VOX GARBLER

« RYON EVERETT'S CYBERSPACE JUNKYARD

ML-DRIVEN SPIRITBOX

A curio, sampling and mixing some of video gaming's cheesiest voice lines into a spirit-box tool for producing outlandish, yet familiar text-to-speech. A passion project, anchored in both learning and entertainment.

Dev logs

Summary

Here for the data-labeling and mining game? Go check out: Garbler Mines

THUNKER: If you've ever been surprised by your own voice when recorded, just imagine how I feel knowing it's my only voice.

STATUS: PRE-ALPHA (WORKS GREAT SOME OF THE TIME)

A curio, sampling and mixing some of video gaming's cheesiest voice lines into a spirit-box tool for producing outlandish, yet familiar text-to-speech. A passion project, anchored in both learning and entertainment.

Back of the box

  • Voices from the past: Spiritbox, but its cheesy gaming voice lines.
  • Text-to-(garbled)speech: Input text, get garblerized sentences vocalized.
  • Spiced for life: Variety of over 800 000 words and 1000s of characters.
  • Curated for flair: Sourced from 25+ gems in Ryon's game collection.
  • Eventually usable: Upgrading regularly with improved boundaries.
  • Garble your way: Highly customizable render engine
  • Confuse your friend: Stitch together almost recognizable phrases.

Systems

  • Automatic speech recognition AI models
  • Local & autonomous model execution
  • MFA + fine-tuning + domain fitting
  • Tunable voice render
  • Library explorer
  • Golden training reference generation and labeling
  • Game audio extractor
  • Custom APIs & render commands
  • Typescript, python, pytorch,
  • Open-weight ASR models
  • llama.cpp