a $24 Waveshare ESP32-S3 development board—the kind of component you can order online and have in your hands within days. The board came with everything he needed: a microphone, a speaker, a microcontroller, and a small ring of LEDs to show when the device was listening.
The mechanics are straightforward. The speaker waits for the wake phrase "Hi ESP," listening through a local speech recognition system called ESP-SR WakeNet. When the child says those words, the LED ring lights up, a tone sounds, and recording begins. The device captures audio until it detects several seconds of silence, then processes what was said locally before sending it to Telegram. Clark receives the message on his phone. He responds with a voice message of his own. The speaker downloads his reply and plays it back through its speaker, all without ever touching a cloud server or training data set.
Clark wrote the firmware himself using ESP-IDF, Espressif's development framework, and integrated components from the company's speech recognition toolkit. The result feels like using any commercial smart speaker—the lag is minimal, the interaction is natural, the experience is seamless. But the intelligence powering it is not artificial. It is biological, parental, and present.
The device has other tricks. Clark added support for local voice commands that trigger music playback directly from a microSD card. A manifest file maps spoken phrases to MP3 files, so commands like "play milkshake" or "play bedtime" work without involving Telegram or Clark at all. Once the device connects to Wi-Fi, it hosts a small web interface where new MP3 files can be uploaded, assigned custom trigger phrases, and stored on the card. After a reboot, the new commands are live.
What makes Dadlexa notable is not its technical sophistication—it is relatively simple—but what it refuses to do. In a technological landscape increasingly built around replacing human interaction with automated systems, Clark used modern hardware and software to accomplish the opposite. He did not put an AI assistant between himself and his son. He created a direct line. The smartest thing inside this smart speaker is still a person, and that person is paying attention.
Citas Notables
In a world increasingly focused on replacing human interaction with automated systems, Dadlexa uses modern hardware and software to accomplish the exact opposite.— Project description
La Conversación del Hearth Otra perspectiva de la historia
Why build this instead of just using a regular smart speaker and talking to your son directly?
Because sometimes you're not in the room. You're at work, or upstairs, or somewhere else. A regular speaker would just give him a generic answer from a database. This one gives him you.
But doesn't that require you to be constantly available? Waiting for questions?
Not constantly. The messages come through Telegram. I respond when I can. It's asynchronous, but it's still me. There's no pretense that a machine is thinking.
What made you decide to strip out the AI entirely?
I think there's something lost when every interaction gets mediated by an algorithm. I wanted my son to know he was talking to a person, even if that person wasn't in the room.
The device also plays music locally. Why include that?
Because not every interaction needs to go through me. Some things—bedtime music, a song he likes—those can just work. It's about balance. Direct connection where it matters, autonomy where it doesn't.
Do you think other parents will build this?
I hope so. The hardware is cheap. The code is out there. It's not about the technology being hard. It's about deciding what you actually want a smart speaker to do.