Licenses and technology
The apps are copyright Yuki Otsue. They use the models and software below. Full copyright notices and license texts are in the app under Acknowledgements & Licenses.
Speech recognition (both apps)
- Haas-ja-v1.0 (third-intelligence, Apache License 2.0) — Japanese speech recognition model, converted to the MLX format and partly quantized to 8 bits.
- Silero VAD v5 (Silero Team, MIT License) — finds the parts of the audio that contain speech.
- Qwen3-ASR 0.6B (Alibaba Qwen, Apache License 2.0) — English speech recognition model, using the 8-bit MLX version published by mlx-community. Caption YOYU Plus only.
Translation (Caption YOYU Plus) — Built with PLaMo
Caption YOYU Plus uses PLaMo 3 Translate by Preferred Networks, Inc., provided under the PLaMo Community License. The app includes “PLaMo 3 Translate 0.8B MLX-4bit compact”, a derived model of PFN’s PLaMo 3 Translate 0.8B MLX-4bit with the embedding and output layers also quantized to 4 bits. Caption YOYU does not include it.
- The first time you use translation, the app asks you to agree to the PLaMo Community License. The full text (Japanese and English) is shown on that screen. One agreement covers translation in both directions.
- The license may be agreed to by people aged 18 or over (or old enough to enter a contract alone in their country). Younger users need the consent of a parent or guardian (Article 2).
- Translations made by the app are output of PLaMo 3 Translate. When you give exported translations to others or publish them, tell the recipients that AI models trained on those translations are subject to the PLaMo Community License (Article 5(4)). Plain-text exports include a sentence saying so, worded for the language of the translation (English or Japanese).
Full text: English / 日本語 (revised November 9, 2026)
Software
- MLX, mlx-swift, mlx-swift-lm, mlx-audio-swift (MIT License)
- swift-transformers and other Swift packages from Hugging Face and Apple (Apache License 2.0)
- Sparkle (MIT License) — checks for and installs updates in Caption YOYU Plus
Parts of the speech engine follow implementations in NVIDIA NeMo, parakeet-mlx, flash-linear-attention, Haas-ASR and speech.cpp.