Moonshine Voice (Moonshine v2 streaming)
MIT-licensed on-device STT library with streaming models (Moonshine v2, 2026) that do work while the user is still talking. Runs on CPU, phones, browsers (WASM) and Raspberry Pi.
Overview
Best for: Private, offline, low-latency captions and voice commands on phones, browsers and single-board computers.
At a glance
On-device library; model sizes range from tiny to large. English plus other languages (list not captured). Legacy non-English non-streaming models are non-commercial.
Microphone or PCM via the library.
n/a
English plus additional languages (list on docs site; not captured).
Paper: bounded time-to-first-token independent of utterance length; third-party claim of sub-200 ms on edge devices is unverified.
On-device.
On-device processing.
CPU-only friendly: laptops, phones, Raspberry Pi, browser WASM.
MIT (code and streaming models); legacy non-English non-streaming models non-commercial.
Features
- streaming partials
- on-device
- intent recognition and TTS in the same library
- cross-platform
Pricing
| What | Price | Unit |
|---|---|---|
| Library and models | $0 |
Runs on the user's device.
Free tier: Free and open source.
Source: github.com
Setup
- pip install moonshine-voice
- moonshine-voice mic --language en (live mic test)
- See moonshine-voice.readthedocs.io for Python/JS/mobile APIs.
Endpoint
n/a (local)Authentication
none
Quick start bash
pip install moonshine-voice
moonshine-voice mic --language en # live transcription from the default microphoneWritten from the current docs. Check the vendor's SDK version before you ship.
Warnings
Licence exception
Everything is MIT except the legacy non-streaming models for non-English languages, which stay under a non-commercial Moonshine licence. Use the streaming models for commercial products.
Benchmarks disagree
The arXiv abstract and the newer PDF on moonshine.ai give different accuracy/speed claims (6x vs 4x model size parity). Test on your own audio and hardware.
Small models trade accuracy
Tiny and micro models suit commands and captions on weak hardware; long-form dictation needs the larger models.
Plus 3 warnings that apply to all open models APIs. See category warnings.
Limits
- Accuracy depends on chosen model size and device CPU.
Models and products
| Name | Status |
|---|---|
| Moonshine v2 streaming models | Released 2026 |
| Legacy non-streaming non-English models | Legacy |
Docs and sources
Docs
Sources used
- raw.githubusercontent.com/moonshine-ai/moonshine/main/README.md
- api.github.com/repos/moonshine-ai/moonshine
- arxiv.org/abs/2602.12241
- download.moonshine.ai/docs/moonshine_streaming_paper.pdf
Supported language list and per-device latency numbers.