🎵 VocalRender: Score-Native Singing Voice Synthesis

Synthesize 48 kHz singing audio from lyrics, MIDI pitches, note values, and tempo. Upload a clean 2-8 second singing clip as a timbre prompt.

Model · Paper · Code

Prompt Audio (2-8s singing clip)

Lyrics (pipe-separated syllables)

Use | to separate syllables. Use SP for breath/silence.

Pitches (MIDI numbers, comma-separated)

MIDI pitch numbers (0-127). Use 0 for rests.

Notes (duration tokens, comma-separated)

Available: <NOTE_1>, <NOTE_2>, <NOTE_4>, <NOTE_8>, <NOTE_16>, <NOTE_32>, <NOTE_DOT_1>, <NOTE_DOT_2>, <NOTE_DOT_4>, <NOTE_DOT_8>, <NOTE_DOT_16>, <NOTE_DOT_32>

Pitch→Word mapping (comma-separated)

Maps each pitch index to a word index. Leave empty for 1:1.

BPM

Synthesized Singing

Examples