No Need to Put a Bot in Your Meetings: This Open-Source Cicada on Mac Does Simultaneous Interpretation Directly

Cicada, whose Chinese name is 知了 (cicada), is a native macOS app. It doesn't send a bot into meetings, nor does it require the other party to install a plug-in. It directly reads the system audio currently playing on the Mac, converts the audio to text locally, then produces bilingual subtitles, real-time notes, and simultaneous interpretation. When you need to speak, hold down the right ⌥ key and talk; the translated speech is sent into the meeting app through the virtual microphone BlackHole 2ch, so the other party hears the translated voice.

This approach bypasses the subtitle systems of meeting apps. Built-in subtitles in Feishu and Teams usually only work inside their own apps; with X Spaces, YouTube livestreams, Discord, or podcasts, they are useless, and often the host has to enable them or a paid version is required. Cicada listens directly to the entire Mac's audio, so as long as the computer can play sound, it works in scenarios such as Zoom, Tencent Meeting, Feishu, Google Meet, and Bilibili livestreams. The other party doesn't need to install anything, and no bot joins the meeting.

Real-time subtitles automatically detect the language locally and translate it into the system language within seconds. Press ⌃⌥S to switch between the notes window, the subtitle bar, and no display. The subtitle bar appears only when someone is speaking; when the mouse passes over it, it automatically becomes transparent, so it won't block the buttons underneath.

After you fill in a Ling key, the notes window listens while organizing summaries, topics, conclusions, to-dos, fishbone diagrams, and a glossary. The author says it updates every 45 seconds. Each conclusion and to-do is marked with the source of the original words and the time; click one to jump back to the bilingual record. Names and terms are normalized, and misheard variants can also be found with ⌘F.

Simultaneous interpretation has more than 10 built-in voices, and you can also clone your own voice with an 8-to-30-second recording. Without a key, you can still use the Mac's built-in voices, but you can't clone. The first time you open interpreting, it will guide you through installing BlackHole.

The bilingual transcript, notes, and original recording of each meeting are saved locally. The meeting automatically ends when the meeting app hangs up or no one speaks for 10 minutes, and blank sections are trimmed. After the meeting, you can generate notes afterward and export Markdown, Word, and SRT. Speaker recognition is an experimental feature and is off by default; when enabled, it can distinguish different speakers, and with a Ling key you can also fill in names. Currently, "you" in the record only includes what is said while holding the speak key; parts spoken directly in the meeting are not yet recorded.

Audio capture, language recognition, speech recognition, and recording are done locally. The only things that leave the computer are the recognized text and the few utterances spoken while holding the speak key during interpretation. Translation connects to Google; after you fill in a key, it connects to Ling and Ming. The app has no analytics tracking and no self-built servers. When sharing your screen, the other party cannot see the notes window by default.

Cicada is open source under Apache-2.0; you can modify it for personal use, use it commercially, or republish it closed-source. Without a key, real-time subtitles, translation, meeting records, and simultaneous interpretation are available. Real-time notes, one-step speech translation, and voice cloning require your own Ling API key. The README says an hour-long meeting costs about 0.15 to 1 yuan, Ling gives 10 yuan per month, which is about enough for 10 to 50 hours; the author's tweet says 50 hours per month if used sparingly. In replies, the author mentioned that new users get 20 yuan upon registration, and then 10 yuan on the 1st of each month; the platform's terms prevail.

The requirements are also clear: macOS 26 or later and an Apple silicon Mac. Intel Macs are not currently supported. The app is currently locally signed and not yet notarized by Apple; the first time you open it, you need to go to System Settings > Privacy & Security and click "Open Anyway". Permissions are requested by feature: system audio permission is used for subtitles; the microphone records only while you hold the speak key; input monitoring is only used to determine whether the right ⌥ is pressed; installing BlackHole requires an admin password once.

The usage steps are not complicated:

1. Download the DMG from GitHub Releases or the official website, drag it into Applications, and allow it the first time you open it.
2. Go to the Ling open platform to register, create a token, and copy the key starting with sk-.
3. Open Cicada, follow the first-run guide to grant system audio permission, and paste the key. If you want to use simultaneous interpretation, then install BlackHole.

The roadmap puts these first: recording my speech, meaning that words spoken directly without pressing the translation key also go into the notes; support for macOS 14.4 and later; official signing and notarization; voiceprint recognition; notes without a key; and integration with other services compatible with the OpenAI API. Windows and Android are further out; iPhone and iPad are not currently planned due to system restrictions.

Its trade-offs are clear: by bypassing the meeting-app ecosystem, it gains invisibility to the other party and cross-app availability; the cost is the system version, chip, and a one-time manual approval. People who frequently hold cross-language meetings or listen to livestreams and don't want to deal with meeting plug-ins should install it and try it. Those still on an old macOS or Intel Mac can only wait for the roadmap for now.

Open-source address: https://github.com/wquguru/cicada

Official website: https://cicada.indaloop.app

Get started: https://cicada.indaloop.app/start/