VladUZH 开源了 Qwen Scribe，一款专为 Apple Silicon Mac 设计的本地语音转录和系统级听写工具 [1]。该项目采用 Qwen3-ASR 模型，通过 MLX 框架在 Mac 的 Metal GPU 上运行，无需依赖云服务或 API 密钥，所有音频和转录文本均保存在本地 [1]。
项目目前已发布 v0.1.0-beta.1 版本 [1]。Qwen3-ASR 1.7B 模型需约 3.4 GB 统一内存，0.6B 模型需约 1.2 GB，系统要求为 macOS 14 或更新版本 [1]。工具支持通过拖放导入音频和视频文件进行转录、自动语言检测、SRT 字幕导出，以及使用右 Command 键快速启动听写功能 [1]。该项目采用 Apache-2.0 许可证，源代码和问题追踪位于 VladUZH/qwen-scribe 仓库 [1]。开发者计划在 v0.2 版本中发布签名和公证的二进制文件 [1]。
VladUZH has open-sourced Qwen Scribe, a local speech-to-text and system-level dictation tool designed specifically for Apple Silicon Macs [1]. The project leverages the Qwen3-ASR model running on Mac Metal GPU through MLX, eliminating the need for cloud services or API keys while keeping all audio and transcribed text stored locally [1].
The v0.1.0-beta.1 release is now available with source code accessible, and the developer plans to distribute signed and notarized binaries in v0.2 [1]. The application requires macOS 14 or later and supports multiple features including drag-and-drop transcription for audio and video files, automatic language detection, SRT export, and quick dictation via the right Command key [1]. The Qwen3-ASR 1.7B model requires approximately 3.4 GB of unified memory, while the 0.6B model needs about 1.2 GB [1]. The project is distributed under the Apache-2.0 license, with source code and issue tracking available at VladUZH/qwen-scribe [1].