-4

Throughout history mankind has created musical genres. This probably had its origins thousands of years ago when someone first banged 2 rocks together in a cave to accompany someone grunting. Or maybe it emerged as a byproduct of communication. Genre-creating has continued to this day. For example, according to Wordnik, Rockabilly is "A form of popular music combining features of rock music and bluegrass."

I believe that it is possible to write or train a computer algorthm that can take audio samples from 2 different genres (such as rock and bluegrass) and blend them together to produce a "new" genre similar to what we know as Rockabilly.

My questions are:

  1. Has this been accomplished yet?
  2. If yes or no, what major steps would be required to write such an algorthm?
  3. If these questions are too large in scope to be easily answered, then is it possible to define what a genre is in computer code? I'm thinking the answer is yes if such things as tempo, rhythm, swing, musical instruments, etc. are identified.

Thanks for any insight into this.

1
  • Comments are not for extended discussion; this conversation has been moved to chat.
    – Richard
    Commented Aug 7, 2022 at 14:04

1 Answer 1

2

There are several programs already that generate music- for example, Dadabots (known for "relentless doppelganger" on Youtube), OpenAI Jukebox, and Google's Magenta. Some skip the whole "music" part and go straight for generating audio, but some (like AIVA) do generate music based on musical scores.

As far as I can tell, there is much more focus on generating convincing songs right now, rather than having AI make any new styles of music. At least, the "new" elements of AI music are mostly accidental.

If you wanted to create new genres instead of just new songs, I suppose you could feed such an algorithm only content from 2 genres, but it's likely that its output will be in one genre or the other, rather than right in between. You'd probably have to modify any open-source program you used as a starting point (working-as-intended adversarial models will reject genre-benders if they don't exist in the original data). If you had an AI that generated musical scores rather than audio, you could certainly design it to pull e.g. "rhythm" data from one genre, and "harmonic" data from another. You probably want the AI to learn what each genre means on its own, rather than trying to define it yourself. I do believe that your goal could be done. AI is a huge topic that I barely understand, but maybe with a few years of research, you may be able to answer your second question.

7
  • Thanks Edward, if my original post wasn't clear I'm not asking about software code that can produce sheet music or chord charts in a "new" genre but rather audio. So it would be given an MP3 audio sample from genre A and an audio sample from genre B and mix or blend the two to produce audio in genre C. But step 1 is to be able to define (in computer terms), what a "genre" actually is. I'm not sure that musicologists and computer scientists have accomplished even that yet. Maybe someone here knows that answer.
    – Steve
    Commented Aug 7, 2022 at 1:13
  • 1
    Generally, these AIs require large data sets. I don't think it's even possible to quantify a genre from a single song, much less create a program that can do it. A more realistic goal might be an AI that will generate a new song from 1000 examples in genre A, and 1000 examples in genre B. Maybe an ideal program would only need 20-30. OpenAI Jukebox used over 1 million examples in their dataset.
    – Edward
    Commented Aug 7, 2022 at 1:19
  • I'm not limiting the size of the training sets in any way. Let's say it is trained on 1000 examples of rock and 1000 of blue grass; or 10 million of each. When asked to blend 1 example of rock with 1 example of bluegrass the question is would it arrive at rockabilly? If this has been attempted and the answer is no, then I think we can say more AI research is required. I just don't know if it has been attempted.
    – Steve
    Commented Aug 7, 2022 at 1:57
  • To whoever downgraded because he doesn't see how this would be useful. Here are 2 answers 1) basic music and human psychology research with possible benefits in music therapy and 2) new genres can take years or decades or more to develop often by accident and random chance. If such a tool existed new genres could be developed and explored much faster, perhaps in days.
    – Steve
    Commented Aug 7, 2022 at 2:05
  • 2
    @Steve I think he means that he doesn't like the or the answer question but doesn't understand enough that he could express in words why he doesn't like it. To me the question sounds more like profound lack of research, but this answer is really good.
    – ojs
    Commented Aug 7, 2022 at 7:26

Not the answer you're looking for? Browse other questions tagged or ask your own question.