Nano Typeless

按下即说，语音秒变文字
基于本地 AI 的 macOS 原生语音输入工具

功能特性 • 安装方法 • 使用方法 • 项目架构 • 参与贡献

English

---

🎬 演示视频

TypelessPromoV2-4K.mp4

✨ 功能特性

功能	描述
🎤 按键说话	按住 Fn 键录音，松开即转文字
🔒 完全本地	所有模型完全在本地运行，数据不出设备
🌐 中英混合	双引擎均原生支持中英文混合输入
⚡ 快速轻量	菜单栏应用，资源占用极低
🎯 通用输入	任意应用可用 - 光标在哪，文字就输入到哪
💻 通用版本	同时支持 Apple Silicon (M1/M2/M3/M4) 和 Intel Mac

🖥️ 系统要求

要求	规格
系统	macOS 14.0 (Sonoma) 或更高版本
芯片	Apple Silicon (M1/M2/M3/M4) 或 Intel - 通用版本支持
内存	建议 8GB 以上

说明：Apple Silicon Mac 将利用神经网络引擎加速推理。Intel Mac 使用 CPU 推理，功能完整。

📦 安装方法

通过 Homebrew 安装（推荐）

# 安装
brew tap ZhaoChaoqun/typeless && brew install --cask nano-typeless

升级

# 升级到最新版本
brew update && brew upgrade nano-typeless

从源码编译

# 克隆仓库
git clone https://github.com/ZhaoChaoqun/nano-typeless.git
cd nano-typeless

# 用 Xcode 打开
open Typeless.xcodeproj

# 或命令行编译
xcodebuild -project Typeless.xcodeproj -scheme Typeless build

首次启动设置

首次启动需要授予两个权限：

权限	用途	如何开启
🎙️ 麦克风	录制语音	系统弹窗（自动）
♿ 辅助功能	监听全局 Fn 键	系统设置 → 隐私与安全性 → 辅助功能

提示：授予辅助功能权限后，可能需要重启应用。

🚀 使用方法

快速开始 启动 Nano Typeless - 出现在菜单栏 按住 Fn 键开始说话 松开 Fn 键完成录音 文字 自动插入到光标位置 使用流程 [按住 Fn] → "你好，这是一个测试" → [松开 Fn] ↓ "你好，这是一个测试" 出现在光标处

状态指示 状态 指示器 就绪 🎵 菜单栏图标 录音中 🔴 视觉遮罩 处理中 ⏳ 加载指示器

🏗️ 项目架构

typeless/
├── Sources/
│   ├── TypelessApp.swift          # 应用入口和生命周期
│   ├── RecordingManager.swift     # 音频录制和引擎调度
│   ├── RecordingState.swift       # 录音状态机管理
│   ├── ASREngine.swift            # ASR 引擎统一协议
│   ├── ASRStreamRecognizing.swift # 流式识别器协议
│   ├── SherpaOnnxOnlineRecognizer.swift # Streaming Paraformer 流式识别
│   ├── QwenASRRecognizer.swift    # Qwen3-ASR 流式识别
│   ├── SherpaOnnxManager.swift    # 模型下载和管理
│   ├── SherpaOnnxPunctuation.swift # CT-Transformer 标点模型
│   ├── CloudRewriter.swift        # 云端 LLM 文本纠错
│   ├── Qwen3TextRewriter.swift    # Qwen3-0.6B 本地文本纠错
│   ├── TermNormalizer.swift       # 词典后处理（术语规范化）
│   ├── KeyMonitor.swift           # 全局 Fn 键检测
│   ├── TextInserter.swift         # 光标文字插入
│   ├── OverlayWindow.swift        # 录音 UI 遮罩
│   └── SettingsView.swift         # 偏好设置 UI
├── Frameworks/
│   ├── sherpa-onnx/               # Paraformer 推理框架
│   ├── qwen-asr/                  # Qwen3-ASR Rust FFI 库
│   └── qwen3-rewrite/             # Qwen3-0.6B 文本纠错 C FFI 库
├── Package.swift                  # Swift Package 依赖
└── Typeless.xcodeproj/            # Xcode 项目

