Moonshine Voice — распознавание речи на устройстве

★ 11.2K

Moonshine Voice is an open-source toolkit for developers building real-time voice agents and applications; speech recognition runs entirely on-device. Reach for it when you need a voice assistant that reacts while the person is still talking, when audio cannot leave the device, when recognition must run on a Raspberry Pi or a phone without internet, or when Whisper is too heavy. No API keys or account are required. It ships speech-to-text models trained from scratch, from accuracy higher than Whisper Large V3 down to tiny 1 MB models, with streaming that processes audio during speech. One library covers Python, JavaScript/WASM, iOS, Android, macOS, Linux, Windows and Raspberry Pi; quick start is pip install moonshine-voice and moonshine-voice mic --language en. Text-to-speech and conversational agents are covered in the docs section Using the Library. The code is MIT and the models are MIT too, except legacy non-streaming models for languages other than English, which use the non-commercial Moonshine Community License. The main direction is speech to text; the repository description also lists intent recognition and text to speech. It suits voice interfaces; check the license of the specific model before commercial use.