How to run? - Githubissues

joerenner / sampleChop

Python code for neural network that chops audio samples for hip hop production

2 stars 1 forks source link

How to run? #1

Open jeremydeanw opened 3 years ago

jeremydeanw commented 3 years ago

Hi Joe, what are the steps for running the script? (for someone who has only used python a couple of times)

jeremydeanw commented 3 years ago

I downgraded numpy to 1.11.3 (seems only the earlier versions work) But now its a different error that im trying to figure out:
File "C:\Users\Jeremy\Desktop\sampleChop-master\sampleChop\get_data.py", line 19, in cqt = np.abs(lb.core.cqt(y, sr=sr, fmin=lb.note_to_hz('F2'), AttributeError: 'module' object has no attribute 'note_to_hz'

joerenner commented 3 years ago

hmm it seems to not be finding this librosa function: https://librosa.org/doc/latest/generated/librosa.note_to_hz.html

I would make sure you have the correct librosa version (0.5.1) installed.

If you just want to run a song through the system, I would use this function: https://github.com/joerenner/sampleChop/blob/0eac5bf54b55168def11bcb894b43354573402ce/sampleChop/classify.py#L165 just replace the path to the wav or mp3 file.

This library is quite old and I didn't really know what I was doing when I created it... but hopefully it will get some interesting outputs for you!

jeremydeanw commented 3 years ago

The order to run the program is how you have it written already right? I should run get_data, data_ops, sampleChop, then classify? Or do I just run classify?

jeremydeanw commented 3 years ago

Ok.....so after alot of scratching my head, I reinstalled everything....and the issue was that I was running 0.3.0 not 0.51. I ran into a bunch of errors, and what fixed it and allowed me to install 0.5.1 was installing llvmlite-0.31.0-cp27-cp27m-win_amd64.whl I also needed to fall back joblib per: https://github.com/librosa/librosa/issues/729

Seems to work now! Running classify.py. Not sure if its working correctly though...the cuts seem to be random. Not sure what order I'm supposed to go in? How do I train on multiple drum breaks for instance? I have a blank text file in my input folder.....what does the text file do? Do you have your data points somewhere?

When running get_data.py on its own I get the error:

File "get_data.py", line 67, in truth_times = [float(x) for x in truth_times] ValueError: could not convert string to float:

jeremydeanw commented 3 years ago

Would I need to resolve the error above in order to create data points?

joerenner commented 3 years ago

Hi Jeremy,

a couple points:

the network checked in was not trained on drum breaks, so this is probably why the cuts seem random.
if you want to train on your own ground truth cuts, you'll need to write the cut times (in seconds) in the associated wav file to a text file with a comma between each time, then run the get_data script (for each wav/txt file)
I'm not sure where my data is.

Sorry I can't be more help, this was a side project years ago when I was in school and just messing around getting to know machine learning.

jeremydeanw commented 3 years ago

Thanks, I'm more just needing to know what the workflow order is....so a temp-plot.html, as well as features + targets txt files were created after running get_data. I see that in sampleChop.py that it uses these files. I tried running sampleChop but ran into an error, I have the seconds as 05,10,15.....is this the correct format? I'm sure I'll figure it out eventually

ValueError: Found input variables with inconsistent numbers of samples: [251802, 4663]

(when running sampleChop)