Project · In development

MÉZIO Voice Engine

Your own voice on your own songs, trained and run locally.

MÉZIO Voice Engine is a local Windows application that replaces the AI-generated vocal in a song with the owner's own voice, and keeps training that voice as more recordings are added. Everything runs on the PC's own GPU: no cloud, no accounts, no subscriptions or paid APIs.

Why it exists

To make songs that sound like their author, without sending a voice or a song to anyone's servers.

What was built

  • One-button conversion: Conversion, timing alignment, tuning, vocal effects and mix in one pipeline, with each stage kept for comparison.
  • A voice that improves: Fine-tunes the voice model further as more recordings are added, with versioned voices to choose from.
  • Side-by-side listening: Final mix, processed vocal, raw conversion and the original, next to each other.
  • Reproducible renders: Every render is kept with a manifest of the voice version, reference and settings. Nothing is overwritten.

Still in development

  • A personal tool. Not available to download. Some of its components are GPL-3.0, which would set the terms for any distribution.

Technology

  • Python
  • PyTorch (CUDA)
  • Seed-VC
  • FFmpeg
  • Tauri

In the Journey

Related