README.md

metadata

title: Mic Translator
emoji: 🌍
colorFrom: indigo
colorTo: purple
sdk: gradio
sdk_version: 4.19.2
app_file: app.py
pinned: false

MIC Translator – Offline AI-Powered Real-Time Multilingual Voice Translation System

MIC Translator v3.0 is a production-grade, modular translation system. It performs high-precision voice transcription, text correction, automatic source language detection, offline/online hybrid text-to-speech synthesis, and multi-language translation entirely on your local machine.

---

🚀 Quick Start

Step 1 — Install Dependencies

Run the install batch script to set up packages (Flask, Whisper, NLLB-200, Piper support, and dependencies):

install.bat

Step 2 — Start the Translation Dashboard

Launch the server:

run_dashboard.bat

Then open your web browser at: http://localhost:5000

---

📁 Modular Project Structure

MIC dashboard/
├── app.py                  ← Clean Flask router (routes & endpoints)
├── config.py               ← App configuration & NLLB-200 language mapping
├── speech.py               ← Offline Speech-to-Text (OpenAI Whisper)
├── translator.py           ← Offline Translation Engine (Meta NLLB-200)
├── language_detector.py    ← Offline language script analysis
├── correction_engine.py    ← Real-time slang/short-form/spoken-word correction pipeline
├── dataset_loader.py       ← FLORES, OPUS, Tatoeba, and custom dataset manager
├── history.py              ← Saved translations, favorites, CSV/JSON exports, and analytics
├── tts.py                  ← Hybrid TTS (Offline Piper + Online gTTS fallback)
├── static/
│   ├── app.js              ← Searchable selection, recorder, history, & status polling
│   └── style.css           ← Premium dark mode user interface
├── templates/
│   └── index.html          ← Main dashboard markup
├── datasets/
│   ├── custom/             ← Directory for custom parallel translation files (JSON/CSV)
│   └── corrections.json    ← 180+ pre & post-translation correction dictionary rules
├── tests/
│   └── test_all.py         ← 80-test verification suite
└── README.md

---

🎙️ Core Pipelines

Voice Speech
     ↓
OpenAI Whisper (Offline STT)
     ↓
Correction Engine (Pre-translation cleanup: slang, abbreviations)
     ↓
Language Detector (Script/Heuristic auto-detection)
     ↓
Meta NLLB-200 (Offline Translation)
     ↓
Correction Engine (Post-translation refinement)
     ↓
TTS Engine (Offline Piper Voice -> fall back to gTTS Online)

---

⚙️ Requirements & Offline Compatibility

• Python: Version 3.9 or higher
• FFmpeg: Required for offline Whisper audio processing (Download FFmpeg) and added to your system's PATH.
• Microphone: Built-in or external mic.
• Fully Offline Support:
  • Speech Recognition: 100% Offline (Whisper base model).
  • Translation: 100% Offline (NLLB-Distilled-600M).
  • Text-to-Speech: Offline voice synthesis is configured for English (en) and Hindi (hi) using local Piper ONNX files.
  • Adding Offline Voices: To speak other languages offline, download .onnx and .json model files from the Piper Repository and drop them inside the voices/ directory. If a local model is not present, the system automatically uses the online gTTS fallback to speak.

---

🧪 Verification

Run the automated test suite to verify configuration, translation, history caching, script detection, and corrector pipeline components:

python tests/test_all.py

Built for high-performance offline voice and text translation.