# moonshine-ai/moonshine

Very low latency speech to text, intent recognition, and text to speech, for building voice agents and interfaces

Repository: https://github.com/moonshine-ai/moonshine
Canonical: https://ross.abutalabs.com/products/moonshine
Homepage: https://moonshine.ai
Language: C++
License: NOASSERTION
License Family: other
Topics: intent-recognition, stt, tts, voice, voice-recognition
Last push: 2026-08-25T00:31:10+00:00

## Health v2 (maintenance only)
Score: 89/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 99, longevity 49
- inputs: {"age_days": 698, "days_push": 9, "days_rel": 9, "gap_med": 5.5, "n_releases_24m": 23}
- flags: no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 10944, forks 598 (observed 2026-08-28T04:10:44.779269+00:00)

## What it is
Moonshine Voice is an open-source on-device AI toolkit providing very low latency speech-to-text, intent recognition, and text-to-speech for building real-time voice agents and interfaces. It ships models from 1MB up to Whisper-Large-V3-beating accuracy, with a unified API across Python, JavaScript/WASM, iOS, Android, macOS, Linux, Windows, and Raspberry Pi.

## Use cases
- transcribe live microphone audio to text with low latency
- build a real-time voice agent that responds while the user is still talking
- run speech-to-text entirely on-device without API keys or cloud calls
- convert text to natural speech in the browser or on mobile
- clone a custom voice from a few seconds of recorded speech
- add intent recognition to a spoken voice interface
- build a dictation or meeting-notes app with speaker-aware transcripts

## When to choose
- you need low-latency streaming transcription for live voice agents
- privacy matters and audio must never leave the device
- you need one speech stack across many platforms including embedded and WASM
- you want tiny models for resource-constrained devices like Raspberry Pi
- you want MIT-licensed speech models without per-call cloud costs

## When to avoid
- you need batch offline transcription of large audio archives where latency is irrelevant
- you need the widest language coverage, since streaming models focus on English and a few languages
- you require commercial use of the legacy non-streaming non-English models, which carry a non-commercial license
- you need heavy-duty NLU beyond intent recognition in a full conversational AI platform

## Facets
- artifact type: library
- maturity: active
- function: speech-recognition, tts, nlp, machine-learning, sdk
- domain: speech-processing, artificial-intelligence, cross-platform
- platform: python, cpp, wasm, windows, browser, cross-platform
- tags: voice-agents, on-device, low-latency, streaming-asr, voice-cloning, intent-recognition, dictation, natural-language-processing, ios, android, macos, linux

## Member repositories
- moonshine-ai/moonshine (main) score 89

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:10:44.779269+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:17:18.867288+00:00, confidence not recorded.
  - readme: https://github.com/moonshine-ai/moonshine (fetched 2026-08-28T04:10:44.779269+00:00, sha d666f3f97157)
  - homepage: https://moonshine.ai (fetched 2026-08-29T08:16:01.945331+00:00, sha 09e9140d4977)
  - site_page: https://moonshine.ai/tts (fetched 2026-08-29T08:16:01.954155+00:00, sha 80ca4f7c7b43)
- Data as of 2026-08-30T08:39:29.467469+00:00.