no more filtering noisy bins or vocs

get_noisy_labels.py

-from glob import glob
-from p_tqdm import p_umap
-from scipy import signal, interpolate
-import pandas as pd, numpy as np
-from metadata import species
-import librosa, mir_eval
-import os, shutil, argparse
-
-parser = argparse.ArgumentParser()
-parser.add_argument('specie', type=str, help="Species to treat specifically", default=None)
-args = parser.parse_args()
-
-for specie in species if args.specie is None else [args.specie]:    wavpath, FS, nfft, downsample = species[specie].values()
-    os.system(f'rm -R \"noisy_pngs/{wavpath.split("*")[0]}/*\"')
-
-    dt = nfft / 8 / FS # winsize / 8
-    def fun(fn):
-        df = pd.read_csv(f'{fn[:-4]}_preds.csv')
-        if 'salience' in df.columns and (df.salience.quantile(.25) < 0.2 or df.SHR.quantile(.75) > 0):
-            # print(fn, snr, subharSNR)
-            if not os.path.isdir(f'noisy_pngs/{fn.rsplit("/",1)[0]}'):
-                os.mkdir(f'noisy_pngs/{fn.rsplit("/",1)[0]}')
-            shutil.copyfile(f'annot_pngs/{fn[:-4]}.png', f'noisy_pngs/{fn[:-4]}.png')
-            return 1
-        else:
-            return 0
-    count = p_umap(fun, glob(wavpath), desc=specie)
-    print(f'dropped {sum(count)} vocs out of {len(glob(wavpath))}')