🌐 English | Русский

🎧 Hariko

Local media pipeline: Audio/Video → Whisper → Text Summary → Jarvis
100% offline, zero network latency, fully configurable.

---

✨ Features

• 🎤 Speech-to-Text via openai/whisper (models: tiny → large-v3)
• 🌍 Language Detection (auto, ru, en, etc.)
• 💻 Hardware Acceleration (CPU / CUDA auto-fallback)
• 📁 Multi-format Support (.mp3, .wav, .mp4, .mkv, .webm)
• ⚙️ GUI Configuration (theme, model size, compute device, paths)
• 🔌 Jarvis-Ready (exports clean text for downstream LLM processing)

---

Quick Start

Prerequisites

• Python 3.10–3.12
• ffmpeg (installed system-wide)
• NVIDIA GPU with CUDA drivers (optional, for faster inference)

Installation

git clone https://github.com/Saka1r/hariko.git
cd hariko
python -m venv venv
source venv/bin/activate  # or venv\Scripts\activate on Windows
pip install -r requirements.txt

Run

python src/hariko.py

💡 First run requires config/config.json. See the Configuration section below.

⚙️ Configuration

{
  "theme": "Dark",
  "gguf_path": "./models/model.Q4_K_M.gguf",
  "whisper_lang": "auto",
  "whisper_model": "base",
  "whisper_computing_device": "cpu"
}

Parameter	Description	Default
theme	UI theme	"Dark", "Light"
gguf_path	Path to GGUF LLM model (future use)	""
whisper_lang	Transcription language	"auto", "ru", "en"
whisper_model	Whisper model size	"tiny", "base", "small", "medium", "large-v3"
whisper_computing_device	Compute backend	"cpu", "cuda"

📁 Project Structure

hariko/
├── src/
│   ├── hariko.py      # Main UI & app logic (KivyMD 2.0)
│   ├── stt.py         # Whisper transcription pipeline
│   └── tools.py       # Config I/O & file utilities
├── config/
│   ├── config.json    # User settings
│   └── anime.png      # App icon
├── output.txt         # Generated transcript
├── requirements.txt   # Python dependencies
└── README.md          # Documentation

🔌 Jarvis Integration

Hariko outputs clean text to output.txt. Jarvis can:

1. Read the file directly (tools.get_output())
2. Consume via REST/WebSocket (planned)

📜 License

MIT License. Free to use, modify, and distribute.

🤝 Support

🐞 Report bugs → Issues
💡 Feature requests → Discussions
🛠️ Contribute → Fork & PR

Built for local, private AI workflows.