Live subtitles and translation for system audio or your microphone on macOS 13+ (Apple silicon and Intel) and Windows / Linux x86_64.
Website · Download latest · 简体中文
Mimi turns speech from your computer or microphone into live translated subtitles. Watch films, follow streams, or take lessons with subtitles floating over your screen.
- Choose system audio, microphone, or both. System audio is the default; microphone capture requires explicit selection.
- Translate audio from a selected app on macOS or Windows 11.
- Show original text, translations, or both.
- Adjust subtitle position, size, and color, or let mouse clicks pass through the window.
- Save subtitles or audio locally and export TXT / WAV when needed. Saving and recording are off by default.
For your first setup, we recommend starting with Alibaba Cloud. We’ve done more real-world testing with it. We’re still improving support for other services, so results and reliability may vary.
You can also try Google Gemini; it works well with a stable connection. It currently offers a free tier. Check your quota and billing in Google AI Studio, where you can also get an API key.
- Open Settings → Speech & Translation, add a configuration, enter the requested provider credentials, and save.
- Choose the recognition and translation languages.
- Play something and turn on Live Subtitles under Subtitles. On macOS, allow Screen & System Audio Recording when asked.
Cloud speech services require your own credentials and receive your audio; usage charges may apply. Apple Speech recognizes audio locally on supported Macs. Independent text translation sends recognized text to your chosen translation service.
Setup & help · Android · Report a bug · Contributing
Apple Speech appears when the system supports it: Apple silicon, macOS 26 or later, and an available system transcriber. It needs no speech API key. The configuration shows this Mac's supported languages and resource status. Stop subtitles, choose the target language under Speech resources, and click Prepare speech resources if it is missing; this explicit action may download system assets. Then choose the same Recognition Language and start subtitles. Automatic language detection is not offered.
Use original-only subtitles, or configure a separate text translator such as Index-Translate. Apple recognition does not supply the translation model, and a remote text translator still receives recognized text.
Bilibili's Index-Translate currently offers a free public translation API (as of October 5, 2026). You can try it in Mimi and compare the subtitle translations with your usual service.
In Settings → Speech & Translation, open an Alibaba Cloud configuration and select OpenAI-compatible API under Text translation. Use the values from the official example:
| Field | Value |
|---|---|
| Service address | https://index-translate.bilibili.com/v1 |
| Model name | Index-Translate-35B-A3B |
| API Key | Leave empty; the public API currently requires no authentication. |
Save, then run the connection check beside Text translation. If this address already has a saved key, remove it with Remove translation key and save again.
Index-Translate handles text translation only: keep your Alibaba Cloud speech-recognition credentials configured; recognition may still incur charges. Free API availability is subject to the upstream service.
macOS keeps asking for recording permission even though it is enabled? Quit Mimi, then remove and re-add only its entry in System Settings → Privacy & Security → Screen & System Audio Recording. Use /Applications/mimi.app for the release app or /Applications/mimi-dev.app for development, enable it, and reopen the same app. See permission recovery.
Thanks to everyone who writes code, reports issues, tries Mimi, or shares it (๑•̀ㅂ•́)و✧
Special thanks to @yebuwudong for the Android app, and @LLLin000 for subtitle animation and Windows audio improvements.
Thanks also to @Chtholly000 for improving Gemini's continuous captions and planned connection rotation.
Thanks to the people in V2EX, LINUX DO, Appinn, NodeLoc, Solo, Xinquji, and Eleduck for trying Mimi, sharing feedback, and spreading the word.
MIT © 2026 yuxino