This repo is in WIP state. Additions and improve pending are:
The aim of this project is to create an AI live host that ideally be hosted on social media sales platform aka Shopee/Tiktok Live.
This idea is to use RAG for product information and pass it to an LLM where it'll download a .wav file which would be played into a virtual model therefore 'speaking' to us.
This is reduce the workload of an actual person promoting the/a product.
We do not intend to replace human creative works with this, just to assist them in their downtime.
- An indexer directory called
pre_processing/where documents and csv should be located to create a vecstore. retrieval/to fetch the data from vecstore and pass it to the LLM.TTS_voice/to generate a.wavfile intovoicebank/based on LLM output depending on whichvoicebank/voice used.VTube_connection/contains scripts for connecting toVTube Studioavaialable onSteam.main.pyis the python script that merges everything together (Currently it just loads the vectors, runs the LLM and download a .wav file)