BAD APPLE!! (Orchestral arr.) // DiffSinger PL Megamodel demo

Hello Hello ! I've been experimenting a little with DiffSinger (also know as Diffusion-based SVS) and wanted to share the results !
As said in the video, to respect the wishes of voicers I am keeping the model private thus not releasing it (even for pre-training purposes).
"Data" annotations without a language code are Polish by default by the way
All vocals in the video are produced from the same model
I think most of the story and credits are already present in the video so I won't add much to the description I think,,,
- - - - -
Credits:
BAD APPLE!! - Team Shanghai Alice
USTX / Mix - PixPrucer
Polish lyrics - • BAD APPLE!! (Polish Fa...
Tagalog lyrics - • 【KURI】Masamang Mansana...
Instrumental - • 【Touhou】 -Bad Apple- (...
Choir adlib inspiration - • 【Multilanguage Cover】T...
Data / Voice providers:
Polish - SzTJ, PixPrucer, rainy, hq, vieri, Scarfmonster, Quake,
Japanese - rev
Tagalog - UtaUtaUtau
Model Specs:
trained on DiffSinger refactor-v2 branch
15 singers divided into multispeaker embeddings
10h Polish singing + 0.9h JP singing + 0.1h TL singing
100k steps from scratch

[@matty_mroz]

Jestem pod wrażeniem głosów: FIlip, Karasu Yuutsukoe, Mat i Pix AI 2. Są to dobre głosy syntetyczne, zbliżone jakością do ludzkich. Świetna robota! Nie mogę się doczekać więcej utworów z tymi głosami.

[@xuusynth]

THIS IS INCREDIBLEEE they all sounds so great 😭💕 you did amazing as always

[@PixPrucer]

Awawaaaaa thank you !! 🙏

[@hair.ballsonaro]

omgg piekne glosy 🥺nigdy nie myslalem ze doczekam PL vocaloidow lepszych niz og yamaha, gratki i podziwiam!

[@PixPrucer]

Wielkie dzięki! 🙏

[@just_roo5483]

Brzmi wspaniale ❤❤

[@_turning_point8939]

cool!

[@chevieri]

💜💜💜💜💜💜

[@user-xr8hn2fk7e]

Hello, PixPrucer. I was shocked to see your demo video. You did a great job and brought me a lot of inspiration. I hope this video can be shared with more people. Can I forward this video to other platforms now? I will mark your UA-cam address and look forward to your reply.😀😀

[@PixPrucer]

Hello hello !! I have not noticed this comment until now sorry for that 😭
Yes I am fine with this video being re-uploaded to other platforms! As long as you also include the original description and link back to the original upload

[@waltervd_o3o]

wait how do you make voicebank models for diffsinger ?

[@PixPrucer]

That's a question too complex to explain in a UA-cam comment ! And thus I'm sending you back to the document I made specifically answering your question
docs.google.com/document/d/1uMsepxbdUW65PfIWL1pt2OM6ZKa5ybTTJOpZ733Ht6s/edit?usp=drivesdk

[@Twiddle_things]

What. The. Fuck.
(I mean that in the best way imaginable)

[@SlayerReduxUTAU]

How does Diffsinger make 1 minute of singing sound so high quality?

[@PixPrucer]

Batch data carrying

[@patrickbernardidefreitas402]

I wanna make an AI VB now… I have no idea on how to make one(or even the time rn… hopefully next year)

[@PixPrucer]

Fortunately for you I have a whole process documented! It can be read here: docs.google.com/document/d/1uMsepxbdUW65PfIWL1pt2OM6ZKa5ybTTJOpZ733Ht6s/edit?usp=drivesdk

[@patrickbernardidefreitas402]

@@PixPrucer omg thank you!!

[@user-sz3bq3pb5s]

Is it free or for sale?

[@PixPrucer]

DiffSinger is open source, anyone can make their own voice models using it