AI Music is becoming advanced, yes, but words could be implemented…
Multiple recordings of voices from different genres could be recorded with all the words in the dictionary and some slang. Cusses and inappropriate language could be excluded. Then, have them recorded in different tones, feed them into the system, and have the AI find the best fit by comparing the tags from the request and the ID of the word. This could help it sound more realistic.
A more complicated solution would be using today-music-generation tech to process the sounds into words, but it’s pretty out there with today’s technology.
Again, you could record instruments in different pitches and do different things to ensure the notes’ diversity, then incorporate those like with the words, or you could do the out-there generation.
You could overlay the normal generation, words, and instruments to create a full song or do them separately to make a good sound.
You could also record sound effects, which would add to the overall feel of the video, such as activity effects, which are caused by something happening, or background effects, like birds chirping.
These are but some ways that sound could be added to OpenAI’s Sora’s Video Generations.
Building on my previous idea and another by h.alesso (See Below), I think lip synchronization would work, especially by h.alesso’s website (Also See Below).
As well, the location of the sound could work as the sounds sync with the source of the noise. (ex. Car, Bird Chiping.)
This could help the video and sound be as realistic as possible.
As well, fakes could be made with SORA, so by tagging the images with something that shows it was generated, people can’t take advantage or even commit crimes with SORA video generation.
