/leek/ - >does nothing >wins How does he do it?

kanru-hua-….png - 1167.51 KB (878x878)

Mikudayo Sat, Jul 18, 2026, 02:47 AM No.30882

>does nothing
>wins
How does he do it?

Mikudayo Sat, Jul 18, 2026, 02:59 AM No.30887

>>30882
Not to sound like a Kanru shill but his engine just works, and you're paying less for a product better than what other companies offer

Mikudayo Sat, Jul 18, 2026, 05:14 AM No.30898

>>30882
Teto

Mikudayo Sun, Jul 19, 2026, 07:10 PM No.30999

>>30887
DiffSinger is better than Synth V but you guys aren't ready for that conversation yet

Mikudayo Sun, Jul 19, 2026, 07:36 PM No.31007

>>30999
I partially agree, some DiffSingers can sound even better than SVs. But that depends on how they're trained. Poorly trained Diffs will sound bad, especially if they use the default vocoders.

Mikudayo Sun, Jul 19, 2026, 09:56 PM No.31018

>>30999
I feel like SynthV works way better on AMD cards, but Diffsinger is way better on NVidia because they can actually handle the load of the software, and therefore make their own high quality VBs with enough patience and resources.

Mikudayo Sun, Jul 19, 2026, 09:57 PM No.31019

>>31018
I have an AMD card and have no issue running DiffSinger but it definitely does suck for training your own voicebanks

Mikudayo Sun, Jul 19, 2026, 10:19 PM No.31022

>>30999
I'd be surprised if any engine had as many tuning parameters as SynthV

Mikudayo Sun, Jul 19, 2026, 10:32 PM No.31024

>>31022
Diffsinger definitely does not. Toriko sounds as high quality as a SV, but she can't be tuned like one, for no other reasons than the resampler not supporting most if any flags. I still really love her and she's free though.

Mikudayo Sun, Jul 19, 2026, 10:35 PM No.31025

>>31024
>but she can't be tuned like one
We're close to that. Tunelab Q can already "autotune" her like in SV to make her sound more realistic and the tension flag can make her voice stronger/weaker.
If she had more singing modes and support for more languages she'd be very close.

Mikudayo Mon, Jul 20, 2026, 10:27 AM No.31034

>>31025
I was moreso referring to the piano roll tuning (stuff like breathiness/gender/tone shift and the like) and vocal modes, the latter of which is possible if the Diffsinger is trained for them, but ultimately not a big deal to lack to me since my first voicebank was for VOCALOID2. One of the things I actually prefer about Diffsinger to SynthV is manual tuning, I just wish my computer didn't take 20 minutes to render 20 seconds of vocals, if that.

Mikudayo Mon, Jul 20, 2026, 12:13 PM No.31035

>>31034
Now you have me looking into distributed computing to see if we could set up a network to share processing power. Can rendering be done via command line? If it can can this might be possible, maybe, but I don't know enough.

Mikudayo Mon, Jul 20, 2026, 02:27 PM No.31041

>>31035
>Can rendering be done via command line?
I think so. Didn't someone make a TTS voice changer using OpenUTAU with command line a few months ago?.

Name
Options	Fortune Sage Nonoko
Comment	Max comment length: 0/4096 Poll duration:
Emotes	Show Emotes
Select	File / Embed / Poll / Tegaki
File	Drag files here to upload or click here to select them, or paste an image
Captcha	Verified — you can post. Name the character shown in the picture. Loading image…
Password	(for deletion)
Please read the Rules and FAQ before posting. Supported file types are: GIF, JPG, OGG, MP3, MP4, PNG, SWF, WEBM, WEBP. Supported embeds are: YouTube, Nicovideo, Twitter/X, Bluesky, BiliBili, Vocaroo. Maximum file size allowed is 20MB. Images greater than 250x250 pixels will be thumbnailed.