Skip to content

OmaScribe 0.3.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 24 Oct 09:26
v0.3.0
c4aa660

OmaScribe 0.3.0

正式版 / Stable release. Offline recording, bilingual transcription and editable meeting notes for macOS, Windows and Linux.

本次更新 / Changes

  • 会议列表、笔记、纪要和转写可独立收合；统一侧栏图标、窄窗口布局与设置导航。
  • 外观、字体、资源预算、模型管理、设备检测和配置导入/导出。模型下载逐文件校验，独立保存，卸载应用不会删除用户模型。
  • Apple Silicon 默认使用可用的 Metal 加速；CPU 后端仍可选择。低内存时暂停整理并保留录音与转写，任务结束后卸载模型。修复 macOS 可用内存误判。
  • 笔记、知识库、专有名词表与手动记忆；背景资料需明确启用，不作为会议事实出处。
  • 可选本地说话人区分、手动命名与片段调整；改名保留历史并标记相关纪要待复核。说话人模型需先手动下载。
  • 配置文件与官网配置链接均先预览后应用，不包含录音、笔记、密钥、路径或联网许可。
  • 官网恢复 macOS、Windows、Linux 图标；正式版本、安装包大小和下载地址随 GitHub Release 自动同步。

Collapsible workspace panels, appearance and model settings, hardware recommendations, local library, optional speaker clustering and safe configuration sharing. Apple Silicon uses Metal when available; CPU remains supported. Recording, transcription and AI notes stay on this computer. Network access is limited to explicitly enabled model downloads and configuration browsing.

安装 / Install

Computer Package
macOS 11+, Apple Silicon OmaScribe_0.3.0_macOS-AppleSilicon.dmg
macOS 11+, Intel AVX2 OmaScribe_0.3.0_macOS-Intel.dmg
Windows 10/11, x64 AVX2 OmaScribe_0.3.0_Windows-x64.exe
Linux x64 AVX2, Ubuntu 22.04+ or compatible OmaScribe_0.3.0_Linux-x64.AppImage / .deb

所有安装包包含 Zipformer Small + Qwen3 1.7B Q4_K_M 与本地运行环境。Windows 包含离线 WebView2 安装器。校验值见 SHA256SUMS.txt，各平台验证报告附于 Release。

Each installer includes the default models and native inference runtime. No cloud AI account is required. Installers are published together only after tests on the matching GitHub runners pass.

已知限制 / Limits

  • macOS uses ad-hoc signing and is not notarized; Windows has no publisher certificate. These are distribution limits, separate from the stable release channel. See installation instructions.
  • WAV import is built in. Other audio formats require FFmpeg on PATH. Linux microphone recording requires an active PipeWire session and pipewire-bin; .deb declares these dependencies. AppImage users install them through their distribution.
  • MLX is an optional experimental backend with a separately installed local Python runtime. It is not bundled or installed automatically. Apple Silicon defaults to Metal, not MLX.
  • Speaker separation is post-recording clustering, not identity recognition; two hours maximum per analysis. Review ambiguous voices and all AI notes against the source recording.
  • System-audio capture/mixing is not included. Cross-meeting retrieval uses lexical matching, not semantic embeddings. Physical microphone permissions, multi-monitor overlays and long-session thermal behavior still require checks on the user's hardware.
  • User data is local but not encrypted by the app. Application upgrades retain the separate user-data directory. Back up important meetings before upgrading.
Assets 17
Loading