MyDictionary

Local AI Dictionary · Translation · Synonyms · Examples · AI-Powered Academic Writing

English | 中文

---

English

🏆 Core Technology Stack

State-of-the-art AI models running 100% locally in your browser:

Component	Technology	Highlights	Performance
🧠 AI Framework	Transformers.js v2.17	Official HuggingFace browser runtime · WASM + WebGL acceleration · Zero backend dependency	⚡ Native speed
🌐 Translation Engine	Meta NLLB-200	Unified EN↔CN model (600MB shared) · Universal 200-language support · SOTA translation quality	🏆 BLEU 40+
🎓 Academic Search	BGE-Base-EN-v1.5	BAAI General Embedding · MTEB Top 5 (84.7% accuracy) · 768-dim semantic vectors	🥇 MTEB Rank #5
📚 Academic Database	Manchester Phrasebank	2,500+ curated phrases · 5 paper sections · University-verified expressions	✅ Academic-grade
🔍 Synonym Engine	DistilBERT-base-uncased	Lightweight BERT variant · Context-aware recommendations · 65MB optimized model	🚀 Fast inference
💬 Example Generator	MiniLM-L6-v2	Sentence transformers · Semantic similarity matching · 23MB ultra-light	⚡ <100ms
📖 Local Dictionary	ECDICT	770k+ entries · MIT License · Phonetics + Collins ratings · Instant lookup (<50ms)	⚡ 60x faster
🔊 TTS Engine	SpeechT5 ONNX	Microsoft SpeechT5 · Browser-native TTS · 120MB model · WebGPU accelerated	🎵 Real-time
🎙️ TTS Server (Optional)	Kokoro-82M	TTS Arena #1 · 82M params · Rust Candle · Port 9527	🏆 Native quality

✨ Features

• 🔒 100% Privacy - All AI processing runs locally in browser, zero data upload, no API keys needed
• 📖 Lightning-Fast Dictionary - 7,400+ high-frequency words with instant lookup (<50ms), 60x faster than AI models
• 🌐 Smart Translation - 200 languages with specialized high-quality models for major pairs (EN↔CN)
• 🧠 AI Semantic Search - BGE-powered intelligent phrase discovery with similarity scoring
• 📚 Synonym Suggestions - Context-aware intelligent recommendations powered by DistilBERT
• 💬 Example Sentences - Real-world usage examples from authentic sources
• 🎓 Academic Writing - 2,500+ academic phrases + AI semantic search for research papers
• 🔊 Text-to-Speech - Browser-native TTS (SpeechT5) + optional high-quality server (Kokoro-82M)
• ⚡ On-Demand Download - Only 600MB by default (EN↔CN shared model), other models downloaded as needed
• 🎯 Hardware Detection - Auto-recommends optimal models based on your device capabilities

🚀 Quick Start

Installation

1. Clone the repository (with submodules):

# Clone with TTS server submodule
git clone --recurse-submodules https://github.com/yourusername/MyDictionary.git
cd MyDictionary

# Or if already cloned, initialize submodules
git submodule update --init --recursive

2. Install dependencies:

pnpm install

3. Build the extension:

pnpm run build

4. Load in Chrome:
   • Open Chrome and navigate to chrome://extensions/
   • Enable "Developer mode"
   • Click "Load unpacked" and select the dist/ directory

Optional: Install High-Quality TTS Server

For native-speaker quality TTS (Kokoro-82M, ranked #1 on TTS Arena):

cd tts-server
git checkout dev
cargo build --release

# Start server (port 9527)
./target/release/kokoro-tts-server

See tts-server/README.md for detailed instructions.

Usage

Method 1: Text Selection

1. Select text on any webpage
2. Use right-click menu or translation icon
3. Sidebar slides in from the right with translation

Method 2: Keyboard Shortcut

1. Select text + Ctrl+Shift+D (Mac: Cmd+Shift+D)
2. Sidebar appears automatically

Method 3: Click Extension Icon

1. Click the extension icon in toolbar
2. Enter text manually in the sidebar

📦 Models

Default Installation (600MB)

• ✅ English ↔ Chinese - Meta NLLB-200-distilled-600M (600MB, shared for both directions)

On-Demand Downloads

Models are downloaded automatically when you use the corresponding features:

