Speech Recognition and TTS in less than 500kb

	Speech Recognition and TTS in less than 500kb(github.com)
	560 points by petewarden 5 days ago \| 85 comments
	tl;dr: Moonshine Micro is an open-source (MIT-licensed) voice AI toolkit for microcontrollers, bundling voice activity detection, speech-to-text, and neural TTS into ~468 KB of RAM and ~3.6 MB of flash, with the Raspberry Pi RP2350 (80¢) as its reference platform. The three components share a single TFLM arena and run sequentially, producing classify-and-speak latencies of roughly 0.7–1.0 seconds. A complete example demonstrates voice-driven Wi-Fi setup on the RP2350.
	HN Discussion: ↑Enthusiastic users showcasing extensions or successful hands-on trials of the toolkit ~Skepticism about accuracy tradeoffs in such a small footprint TTS/STT system ↑Comparing favorably to existing low-memory TTS alternatives like flite and nanotts •Questioning practical usefulness of voice interfaces in general workflows •Curiosity about portability, such as compiling to WebAssembly for browser use