Howdy folks! Bearded recently posted about using the AI platform ElevenLabs to create some voicepacks for characters he was working on. I was really impressed by what he produced, so I decided to do a bit of experimenting with it too. It is pretty amazing.
https://elevenlabs.io/
I built a voice for the TMNT character Mondo Gecko and played around with it a bit. The ability of the software to create a character voice essentially out of whole cloth is incredible, and the text-to-speech mode allows you to create and tweak voice lines in that voice, which is tailor-made for voicepacks. Apparently you can even use sample audio to create voices, which means I'm going to have to eventually revisit all of my DC voice packs, haha.
This tech is, quite simply, a game changer for us. AI is still probably going to destroy our civilization, but at least we can have fun with it before it does, ha! :P
I've found the V3 the most useful, as it allows you to add "direction" to how the program "reads" your text-to-speech lines. Adding things like [cocky] and [excited] generate better "readings" and allows you to tweak them easily.
:EDIT: It just occurred to me: if I put the time and resources into it, I could create fully-voiced campaigns for my mods!
Here's the Mondo Gecko voice I generated:
https://storage.googleapis.com/eleven-public-prod/database/user/VI8vLcB9yIP0yctxQaVbJ2bPlJN2/voices/3CKDuyEyvAmxwFA5C8YN/c3f475c5-ccce-4300-a367-186600e45533.mp3
I'll post the voice pack when I finish it.
That is an amazing vp. I think I was using the wrong version, for Ice when I [excited] she would just say the word excited. Do you download it line by line, or all of them at once? I did Plastic Man all at once, that's why his cadence is off.
I suggested that it was best to use eleben labs, "characterai" is other app that do the same, is even better if the sample is good, I've been using it for a few months now to have some characters shout or say iconic phrases that I couldn't get the audio for or that simply didn't have an audiovisual version to hear their voice.
Cranlox, thank you for starting this off with that suggestion. These programs are something of a revelation for us hobbyists.
Bearded, when you're working on text-to-speech, you have a button on the right in the middle that says "Try V3" and warns it is an alpha build. That is what gives you the option to add "direction." Without it, as you say, the AI just reads the directions as well.
The sample I posted is the sample created when the voice was generated, but I got similarly good quality results using V3. Here's the raw audio of the attack lines I generated as a test. They're not perfect, but they're pretty close, and I could have tweaked them further if I wanted to. I'll do some more testing tonight, perhaps try to create a full VP.
https://www.mediafire.com/file/eqk5nip28jjmtxu/MondoGecko_Attack_Lines.mp3/file
I just wrote multiple lines out, added direction, and then the AI read them one after another. I downloaded the resultant file. Notably, you get three free regenerations if you aren't happy with a delivery, so you don't necessarily have to spend credits to tweak (though you can't alter the text before regenerating).
With the "character.ia" app, you must create a character that is an actor who follows the script you give him without any feedback or interaction questions. The model is designed to create a character and freely avoid speaking the text. That way, if you tell the robot to understand itself, it can only interpret an actor acting the text you give it. This is the only way to "master" what it says. The sample must be 30 seconds long to be optimal; for example, 30 seconds of someone's speech, and the voice will be captured and cloned for use.
You can steal audio and voices from interviews, movies, programs, and television contestants. All you have to do is gather a 30-second sample to ensure it is optimally captured from the same person speaking.
I'll have to play with that one sometime as well.
Well folks, here's my proof of concept. Here is a full voicepack for Mondo Gecko, and I'm pretty happy with it.
https://www.mediafire.com/file/2yzhfe3jep04r2u/Mondo.zip/file
Let me know what y'all think. This took several thousand credits out of the free ones you get for signing up, but I imagine I could do it more efficiently for a second character. I'll probably try to voice the entire team of the Mutanimals in the coming weeks.
Whoops, I had the Attack lines wrong:
https://www.mediafire.com/file/2yzhfe3jep04r2u/Mondo.zip/file