Latest update — On-device AI Polish model (fully local)
A new on-device large model that polishes your dictation entirely on your Mac — nothing uploaded: smooths fillers, normalizes numbers (ITN), segments long speech, and fixes self-corrections. Auto-downloads on first enable. (Apple Silicon+ Windows)
本次更新 —— 本地润色大模型(完全本地化)
全新本地大模型润色,全程在本机运行、不联网不上传:顺滑去口癖、数字规整(ITN)、智能分段、改口纠正。首次开启 AI 润色自动下载模型。(Apple Silicon + Windows)
Vibe XASR 2.0 — new "AI Polish"
Speak, and an LLM tidies your dictation into clean text.
- ✨ AI Polish (cloud) — many OpenAI-compatible providers built in, or add your own; selecting one auto-fills the endpoint & model. Removes fillers, fixes restatements, normalizes numbers, applies hotwords; adjustable temperature / max length; save & switch multiple configs.
- 🔒 AI Polish (local · Beta, Apple Silicon only) — optional, fully offline, Metal-accelerated; the model downloads on first use.
- 📦 Smaller installer — the default model has been quantized & compressed; the macOS installer shrank from ~558 MB to ~165 MB with effectively no quality change.
- 🛠 Various bug fixes and HUD polish.
Universal2 — installs on both Apple Silicon and Intel; local polish is Apple Silicon only.
Vibe XASR 2.0 · 全新「AI 润色」
说完话,让大模型把口述自动整理成干净的文字。
- ✨ AI 润色(云端) —— 多家 OpenAI 兼容服务商开箱即用,也可自定义;选中即自动填好接口与模型。自动去口癖、改口纠正、数字规整、热词修正;温度 / 最大长度可调;多套配置一键保存切换。
- 🔒 AI 润色(本地 · Beta,仅 Apple Silicon) —— 可选完全离线,Metal 加速;模型首次启用在线下载。
- 📦 安装包大幅减小 —— 默认模型已做量化压缩,macOS 安装包从约 558 MB 降到约 165 MB,识别质量基本不变。
- 🛠 修复若干 bug,优化悬浮窗。
Universal2 —— Apple Silicon + Intel 均可安装;本地润色仅 Apple Silicon。
macOS is the primary development branch; the Windows build is compiled in sync and not yet fully tested (preview). Feedback welcome.
macOS 为主开发分支,Windows 为同步编译,尚未充分测试(预览版)。欢迎反馈。