What is this?
Moonshine Voice is an open source AI toolkit for building real-time voice agents and interfaces, covering speech-to-text, intent recognition, and text-to-speech.
Moonshine Voice is an open source AI toolkit for building real-time voice agents and interfaces, covering speech-to-text, intent recognition, and text-to-speech.Is this one library or separate SDKs per platform?It's a single library with the same API ported across Python, JavaScript/WASM, iOS, Android, macOS, Linux, Windows, and Raspberry Pi.
It's a single library with the same API ported across Python, JavaScript/WASM, iOS, Android, macOS, Linux, Windows, and Raspberry Pi.How do I get it running?Follow the Quickstart guide at moonshine-voice.readthedocs.io, which covers install steps for every supported platform and links to runnable Examples.
Follow the Quickstart guide at moonshine-voice.readthedocs.io, which covers install steps for every supported platform and links to runnable Examples.Do I need an API key or account to use it?No — the library runs fully on-device with no account or API keys required for any platform it supports.
No — the library runs fully on-device with no account or API keys required for any platform it supports.Are all the models free to use commercially?Almost all models are MIT licensed, but legacy non-streaming models for languages other than English are under the non-commercial Moonshine Community License instead.
Almost all models are MIT licensed, but legacy non-streaming models for languages other than English are under the non-commercial Moonshine Community License instead.Does this work without an internet connection?Yes, since every model runs on-device, Moonshine works fully offline.
Yes, since every model runs on-device, Moonshine works fully offline.Where does my voice data go?Nowhere off-device — all processing happens locally on the device running the library, not on a remote server.
Nowhere off-device — all processing happens locally on the device running the library, not on a remote server.How does Moonshine's accuracy compare to Whisper?Its speech-to-text models are trained from scratch and claimed to have higher accuracy than Whisper Large V3, while also offering tiny 1MB models for constrained devices.
Its speech-to-text models are trained from scratch and claimed to have higher accuracy than Whisper Large V3, while also offering tiny 1MB models for constrained devices.Why is it optimized for streaming specifically?It does transcription work while the user is still talking instead of waiting for them to finish, which is what keeps latency low for live voice interfaces.
It does transcription work while the user is still talking instead of waiting for them to finish, which is what keeps latency low for live voice interfaces.Can I plug this into a coding agent like Claude Code?Yes — copy .agents/skills/moonshine-voice/ into your project, or run 'npx skills add moonshine-ai/moonshine --skill moonshine-voice' so the agent uses the current API instead of defaulting to Whisper or old DialogFlow names.
Yes — copy .agents/skills/moonshine-voice/ into your project, or run 'npx skills add moonshine-ai/moonshine --skill moonshine-voice' so the agent uses the current API instead of defaulting to Whisper or old DialogFlow names.What can I actually build with it besides transcription?Beyond speech-to-text, it includes intent recognition and text-to-speech, so you can build full conversational voice agents end-to-end.
Beyond speech-to-text, it includes intent recognition and text-to-speech, so you can build full conversational voice agents end-to-end.What are similar projects?sherpa-onnx, supertonic, vosk-api, whisper.cpp, runanywhere-sdks, nobodywho, jargo, pocket-tts
sherpa-onnx, supertonic, vosk-api, whisper.cpp, runanywhere-sdks, nobodywho, jargo, pocket-ttsWhich topics is it in?stt, voice-recognition, low-latency, on-device, edge-ai, voice-assistant, voice, webassembly, speech-recognition, raspberry-pi, tts, text-to-speech
stt, voice-recognition, low-latency, on-device, edge-ai, voice-assistant, voice, webassembly, speech-recognition, raspberry-pi, tts, text-to-speech