TL;DR: Cicada hooks directly into system audio to process speech locally on Apple silicon or via configured API endpoints on other platforms. It uses speech-to-text models to generate bilingual transcripts and structured summaries on the fly, while routing voice output through a virtual audio driver when interpreting.
Best for: Developers and remote workers who spend hours on international calls and livestreams.
Repository snapshot: ⭐ 2 143 stars · 🍴 370 forks · Swift · Apache-2.0
Daily picks: Follow GitHub Open Source on Telegram
What it is
Cicada is an open-source Swift application designed to provide real-time speech recognition, live bilingual subtitles, and structured meeting notes directly from your system audio. It eliminates the need for third-party recording bots in video conferencing software.
How it works
The application intercepts system audio streams locally. On Apple silicon Macs, it leverages on-device speech models to process transcription completely offline. On other operating systems, it routes audio through configured endpoints and can broadcast interpreted voice output using a virtual audio driver.
Key features
- Captures system audio without injecting bots into Zoom, Teams, or Google Meet
- Runs offline speech recognition natively on Apple silicon hardware
- Translates speech and broadcasts interpreted voice output in target languages
- Automatically organizes meeting transcripts into summaries, topics, and to-dos
- Separates speakers using local voiceprint models with local data retention
Getting started
Clone the repository from GitHub and ensure you have Swift installed. Build the project locally for your target operating system. Configure external API keys if you are running the software on non-Apple platforms.
When not to use it
Avoid this tool if you require out-of-the-box zero-configuration setup on Windows or Linux, as non-Mac platforms currently require manual API key configuration for speech processing.
Discuss
Which meeting platform do you use most often?
Curated from the repository README and metadata. No hands-on test is claimed; evaluate it in a sandbox before production use.
Top comments (0)