A completely free, open-source Swedish voice chatbot using Whisper, Swedish language models, and text-to-speech.

• 🎤 Swedish Speech Recognition - Whisper with Swedish optimization
• 🧠 Swedish AI Responses - Bellman model for natural conversation
• 🔊 Swedish Text-to-Speech - Google TTS
• 💬 Natural Conversation - Optimized for back-and-forth chat
• 🌐 Web Interface - Easy-to-use Gradio interface
• 🔒 Completely Free - No API keys required for basic functionality

git clone <your-repo-url>
cd swedish-chatbot

# Automatic setup (recommended)
./setup_environment.sh

# Or manual setup:
python3 -m venv chatbot_env
source chatbot_env/bin/activate
pip install -r requirements.txt

For best Swedish models, get a free token from Hugging Face:

export HF_TOKEN='your_token_here'

# Easy way:
./start_fixed_chatbot.sh

# Or manual:
source chatbot_env/bin/activate
export HF_TOKEN='your_token_here'  # Optional
python swedish_chatbot_fixed.py

• Open http://localhost:7861 in your browser
• Click the microphone and speak Swedish
• Listen to the AI's Swedish response
• Continue the conversation naturally!

brew install [email protected] portaudio ffmpeg

sudo apt install python3-venv python3-pip portaudio19-dev ffmpeg mpg123 alsa-utils

• Install Python 3.11+
• Install ffmpeg
• The setup script will handle the rest

Try these Swedish phrases to test the chatbot:

• "Hej, vad heter du?"
• "Hur mår du idag?"
• "Berätta om Sverige"
• "Vad gillar du?"
• "Kan du hjälpa mig?"

swedish-chatbot/
├── swedish_chatbot_fixed.py      # Main chatbot (recommended)
├── swedish_chatbot_whisper.py    # Alternative version
├── requirements.txt              # Python dependencies
├── setup_environment.sh          # Environment setup script
├── start_fixed_chatbot.sh        # Quick start script
├── setup_token.py                # HF token setup helper
├── .gitignore                    # Excludes virtual env and models
├── README.md                     # This file
├── FIXED_VERSION_GUIDE.md        # Detailed fix explanations
└── docs/                         # Additional documentation
    ├── MODEL_SETUP.md
    ├── QUICKSTART.md
    └── FIXES_SUMMARY.md

• The chatbot_env/ folder is NOT included in git (it's in .gitignore)
• Never commit the virtual environment - it's huge (several GB)
• Each user creates their own environment using setup_environment.sh

• Models are downloaded automatically on first run
• Models are cached locally (~2-7GB depending on which models load)
• Model cache is NOT committed to the repository

• Only requirements.txt is committed - lists what to install
• Virtual environment is created fresh for each installation
• This keeps the repository small and portable

• Grant microphone permissions in System Preferences
• Allow your browser to access microphone

# Clear cache and retry
rm -rf ~/.cache/huggingface/
python setup_token.py  # Set up HF token
python swedish_chatbot_fixed.py

# Install audio dependencies
brew install portaudio ffmpeg  # macOS
sudo apt install portaudio19-dev ffmpeg  # Linux

• Uses OpenAI Whisper "small" model
• Optimized Swedish language settings
• Better accuracy than basic Whisper

1. Bellman (neph1/bellman-7b-mistral-instruct-v0.2) - Best Swedish conversation
2. GPT-SW3 (AI-Sweden-Models/gpt-sw3-126m) - Official Swedish model (requires HF token)
3. DialoGPT (microsoft/DialoGPT-small) - Good fallback
4. DistilGPT2 - Basic fallback

• Short, natural responses (40 tokens max)
• Artifact filtering (removes "Datum:", "Kubernetes", etc.)
• Model-specific prompt optimization
• Context-aware conversation

1. Fork the repository
2. Create your feature branch
3. DON'T commit the virtual environment (it's already in .gitignore)
4. Only commit code, documentation, and requirements
5. Submit a pull request

MIT License - feel free to use for personal or commercial projects.

• OpenAI Whisper for speech recognition
• AI Sweden for GPT-SW3 models
• Bellman project for excellent Swedish conversation AI
• Gradio for the web interface
• Google for Text-to-Speech