/vt/ - Virtual Youtubers

>>68379616
i did, i'm not very good at it but it's recognizable so good enough i suppose

>>68380139
to cut down on latency it's as local as it can be. there are certainly people who've paid more attention to the snippets of architecture info he does reveal than me but afaik the main LLM AI certainly runs locally, and while historically he used azure voice recognition and their TTS for neuro's voice and elevenlabs for evil's voice i believe those both offer higher tier local options for latency sensitive operations
you can compare it to something like what dougdoug does for some of his streams, which literally is just glueing online services together, there's a lot of lag between recognizing his voice -> generating a response using chatgpt -> sending the response to elevenlabs TTS -> getting the response back and playing it. not even close to real time