RetroCirce / Zero_Shot_Audio_Source_Separation

The official code repo for "Zero-shot Audio Source Separation through Query-based Learning from Weakly-labeled Data", in AAAI 2022
https://arxiv.org/abs/2112.07891
MIT License
184 stars 31 forks source link

Total lenght of 'dataloader' across ranks is zero. Please make sure that it returns at least 1 batch. #16

Closed lucellent closed 2 years ago

lucellent commented 2 years ago

This is weird, it was working a few days ago when it suddenly spit out this error. In my understanding it's suggesting that the (reference) samples are not enough? Even though I'm using the same song and samples as prior to this error, I haven't changed anything

lucellent commented 2 years ago

(accidentally closed the issue, sorry)

RetroCirce commented 2 years ago

Hi,

You said that the code ran well a few days ago but now it comes out some errors?

I think the problem possibly comes from your data. Your error log basically says that the model cannot find enough data to separate or query. You probably need to check how much your data’s duration is, or how many samples you have.

lucellent commented 2 years ago

Imagine I separated a song into L and R channels (since zero-shot only works with mono audio) and there are 3 samples in my sample folders, with different lenghts.

I successfully used zero-shot for the R channel file, but then when I tried to run the L channel file, that's when it started giving me the error

That's what I mean by I didn't change anything

RetroCirce commented 2 years ago

Would you send me the file of your samples? If the model works well on your L channel file, I believe there must be something wrong with your R channel file.

lucellent commented 2 years ago

I already deleted them sadly, but still - I exported both the same way and with same settings

I could send you the files I tried to run today if that will work also

RetroCirce commented 2 years ago

Sure, you can send it to me and let me check

lucellent commented 2 years ago

https://we.tl/t-hmjs4k6FKO

lucellent commented 2 years ago

Ah, something came to my mind - I just exported the sample in 16bit wav, it was previously 32 bit mp3... it works now

Was it because it was a 32bit file? I don't usually work with such

RetroCirce commented 2 years ago

Aha, I think the problem might lie in here. I used the librosa audio library to process the audio. By default it might process on 16bit file (I’m not sure). The error says it cannot find any data so it possibly is because the librosa does not process the 32bit file and output none. But i think this can be revised by changing hyperparameters. You can Google it to see more details.

Glad to hear that you fix this problem!