AsukaAI

AsukaAI is your personal offline AI companion, ensuring privacy while offering powerful features. This local AI waifu seamlessly integrates speech-to-text, text generation, and text-to-speech functionalities:

• RealtimeSTT with faster_whisper under the hood for speech-to-text
• Ollama with Llama3 under the hood for text generation
• RealtimeTTS with CoquiTTS under the hood for text-to-speech

🌟 Features

• Privacy: All processing is done locally on your machine, ensuring your data never leaves your device.
• Customizability: Create and integrate your own models according to your needs.
• Modular Design: Each component (STT, text generation, TTS) can be independently configured and customized.
• User-friendly: Easy to set up and use with straightforward installation and configuration.
• Real-time processing: Speech-to-text and text-to-speech are performed in real-time, enabling almost seamless interaction with AI.
• Long-term memory: AI remembers and recalls information across sessions, enabling personalized and context-aware interactions, all while keeping data stored locally.

🖥️ System Requirements

Minimum Requirements:

• VRAM: 0GB (GPU not required)
• RAM: 16GB
• Free Disk Space: 10GB

Recommended Requirements:

• VRAM: 12GB
• RAM: 16GB
• Free Disk Space: 10GB

Notes:

• GPU: Nvidia GPU recommended; CPU can be used if Nvidia GPU is not available.
• AMD GPUs Not supported yet.
• OS: Windows is preferable; Linux and macOS have not been tested yet.

🚀 Installation

Prerequisites

1. Python 3.11: Make sure Python is installed on your system. You can download it from python.org.

2. CUDA Toolkit (if using GPU): Ensure you have the CUDA toolkit installed to leverage GPU acceleration. Download it from NVIDIA's website.

3. Ollama: Ensure Ollama is installed on your system. Download it here Ollama.

4. Miniconda: Ensure Miniconda is installed. Don't forget to check Add to PATH during installation. Download it here Miniconda.

Steps

1. Clone the Repository:

   git clone https://github.com/vancoder1/AsukaAI.git
   cd AsukaAI

2. Install everything using install_windows.bat:

   .\install_windows.bat

3. Wait for the installation to complete.

📖 Usage

1. Start the Application:

   .\start_windows.bat

2. Speech-to-Text:

   • Just start speaking to start transcription.
   • Wait for faster_whisper to process and extract text from your recording.

3. Text Generation:

   • Ollama3 provides human-like interaction. If you prefer a different model, you can change it in models/modelfile.md.
   • Check the Ollama library for available models.
   • For customizing model behavior, refer to the Ollama documentation on their GitHub page.
   • Important: After modifying models/modelfile.md, run the following command to update the model:

     .\update_model.bat

4. Text-to-Speech:

   • Convert generated text to speech in real time while text is being generated.
   • To change the voice, place your desired WAV audio file (approximately 10 seconds, clear voice, no background noise, 22050hz, mono) into the data/reference_voices directory and name the file reference.wav. Make sure you removed previous reference.wav and reference.json files.

5. Configuration:

   • You can customize various settings in the config.json file:

   {
       "version": "1.0.0",
       "description": "Configuration file for AI app",
   
       "stt": {
           "model": "distil-small.en"
       },
   
       "tts": {
           "reference_file": "data/reference_voices/reference.wav"
       }
   }

   • stt.model: Defines the model used for speech-to-text processing. You can change this to other available models such as "base.en" or "distil-large-v3" etc.
   • tts.reference_file: Sets the path to the reference audio file for text-to-speech processing. Ensure the specified file exists and is correctly formatted.

🤝 Contributing

Contributions are welcome! If you have any ideas, suggestions, or bug reports, feel free to open an issue or submit a pull request.

📜 License

This project is licensed under the Apache License 2.0.

🙏 Acknowledgements

• RealtimeSTT
• Ollama
• RealtimeTTS

📬 Contact

For any questions or feedback, please open an issue on this repository or reach out to [email protected].