it looks like the guy used 11 labs for the voice, and the code looks like Python
the algorithms seems to be: generate voice when detection of fuck, while the dirty talk is anybody's guess.
the problem with 11 labs is their API requires an internet connection, so the AI might not be offline.
his code features detection of the movement attributes(in, out, depth, direction)
the conversation algorithm seems to be a combination of chatGPT and possibly the PerChance chatbot, and hardcoded responses for the fuck part.