English | 中文 |

LocalAI-Translator

This is a simple and modern translation app that supports Google Gemini and OpenAI compatible APIs; it also supports offline open-source models to prevent sensitive data leakage. It offers fast and accurate multilingual text translation with a clean, mobile-first user interface.
🚀Live Demo

📸 Screenshots

## Design Features

Core Translation Functionality💬

• Text Translation: Supports real-time streaming translation, displaying the translation result immediately as you input text.
• Speech Translation:
  • General Recording: Converts the user's voice into text and fills in the input box.
  • Real-time Speech Translation (AST): Directly translates the user's voice into the target language and displays it as text, achieving a more fluid conversation-style translation.
• Image Translation: Through the device camera or importing images from the album, the application can automatically detect and extract text from images and then translate it.

Online and Offline Modes🌐

• Online Mode: Utilizes a powerful cloud AI model (Gemini or OpenAI) for high-quality translation. This mode requires network connection and valid API keys.
• Offline Mode: Translates using the Gemma model downloaded locally on the device (run through MediaPipe). Offline mode supports text, voice-to-text, and image text extraction.

Multilingual Support🌟

The application supports a variety of mainstream languages, including English, Traditional Chinese, Simplified Chinese, Japanese, Korean, Spanish, etc., and provides "automatic detection" of the source language.

Personalization Settings🔒

• API Keys and Model Management: Users can input their own Gemini or OpenAI API keys and customize the model names they want to use.
• Offline Model Download: Provides an interface to download, manage, and delete local models for offline mode.
• Offline Text-to-Speech (TTS): Users can customize the voice, speed, and pitch of offline TTS for a personalized listening experience.
• High Accuracy Japanese to Chinese Translation Mode: For Japanese translation to Chinese scenarios, a special "two-step translation" mode (JP → EN → CN) is provided to enhance translation accuracy and fluency.

History Record📊

Automatically saves the last 50 translation records for users to view and reuse at any time.

Document Structure🏗

• / (root directory): Contains the entry point and main configuration files for the project, such as index.html, index.tsx, package.json, vite.config.ts, App.tsx, constants.ts, types.ts, i18n.ts.
• /components: Stores all React UI components. Each component is responsible for rendering and interacting with a portion of the UI.
  • TranslationInput.tsx: Source language input interface.
  • TranslationOutput.tsx: Target language output interface.
  • SettingsModal.tsx: Settings pop-up window.
  • HistoryModal.tsx: History record pop-up window.
  • CameraView.tsx: Camera translation interface.
  • LanguageSelector.tsx: Custom language selection dropdown.
• /services: Stores the application's core business logic, particularly for interacting with APIs and backend services.
  • geminiService.ts: Handles all communication with the Google Gemini API.
  • openaiService.ts: Handles all communication with the OpenAI API.
  • asrService.ts: Manages audio processing and transcription services.
  • downloadManager.ts: Responsible for downloading, pausing, resuming, and deleting offline models.
• /workers: Contains Web Workers for background processing to keep the UI responsive.
  • litert-lm.worker.ts: Main inference worker for offline LLM using LiteRT.
  • mediapipe.worker.ts: Legacy worker for offline LLM using MediaPipe.
  • offline.worker.ts: Background worker for offline model management.
  • nemotron.worker.ts: Worker for real-time speech recognition (ASR) using ONNX Runtime (Nemotron).
  • transformersASR.worker.ts: Worker for speech recognition (ASR) using Hugging Face Transformers.js (Whisper).
  • ocr.worker.ts: Worker for local image text extraction (OCR) using esearch-ocr and ONNX Runtime.
• /hooks: Custom React hooks for specialized logic.
  • usePaddleOcr.ts: Hook for local OCR using PaddleOCR.
  • useWebSpeech.ts: Hook for Web Speech API integration.
• /utils: Utility functions and shared helpers.
  • db.ts: IndexedDB management for local storage.
• /types.ts: Defines the TypeScript types used throughout the application.
• /constants.ts: Stores application constants, such as supported language lists, offline model information, etc.
• /i18n.ts: i18next internationalization configuration file, manages multilingual interfaces.

UI Feature Explanation

Main Interface

The main interface is divided into two blocks, one for input and one for output.

• Source Language Block (TranslationInput):

  • Text Input Area: For inputting or pasting the text to be translated.
  • Language Selector: Select the source language, with automatic detection supported.
  • Function Buttons:
    • Translate: Execute the translation.
    • Record: Activate the microphone for voice input.
    • Camera: Open the camera for image translation.
    • Settings: Open the settings window.
  • Status Display: Displays the character count and current network connection status (online/offline).

• Target Language Block (TranslationOutput):

  • Translation Result Display: Displays the translated text.

  • Language Selector: Select the target language.

  • Function Buttons:

    • Swap Languages: Quickly swap the source and target languages.
    • Flip Screen: Mirror the text in this block for easy display to someone across from you.
    • Copy: Copy the translation result to the clipboard.
    • History: Open the history window.
    • Clear: Clear the translation result.

  • Text to Speech (TTS): Provides male and female voice (or custom voice) buttons to read the translation result aloud.

  • Real-time Speech Translation (AST): When activated, it translates speech input into the target language.

Pop-up windows

• Settings (SettingsModal):

  • Online Pages: Set API keys, service providers (Gemini/OpenAI), and model names.
  • Offline Pages: Manage Hugging Face API keys, download/delete offline models, enable offline mode, and high-accuracy mode during the day.

• History (HistoryModal): Displays a list of past translation records, which can be clicked to reload.

• Camera (CameraView):

  • Provides a real-time preview of the camera.
  • Controls include: taking a photo, importing from the gallery, flash switch, and focus zoom.

Technical Stack🚀

• Frontend Framework: React 19
• Language: TypeScript
• Build Tool: Vite
• Styles: Tailwind CSS
• Offline Models: Google Gemma, Whisper, Nemotron, PaddleOCR
• Offline Inference Engine: LiteRT, MediaPipe, Transformers.js, ONNX Runtime Web
• Online Translation: Google Gemini API (@google/genai) / OpenAI API
• Internationalization: i18next

Notes⚠️

GitHub Pages settings

1. Modify package.json

   "homepage": "https://<your-github-username>.github.io/repository name", ← Add this
   "deploy": "gh-pages -d dist", ← Add this  
   

2. Modify vite.config.ts

   base: '/repository name/', // ← Add this
   plugins: [],
   

           icons: [
          {
            "src": "images/icon-192.png", ←Modify "images/icon-192.png" not "/images/icon-192.png"
            "sizes": "192x192",
            "type": "image/png",
            "purpose": "any maskable"
          },
          {
            "src": "images/icon-512.png", ←Modify "images/icon-512.png" not "/images/icon-512.png"
            "sizes": "512x512",
            "type": "image/png"
          }
        ]
   

3. Install gh-pages
   npm install gh-pages --save-dev or yarn add gh-pages -D
4. Output static website files
   npm build or yarn build
5. Run to GitHub Pages
   npm run deploy or yarn run deploy