-
Notifications
You must be signed in to change notification settings - Fork 79
Open
Description
It seems to me even when I only need "mix_clean", the data preparation code is still highly entangled with noise
LibriMix/scripts/create_librimix_from_metadata.py
Lines 158 to 169 in 60199d7
| mix_id, gain_list, sources = read_sources(row, n_src, librispeech_dir, | |
| wham_dir) | |
| # Transform sources | |
| transformed_sources = transform_sources(sources, freq, mode, gain_list) | |
| # Write the sources and get their paths | |
| abs_source_path_list = write_sources(mix_id, | |
| transformed_sources, | |
| subdirs, dir_path, freq, | |
| n_src) | |
| # Write the noise and get its path | |
| abs_noise_path = write_noise(mix_id, transformed_sources, dir_path, | |
| freq) |
Hence, it might make more sense to always create the "noise" directory here
LibriMix/scripts/create_librimix_from_metadata.py
Lines 104 to 109 in 60199d7
| # Create subdir | |
| if types == ['mix_clean']: | |
| subdirs = [f's{i + 1}' for i in range(n_src)] + ['mix_clean'] | |
| else: | |
| subdirs = [f's{i + 1}' for i in range(n_src)] + types + [ | |
| 'noise'] |
In my use case, I find that when I only use "mix_clean" in the generate_librimix.sh script, the script will crash since the "noise" directory is not created. Please correct me if I am wrong :)
Thanks!
Sincerely,
Leo
Metadata
Metadata
Assignees
Labels
No labels