技术栈

组件	技术
UI 框架	SwiftUI
语音识别	sherpa-onnx (Streaming Paraformer) / Qwen3-ASR (Rust FFI)
文本后处理	Cloud LLM Rewriter / Qwen3-0.6B 本地纠错 / 词典规范化
音频采集	AVFoundation
按键监听	CGEvent Tap API
文字插入	CGEvent（键盘模拟）

工作原理

┌─────────────┐     ┌──────────────┐     ┌─────────────┐     ┌──────────────┐
│  Fn 键      │────▶│   录制       │────▶│   ASR 引擎  │────▶│   插入       │
│  监听       │     │   音频       │     │   转写      │     │   文字       │
└─────────────┘     └──────────────┘     └─────────────┘     └──────────────┘
    CGEvent           AVFoundation         本地 AI           CGEvent

🔧 配置说明

应用内置两种 ASR 引擎，可在设置中切换。默认使用 Streaming Paraformer。

引擎	大小	模式	标点	适用场景
Streaming Paraformer	~414MB（含标点模型）	流式	CT-Transformer	日常使用（默认）
Qwen3-ASR	~834MB	流式	自带标点	高精度，长文本

🤝 参与贡献

欢迎贡献！请随时提交 Pull Request。

1. Fork 本仓库
2. 创建功能分支 (git checkout -b feature/AmazingFeature)
3. 提交更改 (git commit -m 'Add some AmazingFeature')
4. 推送到分支 (git push origin feature/AmazingFeature)
5. 提交 Pull Request

📄 许可证

本项目基于 MIT 许可证开源 - 详见 LICENSE 文件。

🔗 关联项目

Nano Typeless 的部分能力由以下子项目提供支持：

项目	说明	语言
QwenASR	Qwen3-ASR 流式推理引擎，编译为 dylib 供本项目调用	Rust
mlx-qwen-asr-streaming	Qwen3-ASR 流式推理的 MLX 实现，用于 Apple Silicon 上的快速实验	Python
Qwen3-0.6B	Qwen3 0.6B 微调模型，支持 ASR 文本二阶段 rewrite（纠错+格式化）	Python
sherpa-onnx	sherpa-onnx fork，修复 Streaming Paraformer 流式尾部截断问题	C++

🙏 致谢

• FunASR - 阿里达摩院开源语音识别模型
• sherpa-onnx - 跨平台语音识别推理框架
• Qwen3-ASR - 通义千问大模型语音识别

---

English

Press. Speak. Type.
A native macOS voice-to-text tool powered by local AI

Features • Installation • Usage • Architecture • Contributing

---

🎬 Demo

TypelessPromoV2-4K.mp4

✨ Features

Feature	Description
🎤 Push-to-Talk	Hold Fn key to record, release to transcribe
🔒 100% Local	All models run entirely on-device, no data leaves your Mac
🌐 Multilingual	Two ASR engines with native Chinese-English mixed input support
⚡ Fast & Lightweight	Menu bar app with minimal resource usage
🎯 Universal Input	Works in any app - just position your cursor and speak
💻 Universal Binary	Runs natively on both Apple Silicon (M1/M2/M3/M4) and Intel Macs - one app, all Macs

🖥️ System Requirements

Requirement	Specification
OS	macOS 14.0 (Sonoma) or later
Chip	Apple Silicon (M1/M2/M3/M4) or Intel - Universal Binary supported
RAM	8GB+ recommended

Note: Apple Silicon Macs will utilize the Neural Engine for faster inference. Intel Macs use CPU-based inference with full functionality.

📦 Installation

Install via Homebrew (Recommended)

# Install
brew tap ZhaoChaoqun/typeless && brew install --cask nano-typeless

Upgrade

# Upgrade to latest version
brew update && brew upgrade nano-typeless

Build from Source

# Clone the repository
git clone https://github.com/ZhaoChaoqun/nano-typeless.git
cd nano-typeless

# Open in Xcode
open Typeless.xcodeproj

# Or build via command line
xcodebuild -project Typeless.xcodeproj -scheme Typeless build

First Launch Setup

On first launch, you'll need to grant two permissions:

Permission	Purpose	How to Enable
🎙️ Microphone	Record your voice	System Prompt (automatic)
♿ Accessibility	Listen for global Fn key	System Settings → Privacy & Security → Accessibility

Tips: After granting Accessibility permission, you may need to restart the app.

