AI music

  • Guest, it's time once again for the hotly contested and exciting FoH Asshat Tournament!



    Go here and fill out your bracket!
    Who's been the biggest Asshat in the last year? Once again, only you can decide!

Tuco

I got Tuco'd!
<Gold Donor>
45,203
73,226
pretty decent convergence on this. I wonder how confusing the bluegrass sample sounds to someone with a better ear than I have. Someone who could play that tune but might be confused about how certain sounds are attained that are unrealistic. Maybe it's all vanilla enough that you just need a whammy bar or some vibrato to accomplish, or if hearing it is like would be like seeing an AI image of a person with two right thumbs.
 

Captain Suave

Caesar si viveret, ad remum dareris.
4,587
7,703
pretty decent convergence on this. I wonder how confusing the bluegrass sample sounds to someone with a better ear than I have. Someone who could play that tune but might be confused about how certain sounds are attained that are unrealistic. Maybe it's all vanilla enough that you just need a whammy bar or some vibrato to accomplish, or if hearing it is like would be like seeing an AI image of a person with two right thumbs.

There's some definite wonkiness to some of the sounds. The "Death Metal" is pretty much just pitched noise with a drum backing. My wife (runs an AI/data science department and is a serious violinist) has played with it some and says it's awful for classical music.

No doubt it will improve, though.
 
  • 1Like
Reactions: 1 user

Kharzette

Watcher of Overs
4,758
3,390
Having more luck with this now that I have an 8 gig gpu. I was roadblocked at the dimensionality problem: maximum size for tensor is 150 but size is 500 etc. This turned out to be the specified duration. Somewhere in the py or configs it seems hard coded for 10 second clips, and that's the very helpful error you get if the sizes don't match.

First pass took 10 minutes, second is looking like it will be around 40. I'm 27 minutes into it. I think there's a third pass, but not sure.
 

Kharzette

Watcher of Overs
4,758
3,390
Dammit
Code:
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 1.51 GiB (GPU 0; 10.91 GiB total capacity; 6.86 GiB already allocated; 1.52 GiB free; 8.31 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation.  See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Got bored waiting. Was an hour into stage 2, said I had 8 minutes to go. I'll fire it up before I sleep next time.
 

Kharzette

Watcher of Overs
4,758
3,390
Ok it worked. It looks like it ran 3.5 hours and generated 4 at a time from a set of prompts in the code:

Python:
prompts = [
    [
        'The main soundtrack of an arcade game. It is fast-paced and upbeat, with a catchy electric guitar riff. The music is repetitive and easy to remember, but with unexpected sounds, like cymbal crashes or drum rolls.',
        'A fusion of reggaeton and electronic dance music, with a spacey, otherworldly sound. Induces the experience of being lost in space, and the music would be designed to evoke a sense of wonder and awe, while being danceable.',
        'A rising synth is playing an arpeggio with a lot of reverb. It is backed by pads, sub bass line and soft drums. This song is full of synth sounds creating a soothing and adventurous atmosphere. It may be playing at a festival during two songs for a buildup.',
        'Slow tempo, bass-and-drums-led reggae song. Sustained electric guitar. High-pitched bongos with ringing tones. Vocals are relaxed with a laid-back feel, very expressive.',
    ],
    ['song with synths and flute', 'crowd cheering', 'piano sonata waltz, glittery', 'house song, 4 on the floor, rhythm'],
    ['chirping of birds and the distant echos of bells', 'cat meowing', 'saxophone with drums', 'beethoven piano sonata']
]

The results are not great :emoji_laughing:

It is a place to start though.

I'll find a place to share the wavs. They are small, but forum doesn't allow wav.
 

Kharzette

Watcher of Overs
4,758
3,390
1.5 hours setting up a conda environment. The very first step in the demo jupiter notebook fails.

Imagine grabbing a python repo called turd. And the very first step you run in the notebook is from turd import blah and get a ModuleNotFound turd...

That's python for you. It can't find the most base level of the thing you are sitting in.