(500 epochs old) (7/2/23) Trained with 10 minutes of high quality voice clips of Elsa from Frozen Fever, Olafs Frozen Adventure, and Frozen II. Including her hit songs from the films. Elsa in this model, she sounds old or aged like in Frozen 2. You have to play and experiment with the search feature ratio to find the right sound for your audio clip. Enjoy! (500 epochs updated 7/23/23) more cleaner and less robotic. (500 epochs updated 8/27/23) 5%-10% cleaner than second one Note: Elsa's vocal range is about 150Hz-630Hz Average vocal range spoken: 180Hz-230Hz Elsa is cute 💯

WE MAKIN OUT OF AHTOHALLAN WITH THIS ONE 🤣🤣

her mother would be proud

Most overrated Disney Princess.

Will say the VA is very underrated. Her singing is consistent good. So the model should be fun to play with different song types.

more raw sample

more improvement coming, the "sibilance" harsh sounds of elsa will be fix soon

Life's Too Short, together with Anna model 💯

Elsa (Frozen) RVC V2 40K (500 Epochs)

updated version on my old elsa model, sibilance have massive improvement and fixed a noisy background

More updates on Elsa model soon lmao. Just slightly cleaner than the 2 previous models. Idk, still not satisfied about the 2. needs to be fixed slight distortion - tearing or delay effect on breathes (robotic breathes) - slight reverb

well uhh, sorry for some long delays about this Elsa model update, I am having a humming sound issue in ALL colabs and ALL pitch extraction algorithms, and it's is badly annoying as hell. I am 100% sure this is not dataset related issue. Colab is the problem here and I have been asking developers about this.

Would you prefer if I tried with local to see if it helps make a difference?

nah no need, I have a fix now hehe

UPDATED! (kinda). Here is a sample, this is studio acapella sample session. This is hybrid model, I reuse the second .pth and replace it with a cleaner .index

Well the breathing is almost impossible to improve as it's a random noise so I need to sacrifice them for quality. 💯

narrative test -4 pitch mangio crepe 20 hops

Elsa and Anna duet, but I can't harmonize them 💯

updated again 9/17/23 haha, I just use previous model but I added the disney infinity dataset in the index file, so file size is a little larger now

tip: play around with search feature ratio, I suggest not to put it at 1 often it creates a little popping sounds or chops or artifacts.

one last sample on latest model

Updated the model again cuz why not (9/24/23) but replace pth and index now, and also removed disney infinity dataset because it was causing issues. I'd say 96% close to studio quality, we're getting better and better Train Info: mangio-crepe 6 hops, 4 batch, 500 epochs, 33.5K steps, 12 mins dataset

samples

Voiced by Idina Menzel.