I did a very short weekly stand-up meeting podcast (17 seconds) using this AI - https://elevenlabs.io/
2 Likes
Amazing 
1 Like
Could probably fully voice a game this way. It might be a bit stiff sounding, but it’d probably also be a lot cheaper than normal.
1 Like
It supports different voices? adult, child…
It looks like it has a LOT of voices, not sure about child though.
It also does different languages - German, Italian, French, Polish, Spanish, Portuguese, Hindi.
1 Like
On the one hand it arguably sounds better than every other TTS voice I’ve heard, on the other hand the intonation’s quite odd and I kind of wonder if the traditional flatness is better than this distracting wrongness.
1 Like
I’ve tried https://elevenlabs.io/ in German with different voices. The intonation differs between the voices. “Charlie” sounds quite natural, while “Bella” sounds like the computer voice in German subway trains.
I’ve used the German translation of @Paul’s text (it’s not a literal translation):
“Willkommen zum wöchentlichen Podcast. Mein Name ist irgendjemand. Mit mir hier am Mikrofon sind Paul, Seguso, Lowlevel, Frenzie, Kaiman und verschiedene andere. Wir sprechen über das Entwickeln von Adventure-Spielen mit künstlicher Intelligenz. Diese Woche habe ich vor allem Sprites erstellt und sie animiert.”
The interesting thing is, that the AI reads English words in English. Even “Lowlevel” is detected as an English word and pronounced in English.
2 Likes
Grmbl, I created an account on Murf AI because they advertised voice changing but it’s just transcribing STT and then passing it to their TTS. So it’s actually not even remotely as interesting as I thought that would be.
https://coqui.ai/ was pretty impressive last time I tried it.
3 Likes
I’ve actually run their STT locally in the past out of curiosity, but their website there suggests it’s also only TTS and STT rather than adjustment of something spoken? I mean more the “I say something, but out comes Sinatra/Obama/a creepy little girl with the same intonations etc.” type of thing.
Gotcha. It seems like voice conversion should work with this: YourTTS: Zero-Shot Multi-Speaker Text Synthesis and Voice Conversion / Blog / Coqui? But I’d have to look into it more to be sure.
A way to fix text on signs -
For some reason, the generative fill thing is really bad at getting anywhere near the correct text, even with tons of attempts.
However, if you just do the signs separately, as a regular image, after a few tries it’ll usually give you the text you want.
You can then Photoshop the sign into a scene.
Sometimes helps to put in “vector art” in the prompt as well.
Example for “fast food” -
Miraculously it even spelled “restaurant” correctly in the instance at the top -
2 Likes
… and the well known Bet beer at the bottom:
1 Like
btw: You could use the “spelling problem” in puzzles or …
… even as a part of the story. For example the hero stranded in a town where everything seems to be weird, including the signs. (Maybe it turns out that he is trapped in a virtual/computer generated world… 
Then you could even use the graphics with weird stairs. )
2 Likes
Sounds like something that could get old pretty fast, except as a diversion for a room or two. 
1 Like
Just posting this to tell you that I have done absolutely nothing beside a few experiments and I have still to read the last 111 posts. 
3 Likes
I’ve basically been working on this sort of idea for a while now - I removed all the Monkey Island things from my game, so it’s now all original stuff.
I’ve got around 20 rooms, 8 characters, 21 items (also variations of items where two items have been combined, etc.), and somewhere between 25-30 puzzles.
The game is wired up so that I can now play through the puzzles and get to the end (I might add a small puzzle or two to the end part).
I still need to add a lot of the dialogues and unique responses to things, the music, some little animations, the start and end cut-scenes, and neaten some things up.
For the main walk animation, I didn’t really like the way any of the rigging/bones systems looked, so I did the main character’s walk cycle by hand, based off looking at other walk cycles. I don’t have a forward/back cycle yet though.
The link about Broken Sword that LowLevel shared has some interesting things -
Particularly this bit - “a tip from a Nvidia engineer about how to use AI to interpolate frames in between hand-drawn key frames prompted a breakthrough”…
That sounds really useful, as then you only have to make two/three main poses and it would fill in the others and then you just fix them up a bit. I don’t know how to get it to fill in frames though.
4 Likes
Can’t wait to play your game! 
1 Like
There’s a new free AI art thing - https://ideogram.ai/
This one can do text a lot better… even if it messes up, you can choose the version you like best and press “remix” and it’ll usually fix any spelling mistakes (and sometimes give you a better design too).
I made a few Thimbleweed Park logos as tests -
1 Like
Very excited! Keep us updated! And/or start a Mastodon for updates.
Are you planning to use AI for generating voiceover and music as well?
1 Like