Moonshine Micro: Speech AI under 500KB for microcontrollers
Original: Speech Recognition and TTS in less than 500kb
Why This Matters
Bringing on-device STT and TTS to sub-$1 microcontrollers could accelerate voice AI adoption in IoT and edge devices.
Moonshine AI has released Moonshine Micro, an open-source speech recognition and TTS toolkit designed for microcontrollers and DSPs, running in under 500KB. The reference platform is the Raspberry Pi RP2350, which retails for just $0.80.
Moonshine AI has published Moonshine Micro, a sub-500KB open-source toolkit that brings real-time speech recognition (STT) and text-to-speech (TTS) to embedded systems such as microcontrollers and digital signal processors. The project is a stripped-down variant of Moonshine Voice, the company's broader AI toolkit for voice agents and applications. The reference hardware is the Raspberry Pi RP2350 chip, available for approximately $0.80 at retail. The repository includes modules for speech-to-text, neural TTS, a Klatt-synthesis-based TTS engine, voice activity detection (VAD), grapheme-to-phoneme (G2P) conversion, and feature generation, all organized for CMake-based builds targeting the Pico SDK. Example code targets the RP2350 platform directly. The project is hosted publicly on GitHub under the moonshine-ai/moonshine repository, which has accumulated 8,900 stars and 481 forks, indicating significant community interest. By fitting both STT and TTS into under 500KB, Moonshine Micro aims to enable voice interfaces on ultra-low-cost, resource-constrained hardware without requiring cloud connectivity.