Any artifical intelligence/machine learning folks in here?

[sawzalot]

I'm a software engineer and work for a company that is very much into ML/AI.

However.

AI and machine learning is the antithesis of the creative process and artistic expression that is music. There is something unknowable in the process of creating music, which machines will never master. I'm not going to waste even one second of valuable time at artificial intelligence/machine learning related to music, when I could be spending that time at real intelligence/human learning/human creating music.

 

[dalahorse]

The tuner was mentioned before and I agree with that. You could shove the data on to GPUs, Python, and TensorFlow all day long, work out variables and tunings, and repeat thousands of iterations to solidify your program. Or you could just watch the tuner. Even easier than that... If there is a MIDI version available, play it and the recording side by side and judge for yourself how out of tune the recording sounds.

In music, intonation can't always be expressed as a boolean value. There are levels of expression - pleasing or displeasing to the human ear - that software probably isn't going to understand. IMHO, today's AI and musical expression aren't yet compatible. Tomorrow? Perhaps.

I have a DGX-1 and a few data scientists at my day job. I should ask them about this tomorrow.

 

[Wood and Wire]

You could do this with the VocAlign plug-in.

Input the original recording as a reference (isolated track of Jaco bass solo, for example).

Record yourself playing the same solo.

Load both files into VocAlign, and the timing & pitch of the original can be applied to your re-recording.





* (Obviously, it's intended for use with vocals / dialogue).

 

[JW56789]

Personally (although I'm NOT a fretless player) for me, that slight occasional just-barely-out-of-tune note here and there is part of the way a fretless sounds to me, like 12-string guitars that are never totally, completely in tune.

Don't know if some AI R&D group would take this on, but in the meantime, if it kept me up nights, I'd try something like this . . . . .

. . . . or something similar. If these things can fix a warbling front man, oughta work on fretless, much LESS to correct than that clown.

 

[wellshuxley]

Lol I hear this every time I present a problem I think is good for AI. I generate terabytes of microscopy data in my job and the AI ML folks always want to analyze it. They then ask me what I want to find out..

I say x

They say...that's not an AI ML problem

A big dataset does not necessarily mean you have a problem suited to AI/ML. The only part of the problem you described that is potentially suited to AI/ML seems to be extracting the bass part from an ensemble recording.

If you wanted to, say, teach a computer to “play” bass like a particular human bass player, mimicking playing characteristics such as situational intonation- that’s a ML problem.

 

[MYLOWFREQ]

If you pull the isolated bass track into an auto-tune software, you'd see exactly how many cents the notes are off. I think this is the most accurate way of analysis.

 

[40Hz]

A big dataset does not necessarily mean you have a problem suited to AI/ML.

Exactly.

And based on the OP’s first post I don’t see where his idea would need to employ AI or machine learning technology at all.

If you pull the isolated bass track into an auto-tune software, you'd see exactly how many cents the notes are off. I think this is the most accurate way of analysis.

There you go!

Assuming they didn’t use autotune on the track - or rerecord it using an uncredited substitute player.

 

[MYLOWFREQ]

Assuming they didn’t use autotune on the track - or rerecord it using an uncredited substitute player.

Generally speaking, autotuned tracks become obvious when you pull the files into a software and analize them. They are just too perfect to be real.