🚀 Usage

Quick Start Launch Nano Typeless - it appears in your menu bar Hold the Fn key and start speaking Release the Fn key when done Text is automatically inserted at cursor position Workflow Example [Hold Fn] → "Hello, this is a test" → [Release Fn] ↓ "Hello, this is a test" appears at cursor

Status Indicators State Indicator Ready 🎵 Menu bar icon Recording 🔴 Visual overlay Processing ⏳ Loading indicator

🏗️ Architecture

typeless/
├── Sources/
│   ├── TypelessApp.swift          # App entry & lifecycle
│   ├── RecordingManager.swift     # Audio recording & engine dispatch
│   ├── RecordingState.swift       # Recording state machine
│   ├── ASREngine.swift            # Unified ASR engine protocol
│   ├── ASRStreamRecognizing.swift # Stream recognizer protocol
│   ├── SherpaOnnxOnlineRecognizer.swift # Streaming Paraformer
│   ├── QwenASRRecognizer.swift    # Qwen3-ASR streaming recognition
│   ├── SherpaOnnxManager.swift    # Model download & management
│   ├── SherpaOnnxPunctuation.swift # CT-Transformer punctuation
│   ├── CloudRewriter.swift        # Cloud LLM text correction
│   ├── Qwen3TextRewriter.swift    # Qwen3-0.6B local text correction
│   ├── TermNormalizer.swift       # Dictionary post-processing (term normalization)
│   ├── KeyMonitor.swift           # Global Fn key detection
│   ├── TextInserter.swift         # Cursor text insertion
│   ├── OverlayWindow.swift        # Recording UI overlay
│   └── SettingsView.swift         # Preferences UI
├── Frameworks/
│   ├── sherpa-onnx/               # Paraformer inference framework
│   ├── qwen-asr/                  # Qwen3-ASR Rust FFI library
│   └── qwen3-rewrite/             # Qwen3-0.6B text correction C FFI library
├── Package.swift                  # Swift Package dependencies
└── Typeless.xcodeproj/            # Xcode project

Tech Stack

Component	Technology
UI Framework	SwiftUI
Speech Recognition	sherpa-onnx (Streaming Paraformer) / Qwen3-ASR (Rust FFI)
Post-Processing	Cloud LLM Rewriter / Qwen3-0.6B local correction / Term normalization
Audio Capture	AVFoundation
Key Monitoring	CGEvent Tap API
Text Insertion	CGEvent (Keyboard Simulation)

How It Works

┌─────────────┐     ┌──────────────┐     ┌─────────────┐     ┌──────────────┐
│  Fn Key     │────▶│   Record     │────▶│  ASR Engine │────▶│   Insert     │
│  Monitor    │     │   Audio      │     │  Transcribe │     │   Text       │
└─────────────┘     └──────────────┘     └─────────────┘     └──────────────┘
    CGEvent           AVFoundation         Local AI           CGEvent

🔧 Configuration

The app includes two built-in ASR engines, switchable in Settings. Default is Streaming Paraformer.

Engine	Size	Mode	Punctuation	Best For
Streaming Paraformer	~414MB (incl. punct model)	Streaming	CT-Transformer	Daily use (default)
Qwen3-ASR	~834MB	Streaming	Built-in	High accuracy, long text

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

1. Fork the repository
2. Create your feature branch (git checkout -b feature/AmazingFeature)
3. Commit your changes (git commit -m 'Add some AmazingFeature')
4. Push to the branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🔗 Related Projects

Some capabilities of Nano Typeless are powered by these sub-projects:

Project	Description	Language
QwenASR	Qwen3-ASR streaming inference engine, compiled as dylib for this project	Rust
mlx-qwen-asr-streaming	MLX implementation of Qwen3-ASR streaming inference for rapid prototyping on Apple Silicon	Python
Qwen3-0.6B	Fine-tuned Qwen3 0.6B model for two-stage ASR text rewriting (correction + formatting)	Python
sherpa-onnx	sherpa-onnx fork with fix for Streaming Paraformer tail truncation issue	C++

🙏 Acknowledgments

• FunASR - Alibaba DAMO Academy's open-source speech recognition model
• sherpa-onnx - Cross-platform speech recognition inference framework
• Qwen3-ASR - Qwen large model speech recognition

---

Made with ❤️ for the macOS community