Acknowledgements

MeetCap is built on the open-source models and software below. Thank you to their authors. The full licence texts are in the third-party notices.

Models

Apart from the local model, these models download to your computer during setup and run only on it.

• X-ASR offline recognition model: transcribes the whole meeting afterward, and recordings you already have. From Shanghai Jiao Tong University, Shanghai Innovation Institute, Fudan University and Huazhong University of Science and Technology, converted by the sherpa-onnx project. Apache-2.0.
• X-ASR streaming recognition model: live captions during the meeting. Same source. Apache-2.0.
• Silero VAD: detects when someone is speaking. From the Silero Team. MIT.
• Nemotron-3 Diarization: tells speakers apart. From NVIDIA, converted by onnx-community. OpenMDW-1.1.
• CAM++ speaker model: recognizes people you’ve named. From the 3D-Speaker project. Apache-2.0.
• Gemma 4 E4B: the local model, which writes summaries and minutes on your computer. Downloaded only if you choose it. From Google. Apache-2.0.

Software

• sherpa-onnx: runs the speech recognition, voice detection and speaker models. Apache-2.0.
• ONNX Runtime: runs models. MIT.
• llama.cpp: runs the local model. MIT.
• WebRTC audio processing: removes echo when you use speakers, so the other side’s words appear only once. BSD-3-Clause.
• parakeet-rs: inference code for telling speakers apart. MIT.
• pinyin-data, pinyin-pro: pinyin for Chinese names. MIT.

Every other open-source library MeetCap uses is listed in the third-party notices.