Feature	Model	Size	Download Trigger	Technology
Other Languages	Meta NLLB-200 (Universal)	600MB	First non-EN/CN translation	NLLB distilled (same as EN↔CN)
Synonyms	DistilBERT-base-uncased	65MB	Click [Synonyms] button	DistilBERT
Examples	MiniLM-L6-v2	23MB	Click [Examples] button	Sentence-Transformers
Academic Phrases	Phrasebank JSON	1.1MB	Switch to Academic mode	JSON database
🧠 AI Semantic Search	BGE-Base-EN-v1.5	270MB	Click "Semantic Search" tab	BAAI Embeddings
AI Semantic (Lite)	BGE-Small-EN-v1.5	130MB	Low-end devices	BAAI Embeddings
🔊 TTS (Browser)	SpeechT5 ONNX	120MB	Click 🔊 button	Microsoft SpeechT5
🔊 TTS (High Quality)	Kokoro-82M	90MB	Install local server (optional)	Rust Candle

Total: 600MB (default) → 870MB (with AI semantic search) → 990MB (with TTS) → 1.8GB (full installation)

💡 High-Quality TTS: Install optional Candle TTS Server for native-speaker quality (Kokoro-82M, TTS Arena #1)

🎯 Core Functions

1. Smart Translation

• Auto-detects 20+ languages (Chinese, English, Japanese, Korean, Thai, Russian, Arabic, etc.)
• Main language pairs (EN↔CN) use dedicated high-quality models ⭐⭐⭐⭐⭐
• Other languages use universal model supporting 200 languages ⭐⭐⭐⭐

2. Synonym Suggestions

• Context-aware intelligent synonyms
• 5-10 relevant replacement suggestions
• Click to re-translate with selected synonym

3. Example Sentences

• 3-5 real-world usage examples
• Auto-translated examples
• Highlighted target vocabulary

4. AI-Powered Academic Writing

• 2,500+ Academic Phrases: Curated from University of Manchester Academic Phrasebank
• 🧠 AI Semantic Search: BGE-powered intelligent phrase discovery with similarity scoring (50-100%)
• Paper Section Specific: Introduction / Methods / Results / Discussion / Conclusion
• Dual Search Modes: Keyword search (instant) + AI semantic search (intelligent)
• Copy & Paste: One-click copy to your paper
• Performance Detection: Auto-recommends BGE-Base (high-end) or BGE-Small (efficient)

5. Text-to-Speech (TTS) 🔊

• Browser-Native: SpeechT5 TTS runs 100% in browser (120MB ONNX model)
• Smart Loading: Lazy download on first use, permanently cached offline
• Academic Phrases: Click 🔊 button next to any phrase to hear native pronunciation
• Button States: Visual feedback (🔊 → ⏳ → ⏸️)
• Optional High-Quality: Install Kokoro-82M server for native-speaker quality
  • TTS Arena #1: Ranked higher than OpenAI TTS and XTTS v2
  • Lightweight: Only 82M parameters, runs on CPU
  • Rust Candle: Fast inference with minimal memory
  • Port 9527: Local server, no internet required

🛠️ Tech Stack

• Frontend: Chrome Extension Manifest V3
• AI Library: Transformers.js (@xenova/transformers)
• Translation Models: Meta NLLB-200-distilled-600M (EN↔CN) + NLLB-200 (Universal)
• Academic Models: SciBERT + Academic Phrasebank
• Model Download: Hugging Face Hub (official recommended method)

📖 Documentation

• Product Design
• Technical Architecture
• UI Wireframes
• Model Strategy
• Academic Writing Models
• Academic Version Plan
• Product Summary

🎨 Logo Design

MyDictionary's logo features a Raccoon 🦝 holding a dictionary, symbolizing:

• 🧠 Smart & Clever - Local AI-powered intelligent translation
• 📚 Well-Read - Supporting 200 languages
• 🎓 Academic Excellence - Professional academic writing assistance

🔧 Development

Project Structure

MyDictionary/
├── manifest.json           # Chrome Extension config
├── package.json            # Dependencies
├── background.js           # Service Worker (model management)
├── content.js              # Content Script (UI)
├── src/
│   ├── config/
│   │   └── models-config.json
│   ├── utils/
│   └── ui/
│       ├── sidebar.html
│       ├── sidebar.css
│       ├── popup.html
│       └── popup.js
├── assets/
│   ├── logo.png
│   └── icons/
└── docs/

Development Commands

# Install dependencies
pnpm install

# Development mode (watch file changes)
pnpm run dev

# Build for production
pnpm run build

# Package as .crx
pnpm run package

🤝 Contributing

Issues and Pull Requests are welcome!

