AI sneeze covers?

[just-a-harmless-snz]

Many if not all of you probably know that people are now making AI covers of songs. This morning I thought of something... what if someone were to plug a sneeze mp3 into one of these AI cover generators instead of a song?

[Blah!?]

I've been biting my tongue on most of the AI stuff going on since I know I'm not gonna change most people's minds, but honestly I think there's something profoundly disturbing about using an AI to reproduce part of a real person's identity for fetish purposes. It's not a matter of how "good" or true to life the AI is able to get, it's more that the subject hasn't given their consent and I can't imagine anyone who'd do this plans to ask them.

[Dawsio]

I completely agree, especially if what AI is covering is another real person.

 

My first thought was "Would an AI even recognize what a sneeze was, and recreate one on command?" I don't know much about the AI audio programs, but I doubt it could.

[Demosthenes]

You can try this pretty easily with something like musicfy.  If you give a talkie wav then what it generates is basically just the same dialogue from the wav but now in a "sing-songy" voice.  The actual sneezes seem to come out sounding distorted or even be mostly edited out (probably because the AI didn't recognize them as words and doesn't know what to do with them).  If you give it a wav that is nothing but sneezes, it often changes the tone so much they don't really sound like sneezes anymore.  You might have better luck trying a speech to speech tool that isn't specifically for making music. 

[Dingle]

ive never really thought of using ai for this purpose but i would like to keep myself away from using real people in my ai stories. while the music is a fun concept it could violate rights.

[just-a-harmless-snz]

On 7/28/2023 at 8:08 PM, Blah!? said:

I've been biting my tongue on most of the AI stuff going on since I know I'm not gonna change most people's minds, but honestly I think there's something profoundly disturbing about using an AI to reproduce part of a real person's identity for fetish purposes. It's not a matter of how "good" or true to life the AI is able to get, it's more that the subject hasn't given their consent and I can't imagine anyone who'd do this plans to ask them.

Absolutely. I'm a hobbyist digital artist and many of my mutuals in the art community, including myself, do not support the use of AI art programs. I've heard stories about people using AI art softwares to generate sexual images of someone who hasn't consented to being depicted in such ways. You can, in fact, ruin someone's life using AI. With these new AI speech tools you can make people say things they never would have said in real life which, in my opinion, is beyond unnerving. I think that we as a fetish community should be aware of the harm these programs could cause if they are misused, especially for fetish/sexual purposes.

On 7/28/2023 at 8:40 PM, Dawsio said:

I completely agree, especially if what AI is covering is another real person.

 

My first thought was "Would an AI even recognize what a sneeze was, and recreate one on command?" I don't know much about the AI audio programs, but I doubt it could.

I have not used any of these AI music platforms but I do know that the AI doesn't recognise laughs, which means that it probably doesn't recognise sneezes either. I wouldn't even call these programs artificial intelligence since they... aren't really intelligent. (AI art programs steal from human artists without consent or credit and mix the different art styles of these artists together, they don't really render original drawings)

On 7/30/2023 at 4:38 PM, Demosthenes said:

You can try this pretty easily with something like musicfy.  If you give a talkie wav then what it generates is basically just the same dialogue from the wav but now in a "sing-songy" voice.  The actual sneezes seem to come out sounding distorted or even be mostly edited out (probably because the AI didn't recognize them as words and doesn't know what to do with them).  If you give it a wav that is nothing but sneezes, it often changes the tone so much they don't really sound like sneezes anymore.  You might have better luck trying a speech to speech tool that isn't specifically for making music. 

I don't think AI will ever be able to understand what all these human bodily functions (laughs, sneezes, coughs, etc.) are. I'm going to stay away from these programs though, I think things are getting out of hand there at the moment.

On 8/2/2023 at 6:26 AM, Dingle said:

ive never really thought of using ai for this purpose but i would like to keep myself away from using real people in my ai stories. while the music is a fun concept it could violate rights.

Same here, I'm asexual and I feel uncomfortable if I think about real people sneezing... I don't really know how to explain it. I do enjoy fantasizing about fictional characters/OCs but real people aren't my cup of tea in that sense. There will definetely be a copyright issue regarding AI music since there is already a similar debate about visual AI art.

[brokensneeze]

On 7/28/2023 at 11:40 AM, Dawsio said:

I completely agree, especially if what AI is covering is another real person.

 

My first thought was "Would an AI even recognize what a sneeze was, and recreate one on command?" I don't know much about the AI audio programs, but I doubt it could.

AI models only can generate the class of "stimuli" they were trained on. So for example ChatGPT, which was trained on text, can't draw or paint, whereas ones that were trained on drawings or paintings can (well, whether they really are "drawing" or "painting" is something one could debate, but at least they produce what *looks* like drawings or paintings.

So if no sneezes were fed into a model for spoken words during training, or even if an audio recording in the training set HAD a sneeze, but without a human labeling at least once, if not dozens of times, "the thing heard at 2:41 in this recording is called a 'sneeze'", there's no chance one would know how to interpret the command "sneeze" or "make a sneeze". The only way an AI art generator can understand "draw a chipmunk" is that many of the images in its training set that had chipmunks had captions with the word "chipmunk" in them. However, if you trained your model only on lots of sneezes, it would presumably generate something like a sneeze, but would never start speaking words, ever.

I've actually thought before about the possibility of making a "sneeze generator". I wouldn't use AI as it's much too "black box" for the kinds of things I like programming, but a "pink trombone of sneezing" might be cool. I've often thought about trying to make my own speech synthesizer, so some day I might try to make a "text to sneeze" lol.

I do agree that while trying to copy a real-life person's voice (or image) for humorous purposes or just curiosity could be cool, doing it for fetish purposes or to falsely assume/manipulate that person's identity can quickly turn unethical.