the audio and gpt-3 requests get cached (gpt-3 burns money like a fossil fuel to run) and over time the cache clears old dialogue to generate new content.Įventually i'll figure out how to automatically accept song requests (after a lot of input sanitization and content filtering) but for now, the playlist needs to be edited manually. though i'm occasionally fixing bugs and pronounciation errors, it pretty much runs by itself now. now it's being hosted on a $20 vps and can generate new audio on its own cpu, and the playlist is in. since then i've removed my desktop and GPU from the equation. The explaination in the link above was one day into development.
new vegas, and xVA Synth to generate the audio for his voice. I'm using gpt-3 to generate the dialogue based on the original transcript from mr. This was a lot more work than it needed to be because i wanted an actual AI radio host to control/voice it. (Buffering, wait a moment.) technical stuff
0 kommentar(er)
