Moozz DiffSinger Synthesis

Score-driven singing synthesis — real inputs, real progress, real output.

Inputs

DiffSinger needs a melody run (pitch/notes) and an alignment run (timed phonemes) from the same song.

Eligible runs

▸ Advanced: provide inputs directly

project_id

alignment_translated_run_id

...or upload / paste the real artifact JSON directly (used instead of the run_id above if not empty):

melody_run_id

...or upload / paste the real artifact JSON directly (used instead of the run_id above if not empty):

Alignment
+ Melody

→

Duration
model

→

Pitch
model

→

Acoustic
model

→

Audio

Language (auto-detected from the alignment run — override if needed)

Voice — real trained singers, defaults to the backend's own pick unless changed

Voice type

Singer

Technique

run