Speech Recognition and TTS in less than 500kb (github.com)

Moonshine Voice, an open-source AI toolkit, has unveiled Moonshine Micro, a streamlined version poised to enable real-time voice applications on resource-constrained devices like microcontrollers and DSPs. Utilizing the Raspberry Pi RP2350 as a reference platform, which costs just 80 cents, this toolkit can run essential functions such as voice activity detection, speech-to-text, and neural text-to-speech within an impressively small footprint—just 470 KB of RAM. This project aims to democratize access to advanced voice technologies, making them feasible for developers working on embedded systems. The significance of Moonshine Micro lies in its potential to broaden the scope of AI-driven voice applications, especially in IoT devices where power and memory limitations are critical. With components running as low as approximately 89 KiB for voice activity detection and 1.3 MiB for speech-to-text, the modular design allows developers to integrate these capabilities flexibly using TensorFlow Lite Micro. By providing this toolkit under the permissive MIT License, Moonshine Voice encourages commercial exploitation and innovation in voice interaction technology across various platforms, enhancing the AI/ML community's ability to create efficient, responsive voice agents in embedded systems.