🎙 AutoTranscribe

Features • Quick Start • Architecture • 中文说明

AutoTranscribe is a fully automated, offline audio/video transcription system for macOS. It monitors your Desktop and Downloads for new media files, prompts for confirmation, then automatically transcribes audio with speaker diarization — all running locally with zero cloud costs.

---

✨ Features

Feature	Description
🎯 Auto-Detection	Monitors ~/Desktop and ~/Downloads via macOS FSEvents (near-zero CPU)
🌐 Language Detection	Automatically detects Chinese, English, or mixed (en_cn) content
🗣️ Speaker Diarization	Identifies and labels different speakers (2–4 people)
⏱️ Timestamps	Sentence-level timestamps for every segment
📝 Markdown Output	Clean, readable .md files with metadata and speaker labels
🔔 Native Notifications	Stage-by-stage progress + result dialog via macOS Notification Center
🔄 Weekly Auto-Update	Automatically updates models and dependencies every Sunday
🚀 Boot on Startup	LaunchAgent ensures the service runs automatically
🔒 100% Offline	All processing happens locally — no data leaves your machine

🚀 Quick Start

Prerequisites

• macOS (Apple Silicon or Intel)
• Miniconda or Anaconda
• ffmpeg (brew install ffmpeg)

Installation

git clone https://github.com/YannJY02/AutoTranscribe.git
cd AutoTranscribe
bash install.sh

That's it! The installer will:

1. Create a conda environment (transcribe, Python 3.11)
2. Install all dependencies (FunASR, PyTorch, etc.)
3. Set up directory structure
4. Register LaunchAgents for auto-start and weekly updates

Usage

Just save an audio/video file to your Desktop or Downloads. A dialog will appear:

1. 📋 Confirm — Click "转录" to start, or "跳过" to skip
2. ⏳ Progress — Notification center shows 4 stages (extract → detect → transcribe → save)
3. ✅ Result — A popup shows full stats: language, duration, segments, speakers

Output

Transcription files are saved to txt/ with standardized names:

txt/2026_2_13_zh_1.md      # Chinese
txt/2026_2_13_en_1.md      # English
txt/2026_2_13_en_cn_1.md   # Mixed Chinese-English

Management

bash status.sh    # View service status
bash stop.sh      # Stop the service
bash start.sh     # Start the service

🏗️ Architecture

AutoTranscribe/
├── scripts/
│   ├── config.py          # Paths, model names, constants
│   ├── notifier.py        # macOS dialogs & notifications
│   ├── transcriber.py     # FunASR engine (LID + ASR + diarization)
│   ├── file_manager.py    # Naming, moving, Markdown generation
│   ├── watcher.py         # FSEvents file monitoring
│   ├── main.py            # Entry point & orchestration
│   └── update.py          # Weekly model & dependency updater
├── install.sh             # One-click installer
├── start.sh / stop.sh / status.sh
├── video/                 # (gitignored) Processed audio/video source files
├── txt/                   # (gitignored) Transcription output
└── logs/                  # (gitignored) Runtime logs

Models Used

Component	Model	Purpose
Language ID	SenseVoiceSmall	Detect zh / en / mixed
ASR	Paraformer-large	Speech-to-text (zh + en)
VAD	FSMN-VAD	Voice activity detection
Punctuation	CT-Transformer	Sentence segmentation
Speaker	CAM++	Speaker diarization

All models are from FunASR / ModelScope and are downloaded automatically on first use (~1–2 GB).

📄 License

MIT License — see LICENSE for details.

