Notes from the people building SyncTrans, the real-time speech translator for macOS — what shipped, what broke and how we fixed it, and the tradeoffs we argued about along the way. Written to be accurate first and flattering second.
Looking for user-facing release notes instead? See What's New →
Version 1.0.0 is tagged and nothing newer has shipped — but the repository has not been idle. An honest look at the two lines of work currently in development: a hybrid engine that lets you mix recognition, translation and speech sources per session, and a round of local-mode hardening around model downloads and honest capability labels. None of it is in a release yet.
Read the postA walkthrough of one real workflow in SyncTrans 1.0.0: the five-step first-run setup, picking the cloud engine with guest credit, setting a Chinese–English pair with two speakers on one microphone, running the conversation, and walking out with an exported Markdown record. Including the parts that go wrong and where the app tells you about them.
Read the postChinese discount phrasing counts what you pay; Japanese counts what comes off — and a literal translation of 七折 inverts the price. Why every model size we tested fell into this trap, why we added a deterministic rule layer as a backstop for numbers, and what that hybrid approach costs us.
Also in: 中文 · 日本語 · Español · Français · 한국어 · Українська
Read the postSyncTrans 1.0.0 is out. The engineering account of how it got there: two engine families developed in parallel, a conversation view designed for face-to-face use, three layouts over one session record, and a seven-language interface — plus the tradeoffs we accepted to ship them together.
Also in: 中文 · 日本語 · Español · Français · 한국어 · Українська
Read the post