README.md

# 🌍 Multilingual Chatterbox TTS Server for Windows PC
# Portable Installer Build by LeeAeron / @li_aeron
# Git: https://github.com/LeeAeron/chatterbox

## ✨ Key Features

🌐 **24+ Languages:** Arabic, Chinese, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swedish, Swahili, Turkish
🌐 **All languages already available for usage**

🎤 **Voice Cloning:** Clone any voice using reference audio files
🎤 **Enhanced reference oice lenght of 10 minutes (beta)**

🎤 **Stress marks detection for russian language and setup** Enter propmpt in russian with stress marks or tap on RUStress button to set them up and get realism-like voice!
🎤 **Edit your own local stress marks vocabulary  by edit accent_fixes.yaml file**

📚 **Audiobook Generation:** Process entire books with automatic text chunking

⚡ **GPU Acceleration:** NVIDIA (CUDA) and AMD (ROCm) support

🌐 **Modern Web UI:** Intuitive interface with real-time audio playback

📡 **FastAPI Server:** RESTful API with interactive documentation

## 🔩 System Requirements

- **Windows 10/11** (64-bit) or **Linux**
- **Python 3.10+**
- **4GB+ RAM** (8GB+ recommended)
- **5GB+ storage** for models
- **GPU (Optional):** NVIDIA with 4GB+ VRAM or AMD RX 6000/7000 series

## 🎤 Voice Management

**Predefined Voices:** Place `.wav`/`.mp3` files in `./voices` directory

**Voice Cloning:** Upload reference audio via Web UI (stored in `./reference_audio`)

## ⚙️ Configuration

Edit `config.yaml` for settings:
```yaml
server:
  host: "127.0.0.1"
  port: 8004

tts_engine:
  device: "auto"  # auto, cuda, cpu
  
generation_defaults:
  temperature: 0.7
  language: "en"
```

## 📜 License

MIT License

## 🙏 Acknowledgements

- Original sources: THX to [Resemble AI](https://www.resemble.ai/) for [Chatterbox TTS](https://github.com/resemble-ai/chatterbox)

## Last versions key changes:
16/12/2025
- Created (replaced) a git repository with the chatterbox code from scratch.
- Updated most of the git repository code, syncing with the official 1.0.6.
- Completely replaced the voice tempo (speed) adjustment code:
* Processing is now performed with local sox (downloaded and unpacked locally into the neural network folder).
* No echo, no metallic voice feedback.
- The update no longer updates configuration files (they are simply missing from the update).

11/12/2025
- Added a local ffmpeg. Downloaded during a full installation or this update, it no longer depends on the system ffmpeg.
- Switching between language emotion presets will no longer replace the text with the example text.

9/12/2025
- Fixed stress placement using the built-in Ruaccent mechanism (using the RUStress button). More accurate stress placement. - Added processing of additional dictionary files from the /dicts folder. Automatic file encoding detection.
WARNING! LARGE DICTIONARIES LEAD TO LONG LOADING AND STRESS PROCESSING!!! NO CLAIMS AGAINST ME!!!
- Changes have been made to the language emote presets - they no longer switch languages, but stay with the one you generate.

08/12/2025
- Reference audio is now up to 10 minutes long
- All 24 languages ​​are already enabled and available
- Automatic saving of the last used language for the next engine launch (no need to re-select the last used language)
- Stress marks in Russian can now be added directly in the neural network using the RUStress button
- Added a local dictionary of stress marks for Russian and other languages. Simply edit the accent_fixes.yaml file, which already contains examples, as the Ruaccent dictionary I use is the best available, but not perfect. Ease of use.
- Automatic saving of output audio files to the /outputs folder. Changing the audio compression format changes the audio file format. There is no local cache of audio files.