---

🧠 InsightKit (Mac Personal Meeting Assistant) — Preview

This repository now includes an InsightKit foundation for a native macOS meeting assistant with differentiated naming and architecture:

• UI modules: 会话总览 / 高光洞察 / 观点图谱 / 决策账本 / 执行清单 / 时间脉络
• Runtime: SwiftUI shell + Python JSON-RPC sidecar
• Schema: insightkit/schemas/insight_package_v1.json
• Storage: SQLite + FTS5 transcript indexing
• BYOK-ready: provider adapter abstraction for cloud insight generation

Run InsightKit sidecar

python3 scripts/insight_sidecar.py

Build macOS app shell (SwiftPM)

swift build --package-path macos/InsightKitApp

Package double-clickable macOS app (.app)

bash scripts/package_insightkit_app.sh --clean
open dist/macos/InsightKit.app

Sync app bundle on each iteration (rebuild + install)

bash scripts/sync_insightkit_app.sh

Default install path: ~/Applications/InsightKit.app
Optional custom path:

bash scripts/sync_insightkit_app.sh --install-dir /Applications

Default behavior of sync_insightkit_app.sh:

• fail-closed (tests/build/verify failure means no successful sync mark)
• clean build by default (--no-clean to disable)
• writes sync artifacts:
  • logs/workflow/sync_status.json (this run status)
  • logs/workflow/latest_sync.json (last successful sync)

Run gap-driven loop (auto package+install enabled by default)

python3 scripts/release_loop.py --max-rounds 1

Useful flags:

• --no-auto-package disable auto package/install for this run
• --install-dir /Applications override install target
• --package-debug use debug package mode
• --skip-sync-verify skip post-install verification (not recommended)

Export unloadable module for AttentionOS (/Users/yann.jy/Desktop/AI/RSS)

python3 scripts/export_attention_module.py --output dist/attentionos-insightkit-module

See integration guide:

• docs/insightkit-architecture.md
• docs/attentionos-integration.md

🎙 AutoTranscribe — 中文说明

AutoTranscribe 是一个全自动的本地音视频转录系统，专为 macOS 设计。它监控桌面和下载文件夹中的新音视频文件，弹窗确认后自动完成语音转文字和说话人分离，全程本地运行，零云端费用。

✨ 功能特点

• 🎯 自动检测 — 通过 macOS FSEvents 监控桌面和下载目录，待机 CPU 占用近零
• 🌐 语言识别 — 自动检测中文、英文或中英混合内容
• 🗣️ 说话人分离 — 自动识别并标注不同说话人（2-4 人）
• ⏱️ 时间戳 — 每句话都有精确的起止时间
• 📝 Markdown 输出 — 包含元信息、时间戳和说话人标签的清晰文档
• 🔔 原生通知 — 转录各阶段进度通知 + 完成结果弹窗
• 🔄 每周自动更新 — 每周日自动更新模型和 Python 依赖
• 🚀 开机自启 — LaunchAgent 保证服务随系统自动启动
• 🔒 完全离线 — 所有处理都在本地完成，数据不会上传

🚀 快速开始

前置条件

• macOS（Apple Silicon 或 Intel）
• Miniconda 或 Anaconda
• ffmpeg（brew install ffmpeg）

安装

git clone https://github.com/YannJY02/AutoTranscribe.git
cd AutoTranscribe
bash install.sh

安装脚本会自动完成所有配置：创建 conda 环境、安装依赖、建立目录结构、注册开机自启。

使用方法

只需将音频或视频保存到桌面或下载文件夹，系统会自动弹窗提示：

1. 📋 确认 — 点击「转录」开始，或「跳过」忽略
2. ⏳ 进度 — 通知中心分 4 个阶段显示进度（提取音频 → 检测语言 → 转录 → 保存）
3. ✅ 结果 — 弹窗显示完整统计：语言、时长、耗时、片段数、说话人数

输出格式

转录文稿保存在 txt/ 目录，文件名标准化：

txt/2026_2_13_zh_1.md       # 中文
txt/2026_2_13_en_1.md       # 英文
txt/2026_2_13_en_cn_1.md    # 中英混合

管理命令

bash status.sh    # 查看服务状态
bash stop.sh      # 停止服务
bash start.sh     # 启动服务

📄 许可证

MIT 开源协议