AI Realtime Intelligent Audio
- Universal Audio Capture - Works with any application (games, videos, calls, etc.)
- Three Recognition Modes:
- Precise Mode: Uses Whisper for high-accuracy transcription (NVIDIA GPU recommended)
- Realtime Mode: Uses Sherpa-ONNX/Vosk for word-by-word streaming (CPU/GPU)
- LiveCaptions Mode: Uses Windows 11 built-in AI (✅ No GPU/CUDA required, works on AMD/Intel/NVIDIA)
- Multi-language Support: Chinese, English, Japanese, Korean, and more
- Real-time Translation: Translate transcriptions with Google Cloud or NLLB (local)
- Customizable Overlay: Draggable subtitle window with adjustable position
- Multilingual UI: English, Traditional Chinese, Simplified Chinese
- Embedded Python: No separate Python installation required
Important
How to Update: Directly replace the src folder in your original project package with the src folder from this project repository.
- Microphone Input: Auto-detects audio input devices with a new audio source dropdown list.
.tsFile Continuous Transcription: Tracks and transcribes continuously growing.tsfiles in real-time (1:1 speed). Scans the selected folder every loop to automatically read the newly appended parts of the latest.tsfile. Includes a caching algorithm to avoid unstable parts at the beginning of the file and to maintain a safe time buffer. Optionally scans subdirectories.- S3 Cloud Storage: Can automatically upload transcriptions to S3 for direct link access.
- Timezone Settings: Control timezone displays in both logs and the console.
- Subtitle Overlay: Added a button to easily toggle the subtitle overlay window on and off.
- Log Management: Added a switch between detailed and brief logs (brief logs only show time and transcribed/translated text). All transcribed contents are fully displayed without truncation. Logs are written to disk directly (immediate flush) without locking the file.
- Language Separation: Separated Simplified Chinese and Traditional Chinese outputs, fixing issues where both were mixed in the text.
- UI Optimizations: Improved layout to prevent visual crowding. The UI can now be freely resized.
- Quick Launch: Added a one-click launcher script to bypass the UI and directly start the program.
Perfect for most users. Zero configuration.
- OS: Windows 11 (version 22H2 or later)
- Hardware: Any standard laptop or PC (Intel/AMD/NVIDIA)
- No specific GPU or CUDA required.
For users needing high-precision offline transcription.
- OS: Windows 10 or 11
- GPU: NVIDIA GTX 10 series or newer (Recommended for best speed)
- Storage: ~10 GB free space
| Version | Size | Description | Google Drive | Baidu |
|---|---|---|---|---|
| Lite 【Recommended】 | ~600 MB | Best for most users. Lightweight, fast, accurate, works on ANY GPU (AMD/Intel). Requires Windows 11 (22H2 or later). | Google Drive | Baidu |
| Full | ~7.6 GB | All Modes (Precise/Realtime/Live). Offline models pre-installed. | Google Drive | Baidu |
| Model | Type | Size | Languages |
|---|---|---|---|
| Whisper Large-v3 | Precise | 3 GB | All |
| Sherpa-ONNX Bilingual | Realtime | 500 MB | Chinese, English |
| Vosk Japanese | Realtime | 1 GB | Japanese |
| NLLB Translation | Translation | 1.2 GB | Many |
- Download and extract the ZIP file
- Double-click
ARIA.bat - Lite Version: Ready to use immediately (Live Captions only).
- Full Version: Select mode and click Start.
| Mode | Engine | Best For | GPU Requirement |
|---|---|---|---|
| Precise | Whisper | Speeches, videos, pre-recorded content | NVIDIA recommended |
| Realtime | Sherpa-ONNX / Vosk | Live conversations, streaming | CPU/GPU |
| LiveCaptions | Windows 11 AI | General Use (Movies/Meetings) | ✅ Any CPU / GPU (No NPU needed) |
| Language | Precise Mode | Realtime Mode | LiveCaptions Mode |
|---|---|---|---|
| Chinese (中文) | ✅ | ✅ (Sherpa-ONNX) | ✅ |
| English | ✅ | ✅ (Sherpa-ONNX) | ✅ |
| Japanese (日本語) | ✅ | ✅ (Vosk) | ✅ |
| Korean (한국어) | ✅ | ❌ | ✅ |
| Spanish, French, etc. | ✅ | ❌ | ✅ |
| + 50 more | ✅ | ❌ | ❌ |
LiveCaptions Mode: Supports 11 languages via Windows 11 built-in AI. No model download required!
Local Models (Offline):
- NLLB-200: Meta's multilingual translation model, supports 200+ languages, runs locally
Online Services (Free APIs):
- Google Translate: Fast and accurate, via web scraping
- Bing Translator: Microsoft's translation service
- Youdao: 有道翻译 (specialized for Chinese-English translation)
Note: Online services use web scraping and may be subject to rate limiting. Local NLLB model recommended for reliability.
ARIA/
├── python/ # Embedded Python (no installation needed)
├── src/ # Source code
├── models/ # AI models (Full version only)
├── ARIA.bat # Launcher
This project is licensed under the GNU General Public License v3.0 - see the LICENSE file for details.
- GitHub: @sayksii
- Email: mark42967151@gmail.com