Live Sub: Local Translator

Live Sub turns your iPad into a live captioning and translation board for face-to-face conversations and meetings. Everyone sees the original speech and its translation side by side.

KEY FEATURES
• Fully on-device recognition: speech recognition runs locally on your iPad. Audio stays in memory only — it is never uploaded or written to disk.
• Auto language detection: Chinese, Cantonese, English, Japanese and Korean are recognised sentence by sentence — no switching needed.
• Adaptive translation: each sentence is translated into the other language that was spoken most recently, so the last two languages in the room translate into each other. Or pin one target language.
• Side-by-side reading: original on the left, translation on the right, with a font-size slider (18–72 pt) built for reading across a room.
• Replay and read aloud: long-press any caption to replay its recording or hear the original or translation read aloud. Recordings stay in memory only, and you can turn them off in Settings.
• AI analysis: tick any of intent analysis, term explanations, background and logical-fallacy analysis for each sentence. Answers come in the translation language, next to the translation or on their own. Uses the on-device model by default.
• Meeting mind map: after each finished sentence, the app updates a mind map of topics, decisions, to-dos and open questions. Choose a small on-device language model so nothing leaves your iPad, or an optional cloud model with your own API key (Zhipu BigModel, OpenRouter or any OpenAI-compatible service). Nodes added or changed by each update are highlighted, "Tidy" reorganises the whole map, and later sentences keep flowing into the tidied map. Pick the mind map's output language.
• Model manager: see the speech model, translation language packs and on-device language models, with the languages each supports. Download or delete on-device models to manage storage.
• Your language: the interface is available in Simplified Chinese, Traditional Chinese, English, Japanese and Korean — switch any time in Settings.
• Privacy by design: no account, no analytics, no tracking. The cloud option is off by default and only sends the text of finished sentences — never audio — after you turn it on.

NOTES
• Requires iPadOS 26 or later.
• Translation uses Apple's on-device Translation. Download each language pack once; it then works offline.
• The optional on-device mind-map model (about 0.5–1.2 GB) is downloaded on demand and runs on the CPU, so each update takes a few seconds to half a minute depending on your iPad.
• Recognition and translation quality depend on microphone distance, background noise and overlapping speakers.