unable to create Libri1mix

[strwbrry0]

Hello!
I can't generate the metadata for Libri1mix, the pairs keep generating and removing duplicates and it never stops. also the csv files don't get created for some reason. I let it run for 2 days and when i came back i found out that nothing was generated it just loops indefinitely.

[ccb01]

Hello! I can't generate the metadata for Libri1mix, the pairs keep generating and removing duplicates and it never stops. also the csv files don't get created for some reason. I let it run for 2 days and when i came back i found out that nothing was generated it just loops indefinitely.
Hello, I also had such a problem. Have you solved it？

[giuuuug]

Hi!
The issue is caused by an infinite loop for dev and test sets.
The code requests 3000 mixtures, but LibriSpeech dev/test only contain ~2300 utterances, so the condition can never be satisfied.

I fixed it by limiting the target number of mixtures to the maximum possible and adding a safety guard to prevent infinite loops:

Old code block:

if 'train' in librispeech_md_file.iloc[0]['subset']:
   utt_pairs = set_utt_pairs(librispeech_md_file, utt_pairs, n_src)
   noise_pairs = set_noise_pairs(utt_pairs, noise_pairs, librispeech_md_file, wham_md_file)
   # Otherwise we want 3000 mixtures
 else:
   while len(utt_pairs) < 3000: 

New code block:

if 'train' in librispeech_md_file.iloc[0]['subset']:
   utt_pairs = set_utt_pairs(librispeech_md_file, utt_pairs, n_src)
   noise_pairs = set_noise_pairs(utt_pairs, noise_pairs, librispeech_md_file, wham_md_file)
   # Otherwise we want 3000 mixtures (or max possible if dataset is smaller)
else:
   # Calculate maximum possible unique pairs
   max_possible = len(librispeech_md_file) // n_src
   target = min(3000, max_possible)
   print(f"  Target mixtures: {target} (requested: 3000, max possible: {max_possible})")
      
   max_iterations = 100  # Safety limit to prevent infinite loops
   iteration = 0
   prev_len = -1
      
   while len(utt_pairs) < target and iteration < max_iterations:

[ccb01]

Very thankful !