Pepys Dictate for Mac
Open-source licences
Pepys Dictate is built on open-source software and AI models. These are the projects it includes and the licences they come under. The same notices ship inside the app.
Software
transcribe.cpp and ggml
MITSpeech inference. transcribe.cpp by handy-computer; ggml by Georgi Gerganov and contributors.
llama.cpp
MITRuns the formatting model on your Mac. By Georgi Gerganov and contributors.
AI models
- CC BY 4.0
Speech recognition model. Attribution: NVIDIA.
Qwen3.5 and Pepys fine-tunes
Apache 2.0Local formatting models: Qwen3.5 by Alibaba Cloud (Qwen team), and Pepys's own fine-tunes of it, released under the same licence.
Speaker models
MIT, Apache 2.0Speaker labels use two small ONNX models downloaded on demand: the pyannote segmentation model (pyannote-audio 3.0, MIT, by Hervé Bredin and contributors) and a WeSpeaker ResNet34 voice-embedding model (WeSpeaker, Apache-2.0), in the ONNX exports published by k2-fsa/sherpa-onnx (Apache-2.0). The grouping of voices into speakers is Pepys' own code.
Other optional speech models are downloaded on request, and the in-app model list shows each one's licence.