📄 License

MIT License - see LICENSE file

🙏 Acknowledgements

AI Models & Libraries:

• Transformers.js by Hugging Face - Run Transformers models in the browser with ONNX Runtime
• Xenova/nllb-200-distilled-600M - English ↔ Chinese translation (Meta NLLB-200, quality: 8/10)
• facebook/nllb-200-distilled-600M - 200 languages universal translation (Meta AI)
• BAAI/bge-base-en-v1.5 - BGE embeddings for AI semantic search (MTEB Top 5)
• BAAI/bge-small-en-v1.5 - Lightweight BGE for low-end devices
• distilbert-base-uncased - Synonym suggestions
• sentence-transformers/all-MiniLM-L6-v2 - Sentence embeddings
• Academic Phrasebank by University of Manchester - 2,500+ curated academic phrases

---

With the wisdom of 🦝, translation made simple

Made with ❤️ by Jason

---

中文

🏆 核心技术栈

业界领先的 AI 模型，100% 浏览器本地运行：

组件	技术	亮点	性能
🧠 AI 框架	Transformers.js v2.17	HuggingFace 官方浏览器运行时 · WASM + WebGL 加速 · 零后端依赖	⚡ 原生速度
🌐 翻译引擎	Meta NLLB-200	统一英中模型 (600MB共享) · 通用200语言支持 · SOTA 翻译质量	🏆 BLEU 40+
🎓 学术搜索	BGE-Base-EN-v1.5	北京智源 BGE 嵌入 · MTEB 排名第5 (84.7%准确度) · 768维语义向量	🥇 MTEB #5
📚 学术数据库	曼彻斯特大学学术短语库	2500+精选短语 · 5个论文章节 · 大学认证表达	✅ 学术级
🔍 同义词引擎	DistilBERT-base-uncased	轻量BERT变体 · 上下文感知推荐 · 65MB优化模型	🚀 快速推理
💬 例句生成器	MiniLM-L6-v2	句子转换器 · 语义相似度匹配 · 23MB超轻量	⚡ <100ms

✨ 特性

• 🔒 100% 隐私 - 所有 AI 推理在浏览器本地完成,零数据上传,无需 API 密钥
• 🌐 智能翻译 - 200 种语言,主要语言对 (英↔中) 使用专用高质量模型
• 🧠 AI 语义搜索 - BGE 驱动的智能短语发现,带相似度评分
• 📚 近义词推荐 - DistilBERT 驱动的上下文感知智能建议
• 💬 例句展示 - 来自真实来源的使用场景例句
• 🎓 学术写作 - 2,500+ 学术短语 + AI 语义搜索助力论文写作
• ⚡ 按需下载 - 默认仅 600MB (英中双向共享模型),其他模型按需下载
• 🎯 硬件检测 - 根据设备性能自动推荐最优模型

🚀 快速开始

安装

1. 克隆仓库:

git clone https://github.com/yourusername/MyDictionary.git
cd MyDictionary

2. 安装依赖:

pnpm install

3. 在 Chrome 中加载:
   • 打开 Chrome 浏览器,访问 chrome://extensions/
   • 开启"开发者模式"
   • 点击"加载已解压的扩展程序",选择项目目录

使用方式

方式 1: 网页划词

1. 在任意网页选中文本
2. 使用右键菜单或翻译图标
3. 右侧滑出面板显示翻译

方式 2: 快捷键

1. 选中文本 + Ctrl+Shift+D (Mac: Cmd+Shift+D)
2. 侧边栏自动出现

方式 3: 点击插件图标

1. 点击浏览器工具栏的插件图标
2. 在侧边栏手动输入文本翻译

📦 模型说明

默认安装 (600MB)

• ✅ 英译中 / 中译英 - Meta NLLB-200-distilled-600M (600MB，双向共享)

按需下载

当您使用相应功能时,会自动提示下载:

功能	模型	大小	下载时机	技术
其他语言翻译	Meta NLLB-200 (通用)	600MB	翻译非英中语言时	NLLB distilled (与英中同模型)
近义词	DistilBERT-base-uncased	65MB	点击[近义词]按钮时	DistilBERT
例句	MiniLM-L6-v2	23MB	点击[例句]按钮时	Sentence-Transformers
学术短语库	Phrasebank JSON	1.1MB	切换到学术模式时	JSON 数据库
🧠 AI 语义搜索	BGE-Base-EN-v1.5	270MB	点击"语义搜索"标签	北京智源 BGE
AI 语义搜索(轻量)	BGE-Small-EN-v1.5	130MB	低性能设备	北京智源 BGE

总计: 600MB (默认) → 870MB (含AI语义搜索) → 990MB (含TTS) → 1.8GB (完整安装)

🎯 核心功能

1. 智能翻译

• 自动检测 20+ 种语言 (中、英、日、韩、泰、俄、阿等)
• 主要语言对 (英↔中) 使用专用高质量模型 ⭐⭐⭐⭐⭐
• 其他语言使用通用模型支持 200 种语言 ⭐⭐⭐⭐

