AIGC-Telegram-Bot

AIGC Telegram Bot

AI-powered voice cover generation bot for Telegram. Upload a song or paste a YouTube link, pick a voice model, and get an AI-generated cover delivered directly in your chat.

Based on AICoverGen by SociallyIneptWeeb.

---

Features

• YouTube & Audio Upload — Paste a YouTube URL or upload MP3/WAV/OGG directly
• RVC Voice Models — Swap vocals using any RVC-trained voice model
• MDX Vocal Separation — Isolate vocals from instrumentals using MDX-Net
• Audio Effects — Reverb, compression, noise reduction, pitch shifting
• Customizable Settings — Pitch, index rate, F0 method, reverb, output format, and more
• 3 Inference Modes — Full pipeline, RVC only, or MDX separation only
• Admin Controls — User management, stats, and cleanup commands
• Docker Ready — Full Docker + docker-compose deployment

---

Project Structure

aigc-telegram-bot/
├── src/
│   ├── __init__.py
│   ├── bot.py                # Telegram bot — commands, handlers, conversation flow
│   ├── config.py             # Configuration management (.env + dataclasses)
│   ├── state.py              # Thread-safe user session / state manager
│   ├── pipeline.py           # Async pipeline runner (ThreadPoolExecutor wrapper)
│   └── core/                 # AI Cover Generation pipeline
│       ├── __init__.py
│       ├── cover_pipeline.py # Main pipeline: download → separate → convert → mix
│       ├── mdx.py            # MDX-Net vocal separation (ONNX Runtime)
│       ├── rvc_voice.py      # RVC voice conversion (SawitProject/rvc)
│       └── my_utils.py       # Audio utility functions (ffmpeg wrapper)
├── assets/
│   └── mdxnet_models/        # Bundled MDX-Net ONNX models (~234 MB)
│       ├── UVR-MDX-NET-Voc_FT.onnx
│       ├── UVR_MDXNET_KARA_2.onnx
│       ├── Reverb_HQ_By_FoxJoy.onnx
│       ├── UVR-MDX-NET-Inst_HQ_4.onnx
│       └── model_data.json
├── data/
│   ├── models/               # Place RVC voice models here
│   ├── temp/                 # Temporary audio files (auto-cleaned)
│   └── output/               # Generated covers
├── Dockerfile
├── docker-compose.yml
├── requirements.txt
├── .env.example              # Configuration template
└── README.md

---

How It Works

The bot runs a multi-stage AI audio pipeline:

Input (YouTube URL / Audio File)
        │
        ▼
┌─────────────────────────┐
│  1. Download / Load     │  yt-dlp for YouTube, pydub for local files
└───────────┬─────────────┘
            │
            ▼
┌─────────────────────────┐
│  2. MDX Vocal           │  UVR-MDX-NET-Voc_FT → vocals + instrumentals
│     Separation          │  UVR_MDXNET_KARA_2 → main + backup vocals
│                         │  Reverb_HQ_By_FoxJoy → de-reverb
└───────────┬─────────────┘
            │
            ▼
┌─────────────────────────┐
│  3. RVC Voice           │  SawitProject/rvc — voice cloning
│     Conversion          │  Configurable pitch, index rate, F0 method
└───────────┬─────────────┘
            │
            ▼
┌─────────────────────────┐
│  4. Audio Effects       │  Reverb, compression, HPF (pedalboard)
│     & Mixing            │  Mix AI vocals + backup + instrumentals
└───────────┬─────────────┘
            │
            ▼
    Output (MP3/WAV) → Sent to Telegram chat

Inference Modes

---

Quick Start

1. Prerequisites

• Python 3.10 or 3.11
• FFmpeg (sudo apt install ffmpeg)
• SoX (sudo apt install sox)
• A Telegram Bot Token from @BotFather
• At least one RVC voice model (.pth file + optional .index file)

2. Install

git clone https://huggingface.co/R-Kentaren/AIGC-Telegram-Bot
cd AIGC-Telegram-Bot

pip install -r requirements.txt

# Install RVC voice conversion engine
pip install git+https://github.com/SawitProject/rvc.git

3. Configure

cp .env.example .env

Edit .env and set at minimum:

BOT_TOKEN=your_telegram_bot_token_here
ADMIN_IDS=your_telegram_user_id

4. Add Voice Models

Place each RVC voice model in its own folder under data/models/:

data/models/
├── my_voice_model/
│   ├── model.pth          # Required: RVC model weights
│   └── model.index        # Optional: FAISS feature index
├── another_voice/
│   ├── model.pth
│   └── model.index
└── ...

