This guide documents experimentation with locally running edge AI models on the Raspberry Pi 5. Two basic projects are included that demonstrate how to use Python to combine local LLMs with text to speech models.

The models used, SmolLM3 and Piper TTS, are both multi-lingual. The projects focus on translation tasks in order to experiment with these capabilities. The models aren't perfect and do sometimes go a little off the rails. The capabilities that they unlock for Raspberry Pi 5 based projects are impressive none-the-less. Here is a video demonstrating the project.

Cloud based LLMs, translation, and TTS services would likely be faster and on average give better output but they require an internet connection, come with privacy concerns, and can be costly. So running these locally on the Raspberry Pi 5 is a nice alternative.

The speaking translator CLI is a basic command line translation utility that accepts text in English, translates it into one of 5 languages, then synthesizes the translated version to speech to play out of the speakers. It also supports keeping a history of translations to quickly and easily replay them without having to redo the translation. Perfect for practicing listening and pronunciation in another language.

The weather and wardrobe assistant fetches weather from weather.gov, uses the current and forecast conditions to generate wardrobe suggestions, then translates the weather and wardrobe text into the specified language before synthesizing it to speech with the TTS model.

All of the experimentation was done on a Raspberry Pi 5 with 8gb RAM.

Parts