2. 近义词推荐

• 基于上下文的智能同义词
• 5-10 个相关替换建议
• 点击可重新翻译

3. 例句展示

• 3-5 个真实使用场景例句
• 自动翻译每个例句
• 高亮显示目标词汇

4. AI 驱动的学术写作

• 2,500+ 学术表达: 来自曼彻斯特大学学术短语库精选
• 🧠 AI 语义搜索: BGE 驱动的智能短语发现,带相似度评分 (50-100%)
• 论文各部分专用: Introduction / Methods / Results / Discussion / Conclusion
• 双重搜索模式: 关键词搜索 (即时) + AI 语义搜索 (智能)
• 即复即用: 一键复制到论文中
• 性能检测: 自动推荐 BGE-Base (高性能) 或 BGE-Small (高效)

🛠️ 技术栈

• 前端: Chrome Extension Manifest V3
• AI 库: Transformers.js (@xenova/transformers)
• 翻译模型: Meta NLLB-200-distilled-600M (英中) + NLLB-200 (通用)
• 学术模型: SciBERT + Academic Phrasebank
• 模型下载: Hugging Face Hub (官方推荐方式)

📖 文档

• 产品设计文档
• 技术架构指南
• UI 设计稿
• 模型策略
• 学术写作模型
• 学术版本开发计划
• 产品总结

🎨 Logo 设计

MyDictionary 的 Logo 采用 小浣熊 🦝 抱词典的形象,象征:

• 🧠 聪明智慧 - 本地 AI 智能翻译
• 📚 博学多识 - 支持 200 种语言
• 🎓 学术严谨 - 专业的学术写作辅助

🔧 开发

项目结构

MyDictionary/
├── manifest.json           # Chrome 插件配置
├── package.json            # 依赖管理
├── background.js           # Service Worker (模型管理)
├── content.js              # Content Script (UI)
├── src/
│   ├── config/
│   │   └── models-config.json
│   ├── utils/
│   └── ui/
│       ├── sidebar.html
│       ├── sidebar.css
│       ├── popup.html
│       └── popup.js
├── assets/
│   ├── logo.png
│   └── icons/
└── docs/

开发命令

# 安装依赖
pnpm install

# 开发模式 (监听文件变化)
pnpm run dev

# 构建生产版本
pnpm run build

# 打包为 .crx 文件
pnpm run package

🤝 贡献

欢迎提交 Issue 和 Pull Request!

📄 License

MIT License - 详见 LICENSE 文件

🙏 致谢

AI 模型与库:

• Transformers.js by Hugging Face - 使用 ONNX Runtime 在浏览器中运行 Transformers 模型
• Xenova/nllb-200-distilled-600M - 英中双向翻译 (Meta NLLB-200, 质量 8/10)
• facebook/nllb-200-distilled-600M - 200 种语言通用翻译 (Meta AI)
• BAAI/bge-base-en-v1.5 - BGE 嵌入用于 AI 语义搜索 (MTEB 排名第5)
• BAAI/bge-small-en-v1.5 - 轻量级 BGE 用于低性能设备
• distilbert-base-uncased - 近义词推荐
• sentence-transformers/all-MiniLM-L6-v2 - 句子嵌入
• Academic Phrasebank by University of Manchester - 2,500+ 精选学术短语
• ECDICT by skywind3000 - 770,000+ 词条英中词典 (MIT License)

📚 词典数据来源与许可

MyDictionary 使用以下开源词典数据,所有数据均遵循其原始许可协议:

ECDICT (英中词典)

• 项目地址: skywind3000/ECDICT
• 作者: skywind3000
• 许可证: MIT License
• 词条数: 770,000+
• 内容: 英文单词、音标、中文翻译、词形变化、柯林斯星级评分
• 使用范围:
  • ✅ 个人学习和使用
  • ✅ 开源项目集成
  • ✅ 商业应用 (需保留版权声明)
  • ❌ 禁止直接转售词典数据

数据分层:

• Tier 1 (7,400 词): 高频词汇,内置于插件,立即可用
• Tier 2 (12,000 词): 扩展词汇 (CET6, IELTS, TOEFL, GRE)
• Tier 3 (751,000 词): 完整词库

版权声明

本插件代码采用 MIT License,词典数据保留原始许可证。

使用本插件即表示您同意遵守以下条款:

1. 插件代码可自由使用、修改和分发 (MIT License)
2. 词典数据需遵守 ECDICT 的 MIT License 条款
3. 商业使用需保留版权声明和许可证文件
4. 禁止单独提取词典数据用于转售

---

用 🦝 的智慧,让翻译更简单

Made with ❤️ by Jason, wish my PhD journal published soon.