5. Run

python -m src.bot

The bot will start polling for updates. Open Telegram and send /start to your bot.

---

Bot Commands

Cover Generation Flow

/cover
  → Choose input: [YouTube URL] or [Upload File]
  → Paste URL or send audio file
  → Select voice model from list
  → Review settings & confirm
  → Wait for progress updates
  → Receive generated cover as audio file

---

Configuration Reference

All settings are managed through .env. Copy .env.example for the full list.

Bot Settings

Pipeline Defaults

---

Docker Deployment

Using Docker Compose (Recommended)

# Clone and configure
git clone https://huggingface.co/R-Kentaren/AIGC-Telegram-Bot
cd AIGC-Telegram-Bot
cp .env.example .env
# Edit .env with your BOT_TOKEN and ADMIN_IDS

# Add voice models
cp -r /path/to/your/models/* data/models/

# Start the bot
docker compose up -d

# View logs
docker compose logs -f

Using Docker Directly

docker build -t aigc-bot .
docker run -d \
  --name aigc-bot \
  --env BOT_TOKEN=your_token \
  --env ADMIN_IDS=your_id \
  -v ./data/models:/app/data/models \
  -v ./data/output:/app/data/output \
  aigc-bot

GPU Support

Uncomment the GPU section in docker-compose.yml if you have NVIDIA GPUs:

deploy:
  resources:
    reservations:
      devices:
        - capabilities: [gpu]

---

RVC Voice Models

Where to Get Models

• AI Hub Discord — Community voice models
• RVC Models Collection — HuggingFace
• Train your own using RVC WebUI

Model Format

Each model is a folder containing:

your_model/
├── model.pth            # Required — trained RVC model weights
└── model.index          # Optional — FAISS index for timbre retrieval

Place them in data/models/<model_name>/.

Pitch Guidelines

---

Adding Custom Voice Models via Telegram

Users can simply add new models by placing .pth (and optionally .index) files in data/models/<name>/ on the server. The bot automatically detects new models when /models is called.

---

Troubleshooting

Bot doesn't start

• Verify BOT_TOKEN is set correctly in .env
• Ensure python-telegram-bot is installed: pip install python-telegram-bot
• Check logs for import errors

"No voice models available"

• Ensure RVC model .pth files are in data/models/<model_name>/
• Check file permissions

RVC inference fails

• Install RVC: pip install git+https://github.com/SawitProject/rvc.git
• If FP16 errors occur, set RVC_HALF_PRECISION=0 in your environment
• Ensure you have enough RAM/VRAM (minimum 4 GB, recommended 8 GB+)

YouTube download fails

• Ensure yt-dlp is up to date: pip install -U yt_dlp
• Some videos may be region-restricted or age-gated
• For YouTube Music, cookies may be needed (place in assets/config.txt)

OOM (Out of Memory)

• Reduce MAX_CONCURRENT_JOBS to 1
• Use shorter songs or lower quality settings
• Enable CPU mode by not having CUDA installed

---

Tech Stack

---

License

This project is licensed under the MIT License — see LICENSE for details.

The underlying AI models (MDX-Net, RVC) have their own licenses. Please refer to their respective repositories for more information.

---

Credits

• SociallyIneptWeeb/AICoverGen — Original AI Cover Generation pipeline
• SawitProject/rvc — RVC voice conversion engine
• openvpi/MDX-Net — Vocal separation models
• Spotify/pedalboard — Audio effects