🎵 VocalRender Demo — Turn a score into a singing voice

VocalRender sings a melody you describe with lyrics, notes, and tempo. A short singing recording supplies the vocal color; it does not need to contain the same song.

Model · Paper · Code

What you need

  1. A voice reference: choose an included voice, or upload 2–8 seconds of clean, unaccompanied singing.
  2. Lyrics: enter Chinese lyrics. They are split character by character; you can also use | to control the split, or import ABC/MusicXML below. Other languages are not supported by this checkpoint.
  3. Melody and rhythm: import a score, or press Create word-by-word score, then set pitch and duration. Use + Melisma note when one lyric unit spans multiple notes.
  4. Generate: choose the tempo and press Generate Singing. Or press 🎲 Random score preset to load a ready-made score, adjust it freely, then generate. The first run may wait in a shared GPU queue.

This checkpoint supports Chinese lyrics only. Other languages are rejected before inference. Only upload a voice recording that you own or have permission to use.

Model checkpoint
VocalRender-Pro is selected by default; switch to VocalRender to compare the base checkpoint.
1. Choose a voice reference
Use an included singing voice, or choose Upload my own voice below.