Acknowledgements
transcribe_cpp is a Flutter package around other people’s work. It would not exist without them.
transcribe.cpp
Section titled “transcribe.cpp”transcribe.cpp, by the transcribe.cpp authors, is the speech engine: it loads the models and turns audio into text. This package ships its builds for iOS and Android, generates its model list from transcribe.cpp’s catalogue, and downloads the models it converts and publishes on Hugging Face. The word error rates in these docs come from that catalogue too.
transcribe_cpp is an independent project, not an official part of transcribe.cpp.
ggml is the machine-learning library transcribe.cpp runs on, including the Metal code that runs models on the iPhone’s GPU.
Silero VAD
Section titled “Silero VAD”Silero VAD decides when someone is speaking. Its model was ported to C for this package, and its weights ship with it.
The models
Section titled “The models”The speech models are the work of their authors, among them
NVIDIA (Parakeet) and the
Moonshine team.
Each model has its own licence; some require apps to credit the authors, and
some are non-commercial. Check ModelDescriptor.license for each one you
ship.
Licences
Section titled “Licences”transcribe_cpp is MIT-licensed. The licences of everything it includes are in THIRD-PARTY-LICENSES.md.