How an 8‑Year‑Old Built an Offline Voice‑Controlled AI Storyteller on a $8 ESP32

Offline AI storytelling device

An eight‑year‑old developer built a compact, fully offline voice‑controlled storytelling system using an ESP32‑S3 microcontroller (approximately $8), an INMP441 I2S microphone, a MAX98357A amplifier, and a small speaker.

Model foundation

The device runs the TinyStories LLM, an open‑source language model created by slvDev for ESP32 that can generate short stories without any network connection. The developer forked the repository and added:

On‑device voice interaction

Playback commands: “play”, “pause”, “continue”, “next story”

Simple on‑device animations displayed while the story is read

Runtime behavior

When the microphone captures a command such as “tell a funny story”, the ESP32‑S3 executes the following steps:

Wake‑word detection triggers the TinyStories model.

The model generates a short narrative.

The generated text is sent to the on‑board text‑to‑speech (TTS) engine, which streams audio through the MAX98357A amplifier to the speaker.

Simultaneously, a predefined animation is rendered on the device’s display.

All processing—including speech recognition, story generation, TTS synthesis, and animation rendering—occurs locally; no Wi‑Fi, cloud services, or external APIs are required.

Current limitation and next steps

The primary limitation identified is the quality of the ESP32‑based TTS output, which sounds basic in English. Improving the naturalness and intelligibility of the synthesized speech is the next development goal.

Source code

The full project, including the modified TinyStories model and firmware, is available at https://github.com/sri9s/tinystories-language-models.