With the introduction of ChatGPT-4o, I found that you can add some customisations to how ChatGPT response your question.
I decided to giver her a personality (002 in my case). Similar to character.ai 🌚.
Also, I thought of making it as a Assistant Chatbot like Google Assistant on Android phones. But as I went on building this project, I found that that cannot be done. Due to the limitations of libraries used in the project. 🫤
- STT - for voice input
- chatgpt.com - for making response (coz LLM on lapot takes too much time to respond. and is not as good as ChatGPT) (* USE IT ON YOUR OWN RISK *)
- TTS (with expressions) - for audio response
- NVIDIA GPU WITH 4 GB OR MORE VRAM
- IF VRAM IS LESS, YOU CAN TRY SMALLER STT MODELS WHICH ARE ABOUT 300 - 500 MB
- OR IF NO GPU, YOU CAN TRY pyttsx3 (but it kills the purpose 🫥) <<--- THIS PART IS NOT CODED YET 😬
- ATLEAST 10 GB SSD SPACE. (HDD WILL BE SLOW)
- STT MODELS ARE AROUND 2 GB AND TORCH+CUDA AND OTHER STUFFS WILL TAKE ABOUT 6-7 GB.
- 16 GB OR MORE RAM (I HAVE NOT CHECKED THE PERFORMANCE ON 8 GB RAM)
- MAKING A NEW ENVIRONMENT WILL BE BETTER
- LOOK HERE TO KNOW MORE AND FOR INSTRUCTIONS
- MAKE A
.envFILE IN THE ROOT FOLDER OF THE PROJECT WITH THESE ENTITIES -USER_EMAIL : [email protected] USER_PASSWORD : your_password- OR YOU CAN MANUALLY PUT YOUR EMAIL IN THE CODE (not recommended)
- PASSWORD CAN BE KEPT EMPTY, DURING THE RUNTIME, THE PROGRAM WILL ASK FOR PASSWORD IN THAT CASE
INSTALL THESE PACKAGES -
- dotenv
- undetected_chromedriver
- selenium
- getpass
- RealtimeSTT
- torch + cuda
- pygame
- pynput
- TTS
- ChatGPT account
RealtimeSTT may not work correctly with recent version of torch+CUDA, just google about it, you should get some solution.
Make sure that Torch + CUDA is setup correctly - check this
Now at the end, you will need a reference "high quality" voice that you want your AI Assistant to emulate. (Find your Waifu/Husbando 😉)