Enunu

[SneezyEevee]

Has anyone here used Enunu to create an ai voicebank? If so, how is it working out? There’s so little info on it so I’m not sure if it can be reliably used or not but I’ve listened to some Utau covers using the plugin and it sounds amazing like synth v or something. This could really take things to the next level
I’d honestly love to download and try it out but the page is in Japanese and I can’t read it to see if there’s any required computer specs for it. I have Windows 8.1 on my old pc I use for utau stuffs. It’s taking OpenUtau pretty well so far but I can’t help but wonder if downloading Enunu would push things a bit too far
I probably should have posted this in Utahelp but too late now lol

Post automatically merged: Apr 17, 2022

I feel like I’m coming up with something new to do Utau-wise all the time but I swear if YouTube doesn’t stop recommending me videos of Utaus using new cutting edge stuff I’m going to blow up my pc for sure

 

[Halo]

I gotcha on the "probably should have been in Utahelp" haha, moved.

Now. There's actually a pretty large amount of resources on this these days.
Notes:
- ENUNU is based on NNSVS so they often get lumped together; sometimes when you can't find a resource for one, there'll be a resource for the other.
- You can download a few Vocal Databases pre-AI training and have a look at their files and labels and stuff, even.
- As of right now, you probably need to not be too shy about some code. There's nothing too difficult, but you do need to do some minor editing of colab notebooks and deal with a fair amount of command line panels if you want to iterate and test quickly lol... with openUTAU its a little easier, but that implementation is still quirky.
- It is better if you know at least a little Japanese to catch some smaller details, but meh. I found it helpful fiddling with NEUTRINO and Sinsy a little just to understand how AI software work.

A Guide to NNSVS / ENUNU

Information regarding the AI synthesizer NNSVS and its related tool, ENUNU. Maintained by xuu.

nnsvs.carrd.co

haru0l's things - Resources

NNsvs Voice making tutorial (WIP)

sites.google.com

The NNSVS/ENUNU discord is a bit messy, but it has a lot of knowledgeable folks and logs known bugs and stuff so I also recommend that. I don't have the link on me, but iirc it's managed by CrazY / oatsu.

Additionally.
- Min specs are unclear off the top of my head, but for renders you can run it all online via colab (I think haru0l has a tut for this), or run via command line and just see how it fares. I can't imagine running it would be too much, though I do think trying to train your own VB might be. It's generally kind of a pain so I do recommend training using the provided colab notebooks or similar.
- Try the instructions for OpenUTAU, it won't permanently break anything, but it is quite heavy right now. Even on decent computers it's got some bottleneck problems from not being fully implemented/ironed out, so there are pretty long waiting/hang times and you have to manually load in the pitch data.
- It's pretty fun! When it comes down to it, you just have to record a lot of songs and take the time to label them, it's certainly no harder than otoing in that respect. Slowly but surely a lot of the other difficult tasks like making MIDI/UST/XML data that matches the labeling (a thing you unfortunately have to do) are being automated. None of the stuff that goes into making the VB before training is intensive on your PC, so even if you had to get someone else to train it and make demos you could still make and configure the data easily even with poor specs.
- It also moves pretty fast. I picked up a bunch of this a few months ago, had to do some uni work, and then came back to find that half the stuff I was doing was outdated LOL

 

[SneezyEevee]

Wow! Thank you so much for the info! This is awesome!!
I had no idea that these resources existed so I’m looking forward to researching this a bit more for in the near future. This will be such a game changer if it all works out
And thank you for moving my thread over to utahelp lol it was bothering me.
OpenUtau is awesome so far, there’s hardly any engine noise on that software, though it’s slightly laggy. I don’t mind it so much since the results make up for it. I’m glad it’s not just me though and it’s cause it’s still under construction, yep I’m always worried about potentially overheating my pc since that would be scary to say the least…

I’m saving the majority of my reading/research for later when I’m able to sit down and study it but I briefly scanned one of the resources on how to make a voicebank and noticed it mentioning google drive. My old pc isn’t exactly a dinosaur but it’s…special. Do I need to have a google drive account/reliable internet access to make the ai voicebank? If so I may have to wait until I can get the pc fixed or save up for a new one. If all possible it would be great if the command line could handle it

 

[Halo]

Do I need to have a google drive account/reliable internet access to make the ai voicebank? If so I may have to wait until I can get the pc fixed or save up for a new one. If all possible it would be great if the command line could handle it

Hm... It'll be more difficult, and IDK if your computer will manage. It will take a while too, as I understand it, even with a pretty powerful comp. That said, it's definitely possible on paper, so worth looking into...? The commandline method doesnt need to you upload anything to GDrive I think. I've never done it that way, so figuring it out might be a bit tough. May just be better to record and label as much data as possible before finding a way to send it to a friend, and have them manage the training.

 

[SneezyEevee]

Paper? I do like writing...
I might just have to wait this one out and just record all the stuff i need for when the time comes.
Thank you for your help and for your time, this will definitely be a project I'm looking foward to doing