Worktone-deaf
Hand it lyrics and a score, and it sings them: an AI voice on the melody, a synthesized band underneath.
The score is one plain text file: melody, lyrics under the notes, and the band's parts. A synthesizer plays the band; DiffSinger, an open model trained on real singing, sings every syllable on its note. It checks its own work by transcribing the song back and measuring every note's pitch. The name is a joke: the pitch is the one thing it gets right every time. The first try bent a speaking voice onto the notes, and it sounded like auto-tune.
Listen
The work behind it
- Lines written
- 4,876
- Files
- 31
- Days worked
- 2
- Commits
- 7
- Started
- 9 Oct 2026
- Last change
- 10 Oct 2026
What it would have cost
$18,000to have this built for you
$500what it is worth to own today
Prototype-heavy singing-synthesis tool; no users/revenue, value mostly as reusable code.
Average of 1 day of estimates, the last on 10 Oct 2026. How this is worked out.
All projects in the showroom · Ask for a seat
Measured 10 Oct 2026.