cursed fact: Adobe Podcast's "Enhance AI", a tool for noise removal and voice boosting, firmly believes that any audio you give it *must* have human speech.
If you upload, say, vocal-free chiptunes playing on a Game Boy, it will *find* the speech.
cursed fact: Adobe Podcast's "Enhance AI", a tool for noise removal and voice boosting, firmly believes that any audio you give it *must* have human speech. If you upload, say, vocal-free chiptunes playing on a Game Boy, it will *find* the speech. 109 comments
@lazerwalker That kind of slaps, though...
0
0
20 Oct 2023 at 19:53 | Open on ak.angelstrapped.com
@lazerwalker it sounds like the fake speech sounds in older CRPG on the SNES. I kinda like it @lazerwalker reminds me a bit of the hatsune miku guitar pedal that turns every note into a japanese phoneme. @lazerwalker there are conspiracy folks frantically pointing at video compression artifacts "proving" world leaders are lizard people. We already had panics about satanic messages hidden in recordings, if you play the record backwards, etc. I'm expecting someone's to start over analyzing media made with this, and find "hidden messages" if they run it through the right filters. Is this going to give us a new moral panic like the Satanic backmasking on heavy metal albums? It is time for it, I guess. @lazerwalker in my days we used to play records at 45 rpm to get our demonic messagesโฆ younger generations donโt know how easy they have it @lazerwalker I hope someone uses this technology as the voices for characters in a videogame, because this sounds awesome. @lazerwalker I don't know if I'm flattered or concerned that it sounds a lot like I do when I'm straightening up and forgot to turn on some music first. I bet it even gets the same "bahma nama nama" phonemes that I use for the SMB underground theme. Sheesh, this is pretty awkward for gamer-ish dads. @yuliyan @lazerwalker Agreed, I was hoping it would fade into the original at some point. @lazerwalker I knew what I was getting into when I clicked that link, but now I too am cursed to have heard that. ๐ @lazerwalker that is _epic_. it's like leaks from dreamspace set to music i love it. @lazerwalker I think Jim OโRourke was doing this or similar on his latest LP https://steamroom.bandcamp.com/album/steamroom-61 @colin_howells @lazerwalker Sounds like what granular synthesis might sound like if seeded with spoken word too. A few different ways to skin that cat. I was surprised to hear that confrontational โcut my earphonesโ vid about 6 minutes in. @Chancerubbage Ha yeah that was hilarious. (He does tag his stuff as โcomedy' even though it's electroacoustic/ musique concrete.) There's an LP a few releases back which was a Fourier transform of Gould's Goldbergs which was also extremely boss @colin_howells oohโฆ I actually enjoy Fourier transforms- the lower resolution they are the better. I havenโt kept up with him much since his being with everyone everywhere at once in the 90s. @Chancerubbage He's had a killer run with these Bandcamp releases, I really encourage you to check them out, I've got like 25 of them This massive interview from 2020 was fascinating https://toneglow.substack.com/p/014-jim-orourke @colin_howells it is probably the first time Iโve heard someone sample a viral video ( infinity mirror tik tok sailor chantyโs donโt count, that stuff is โbuild on THISโ by design) I love this Fourier analysis of Robert Frost reading a line of his poetry. Four sine waves, just jigger their frequency and amplitude. Now if you could reverse engineer what R2-D2 was sayingโฆ. http://www.columbia.edu/~remez/musical-and-poetic-sine-wave-speech.html @lazerwalker I feel like you may have discovered a new art form here... or a way for ghost hunters to claim old games are somehow haunted by the souls of those who created them. Like most software. @lazerwalker โAIโ โHallucinationsโ. @lazerwalker this is great; also, I canโt remember if youโre already familiar with the work of Diana Deutsch, but, if not: @lazerwalker This heralds the next generation of "playing the rock album LP backwards to hear satanic voices." @lazerwalker @lazerwalker A journalist @Ruhrnalist uses whisper to transcribe interview, but only the answer track. He complains from time to time that it invents questions between the answersโฆ. @lazerwalker now we get to find out what the windshield wipers were actually sayingโฆ @lazerwalker @lazerwalker these would be cool for an 8/16 bit game to use as voice sound effects, if the song could be removed somehow. @lazerwalker this is great wonder what would happen if you gave it the midi of a song with lyrics @lazerwalker thats the exact sound i make to myself around the house when no one else is home @lazerwalker wow lowkey brilliant tho. I really wanna hear it chopped up and sprinkled over like a deep minimal tech-house beat @lazerwalker this seems consistent with my theory that at least some AI are being possessed by the Fae. this is my favorite micro-genre of joke remixes and i really hope it takes off soon. somebody put radiohead creep through enhance ai and ive been thinking about "so ferkin special" ever since All I can hear is "Turn me on dead man, turn me on dead man..." :-D https://www.huffpost.com/entry/reverse-speech-revolution_n_5288920 @lazerwalker reminds me of years ago when people would play popular music backwards for subliminal messages, this wouldโve been a riot @lazerwalker Love that. Like the early days of image recognition where you could force the DL to find hallucinatory dogs that didn't exist in anything. @lazerwalker That's why I gave it the sound of a flushing toilet and of the dishwasher. Ran that through #ElevenLabs dub feature and told it to output English. The result is... Well, embedded in this weirdness I created: @lazerwalker I guess it'll try to find some "satanic lyrics" in it... I wounder how well it handles extremely agressive #vocoders like #TWELP, #MELPe or #Codec2... @lazerwalker this reminds me of the game #Oxenfree. Not directly to anything in the game, but this reaaally seems like a mechanic the game could have used somehow. So it's got psychosis. There are modern drugs that control it pretty well in humans. One is quetiapine, also known as Seroquel as a trade name. SeroQueL ๐คญ @lazerwalker @lazerwalker @neckspike lol, that's grand and the absolute opposite of what tools like KRISP do (if you pass any music track through KRISP it turns it into a mostly acapella version) @lazerwalker @esther I learned about that recently. I think itโs called overfitting because the AI is trained only by rewarding output that is speech. Itโs the same for AI translation, and itโs the reason why, if you set Google Translateโs input language to Latin, and type random made-up words ending in um or us, it will โfind senseโ in them. Itโs worth trying for yourself, it would be hilarious if it wasnโt causing do many problems. @lazerwalker itโs parAIdolia! Pretty excellent example of the not always so subtle human biases can infect non-human systems. Someone please feed โEnhance AIโ some r2d2 dialog. I want to find out what Artoo was telling 3p0 @lazerwalker this is what i sound like when i'm working at my computer with headphones on @lazerwalker In Asimov's "Three Laws" setting, I suspect there's assholes who give robots orders which cannot be obeyed (like "translate this document" when the document is random gibberish) just for funsies, because humans are scum. Of course, the risk is that the robot will find the translation and then summon back the Elder Gods. (Something close to this, unintentionally, did happen in one story: https://en.wikipedia.org/wiki/Robot_AL-76_Goes_Astray) |