The Great AI Voice Training Samples Repository

[Antares]

Hello,

this thread should serve as a compilation of AI voice training files for Warcraft III voices or Blizzard voices from other franchises, making it easy to get the optimal output from the AI tools. Because the Sound resource section isn't really made for this, the Sound / Music Art forum was suggested by @deepstrasz as the best place for this topic.

If you have trained voices yourself and would like to share them with the community, reply here and I will add them to the original post, optimally with a small sample output for each. I will add them as well when I'm not lazy.

If you're interested in using AI voice generation in your map, but don't know where to start, check out this tutorial:

Using AI to create custom voice lines and campaign dialogue


Compiled by Antares A :

Jaina, Arthas, Uther, Malfurion, Illidan, ShandrisWc3, TyrandeWoW, ZagaraSC2, Rifleman, Tichondrius, Muradin, Balnazzar, Anub'Arak, Grunt, Knight

Compiled by ImMikeGaming :

WC3: Dryad, Druid of The Claw, Kael´Thas, Malfurion, Tyrande

Showcase - Malfurion Quest

Compiled by Nyctaeus :

Valeera (Hots), NotDwarf (Overwatch), Kharazim (Hots), McCree (Overwatch), Lunara (Hots), Footman

Compiled by ZethHolyblade :

Priest

 

[ImMikeGaming]

Compiled by ImMikeGaming :

WC3: Dryad, Druid of The Claw, Kael´Thas, Malfurion, Tyrande

Showcase - Malfurion Quest

 

[Nyctaeus]

Adding couple moar, used to generate some stuff for Exodus. I do have more, but these are the most refined ones that provide quality output from ElevensLab. More will come one day.

 

[Antares]

Very nice!

There's a 5 minute nothing in one of the Valeera sound files. And I think that adding the death/laugh etc. sounds is detrimental to the output quality. I would remove those.

 

[Nyctaeus]

Very nice!

There's a 5 minute nothing in one of the Valeera sound files. And I think that adding the death/laugh etc. sounds is detrimental to the output quality. I would remove those.

Oooops...

Apparently I forgot to slide everything leftwards before export. It doesn't seem to affect ElevensLab performance.

Regarding death sounds, laughter and other expressions like that, I removed them from McCree and Torbjorn files and left them in the others to test ElevensLab algorithm. It seems to be ignoring them, as there was no strage things like that in output. Valeera is source for major character, so I generated many lines using this source and it seems to be adequately tested.

What really messes up the results is the background noise. Also things like spell sounds and everything overlapping the voices. This often gets crushed with the voice itself.

Anyway, here is the fix. I removed death sounds and laughters as well.

 

[Antares]

The AI probably has an understanding of what laughter is, but I doubt it has been trained on death screams . So, including the former is probably fine and might even be beneficial when you want to generate laughter (sometimes you get sighs when you write "hm", sometimes it's "haitch em" , probably the same with "hahaha")

 

[Kyrbi0]

I'm very excited by the modding prospects of AI Voice Generation, and glad people are jumping on this nascent niche to try & organize the chaos. However I'm a little confused; what exactly is being done/compiled here? Is this a collection of "wc3 soundsets turned into AI voice training modules (i.e. "I plugged all the orc grunt lines into ElevenLabs & here's the voice file output" (so anyone can make the Orc Grunt say something without having to do the actual training themselves))" kinda thing? Or am I misunderstanding?

 

[Antares]

These are training files for elevenlabs, so what you use as input for the A.I. who then produces new voice lines based on those files, hopefully with a good quality output. If the training files are just the unit response sounds, it's trivial and you can just compile them yourself, but it's always better to use campaign voice lines if possible, because they're longer. And for some characters, it may be harder to find good voice lines that are longer and without distortions. For example, for Tyrande using the Warcraft 3 sound files leads to artifacts because of the ambient background sounds in all her voice lines, so I compiled the best ones from WoW as a replacement.

 

[Kyrbi0]

Ah, so this thread serves to help save people the trouble of compiling the requisite collections of voice input files (for the use in AI generation)? Very cool! I love it.

 

[ZethHolyblade]

I'd like to give something back. Here are the samples I used for the Warcraft 3 priest voice with Elevenlabs.

It uses its 3 dialogues from the whole game and his unit sounds. Know that his pissed3.wav had the magic wand sound removed and the pissed6.wav was slowed down to sound like every other quote does.

Here's a sample of how he'll sound like: WC3SC1 Priest/Crusader sample