VoiceMeeting turns long conversations into searchable transcripts and polished meeting notes, while keeping the speech recognition work on your own Mac. Record a meeting, import an audio file, review the timeline, clean up the transcript, and generate Markdown notes without sending raw audio to a cloud ASR service.
Version
0.0.2is a macOS-first release. Windows and Linux are not packaged yet.
- Private by design: speech-to-text runs locally with Whisper-family models, including Apple Silicon MLX models on Mac.
- Built for real meetings: timeline playback, speaker-aware transcript review, editable copies, re-transcription, proofreading, and paragraph cleanup are part of the same workflow.
- No mystery pipeline: downloaded models, transcript versions, notes, and audio chunks are visible and managed locally.
- Bring your own assistant: meeting notes can use VibeAround or a configurable LLM API connection.
- Bilingual interface: switch between Chinese and English from the top bar.
- Record from the microphone or import an existing audio/video file.
- Use local faster-whisper or Mac MLX Whisper models.
- Preload the selected ASR model before recording so the first transcription does not appear stuck.
- Generate transcript versions for re-recognition, speaker recalibration, text proofreading, paragraph organization, and manual editing.
- Play back audio on a visual timeline and jump directly from transcript segments.
- Generate, stream, render, and download Markdown meeting notes.
- Save transcripts and notes through native macOS save dialogs in the desktop app.
- Manage local model downloads with progress and install/delete controls.
The first public build is macOS only:
- Apple Silicon:
VoiceMeeting_0.0.2_aarch64.dmg
The app is signed and notarized for macOS distribution.
Requires Python 3.10 or newer. High-precision speaker separation uses pyannote.audio 4.x, which does not run on Python 3.9. If an existing .venv was created with Python 3.9, move or remove it before rerunning setup.
./scripts/setup.sh
./scripts/dev.shOpen http://127.0.0.1:5199.
bun run build:desktop
./scripts/sign-and-notarize-macos.shThe notarized DMG is written to:
tauri/src-tauri/target/release/bundle/dmg/
Signing credentials live in apple-sign.config, which is intentionally ignored by git.
VoiceMeeting supports two note-generation routes:
- VibeAround: local channel for the meeting assistant.
- LLM API: configurable model access. In Settings, choose a provider preset, enter the model name, API key, and optional API Base URL.
VoiceMeeting 是一个面向会议录音和音频转写的本地桌面应用：录音、导入音频、查看时间线、整理逐字稿、生成 Markdown 会议纪要，都在一个界面里完成。语音识别默认在本机运行，不需要把原始音频交给云端 ASR。
0.0.2是 macOS 首发版本，暂时只提供 Mac 安装包。
- 隐私优先：语音识别在本机完成，Mac 上可使用 Apple Silicon 友好的 MLX Whisper 模型。
- 适合真实会议：音频时间线、逐字稿回放跳转、说话人校准、自动校对、段落整理、可编辑副本都在同一套流程里。
- 流程透明：模型文件、转写版本、音频片段和纪要都保存在本地，可在设置里管理。
- 可接入自己的助手：纪要生成支持 VibeAround，也支持通过模型接口接入各家大模型 API。
- 中英文界面：顶栏可一键切换中文和英文。
需要 Python 3.10 或更新版本。高精度说话人分离依赖 pyannote.audio 4.x，不能在 Python 3.9 环境运行。如果已有 .venv 是 Python 3.9 创建的，请先移动或删除后再重新运行安装脚本。
./scripts/setup.sh
./scripts/dev.sh然后打开 http://127.0.0.1:5199。
bun run build:desktop
./scripts/sign-and-notarize-macos.sh生成的 DMG 位于：
tauri/src-tauri/target/release/bundle/dmg/
如果不用 VibeAround，可以在设置里配置“LLM 模型接口”：在界面里选择服务商预设，填写模型名、API Key 和可选 API Base URL 即可。常见模型名示例包括 openai/gpt-4o-mini、anthropic/claude-sonnet-4-5、openrouter/openai/gpt-4o-mini、deepseek/deepseek-chat；OpenAI 兼容服务一般使用 openai/your-model 并填写对应的 API Base。