Anime AI Waifu is an AI powered voice assistant with VTuber's model, that combines the charm of anime characters with cutting-edge technologies. This project is meant to create an engaging experience where you can interact with desired character in real-time without powerful hardware.

• 🎤 Voice Interaction: Speak to your AI waifu and get instant (almost) responses.

  • Whisper - openai's paid speech recognition.
  • Google sr - free speech recognition alternative.
  • Console - if you don't want use microphone just type prompts with your keyboard.

• 🤖 AI Chatbot Integration: Conversations are powered by an AI chatbot, ensuring engaging and dynamic interactions.

  • Openai's 'gpt-3.5-turbo' or any other available model.
  • File with personality and behaviour description.
  • Remembers previous messages.

• 📢 Text-to-Speech: Hear your AI waifu's responses as she speaks back to you, creating an immersive experience.

  • Google tts - free and simple solution.
  • ElevenLabs - amazing results, tons of voices.
  • Console - get text responses in your console (but VTube model will be just idle).

• 🌐 Integration with VTube Studio: Seamlessly connect your AI waifu to VTube Studio for an even more lifelike and visually engaging interaction.

  • Lipsync while talking.

*Demonstration in real time without cutouts or speed up. This is real delay in answers.

To run this project, you need:

1. Install Python 3.10.5 if you don't already have it installed.

2. Clone the repository by running git clone https://github.com/JarikDem-Bot/ai-waifu.git

3. Install the required Python packages by running pip install -r requirements.txt in the project directory.

4. Create .env file inside the project directory and enter your API keys

   .env template

   OPENAI_API_KEY='YOUR_OPEN_AI_KEY'
   ELEVENLABS_API_KEY='YOUR_ELEVENLABS_KEY'

5. Install VB-Cable

6. Install and set VTube Studio

   Settings:

   • Select CABLE Output as microphone. Select Preview microphone audio to hear waifu's answers

     

   • Select input and output for Mouth Open. Optionally you can set "breathing" to get idle movents.

     

7. Select your required settings in main.py in waifu.initialize

   Arguments:

   • user_input_service (str) - the way to interact with Waifu

     • "whisper" - OpenAI's whisper speech to text service; paid, requires OpanAi API key.
     • "google" - free google speech to text service.
     • "console" - type your promt in console with text (absoulutely free).
     • None or unspecified - default value is "whisper".

   • stt_duration (float) - the maximum number of seconds that it will dynamically adjust the threshold for before returning. This value should be at least 0.5 in order to get a representative sample of the ambient noise. Default value is 0.5.

   • mic_index (int) - index of the device to use for audio input. If None or unspecified will use default microphone.

   • chatbot_service (str) - service that will generate responses

     • "openai" - OpenAI text generation servise; paid, requires OpanAi API key.
     • "test" - returns prewritten message; used as dummy text for developement to reduce time and cost of testings.
     • None or unspecified - default value is "openai".

   • chatbot_model (str) - model used for text generation. List of available models you can find here. Default value is "gpt-3.5-turbo".

   • chatbot_temperature (float) - determines creativity of the generated text. A higher value leads to more creative result. A lower value leads to less creative and more similar results. Default value is 0.5.

   • personality_file (str) - relative path to txt file with waifu's description. Default value is "personality.txt".

   • tts_service (str) - service that "reads" Waifu's responses

     • "google" - free Google's tts, voice feels very "robotic".
     • "elevenlabs" - ElevenLabs tts with good quality; paid, requires ElevenLabs API key.
     • "console" - output will be printed in console (free).
     • None or unspecified - default value is "google".

   • output_device - (int) output device ID or (str) output device name substring. If VB-Cable is used, you need to find device, that will start with CABLE Input (VB-Audio Virtual using sd.query_devices() command.

   • tts_voice (str) - ElevenLabs voice name. Default value is "Elli".

   • tts_model (str) - ElevenLabs model. Recommended values are "eleven_monolingual_v1" and "eleven_multilingual_v1". Default value is "eleven_monolingual_v1".

8. Run the project by executing python main.py in the project directory.

Depending on the selected input mode, program may send all recorded sounds or other data to the 3-rd parties such as: Google (stt, tts), OpenAI (stt, text generation), ElevenLabs (tts).

MIT