A Mutually Incomprehensible AI Conversation System

Overview

This project was created to highlight how wonderful it is when an AI actually speaks your language. To do that, I intentionally built a system where both sides communicate in languages they cannot understand.

Your words are converted into mysterious hiragana and passed on to the AI.

The AI, after struggling with this meaningless string, responds — and its reply is then converted into KeroKero (a Japanese onomatopoeia that represents a frog’s call, like “ribbit ribbit”) audio before reaching you.

I aimed to create the most absurdly funny use case for the amazing LLM Module!

System Flow

Unlike ordinary conversation systems, this one converts both the speaker’s voice and the LLM’s response into incomprehensible formats.

Flow from Speaker to LLM : STT (Speech to "適当-Tekitoh" )

• Only the volume level of the microphone input is detected.
• While the volume stays above a certain threshold, random strings are generated.
• The generated meaningless string is sent to the LLM Module.

Flow from LLM to Speaker:TTK (Text to Kero Kero)

• The system receives a response from the LLM Module.
• The response string is then converted into high-pitched, unintelligible speech (a “KeroKero” voice) and played back.

By repeating these processes, both sides can interact entirely in unknown languages.

Firmware

The source code for this project is published in the examples/llm directory of the kerokerovoice library on GitHub:(https://github.com/misetteichan/kerokerovoice)

Download the ZIP file, install it into Arduino IDE, compile the sample sketch, and upload it to your device.

Stack-chan

While the conversation works with only an M5Stack CoreS3 and the LLM Module, I adopted Stack-chan as the robot’s appearance to give the conversation partner a physical presence.

What is Stack-chan?Stack-chan is a hand-sized, super cute communication robot developed and released by Shishikawa-san.It is released under the Apache version 2.0 license and has an active community contributing to its development.

Project page: https://github.com/stack-chan/stack-chan