kronosta/DimensionalStack-RVC

These are RVC models for the custom voices I crafted for some of the characters in my fictional multiverse called the Dimensional Stack (or alternatively Quatrammotile).

Essentially I found a strategy where you can concatenate audio files of voices to mix them, and use XTTS_v2 to randomize the voices a little while keeping overall tonality (because it's kinda bad at cloning tbh). After crafting fitting voices for my characters usable with CosyVoice, I generated about 4 minutes of output and fed it into the RVC trainer. Note: Pitch-detection has been enabled so these voices can theoretically sing, that's not to say they're very good singers (they're not really, the voices are too abrasive).

Voice Descriptions

• uncovesseltuxe
  • Composed of a complex mixture of Karl Jobst and a brief snippet of Ccarretti. It's clear, mostly neutral, and a bit nerdy. When singing he turns into a harsh-voiced country grandma for some reason, I'm not sure why (well, it's obvious that it's giving him the same singing voice as his talking voice, which is not how it normally works, and it just so happens that his talking voice sings like that. But you wouldn't anticipate him singing that way based on his voice. Tangent over.)
• ievokt
  • Composed of a mixture between Jan Misali and Matt Rose, both pitched down 2 semitones before cloning. Note that the AI's interpretation of this mixture is nothing like its components. Ievokt's voice is harsh, gravelly, and lends itself well to aggressive tones.
• thaneophyros
  • Thaneophyros' voice is literally just Geosquare with a few intermediate cloning steps that change it a tiny bit. But it's still mostly just Geosquare. The voice is calm, warm, and low-pitched. I have not tested the singing on this model, but I think it might work better due to it being a much more simple voice.
• thaneophyros-reconfigured
  • This is completely unrelated to thaneophyros, for reasons relating to the fictional multiverse. It has a nasal, boxy quality, and is calm, moderately warm, and mediumly pitched.
• alanite
  • Alanite is not a character in the stack, but it gets lumped in here since it was also constructed out of a weird chain of AI clones. It was originally a really inaccurate clone of my own voice by XTTS_v2. It's low, robotically neutral, and very cool-sounding.
• banqrrougt
  • Pronounced as "Bancroft", it's a sarcastic, grating female voice that sounds like it's coming through a low-quality speaker.
• macrelydve
  • Pretty much just a clone of jan Misali, the YouTuber. If you listen to one of his videos, that's pretty much exactly how this sounds with a few minute differences. Gravelly and cheerful.
• outzschcrad
  • Smooth, neutral, approachable male voice.