diff --git a/.gitignore b/.gitignore
index 92c288c8878ed50335f579b6219e592643957270..741c5687754bc9341f87f4ffd81cb8d14bba7ccc 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,21 +1,8 @@
-LiuLabData/*
__pycache__/*
-train_set*
-hyenas/*
-dclmmpa2011/*
-wolves/*
-orangs/*
-Synth_birds_database/*
pred_pngs/*
annot_pngs/*
noisy_pngs/*
runs/*
-little_auk/*
-Lion Roar Data/*
-monk parakeet/*
-FCPalmae/*
-little_owl/*
-white_eye/*
+data/*
pesto-full/*
-marcelo/*
-acoustic_data_sets.zip
+crepe_ft/train_set*.pkl
diff --git a/metadata.py b/metadata.py
index 3dcfeb2261451d28d83d3aa4722e9ab3e508f315..30291d0a5fcb335914ad7b5d53631defb75a20d0 100644
--- a/metadata.py
+++ b/metadata.py
@@ -1,86 +1,93 @@
species = {
'wolves':{
- 'wavpath': 'wolves/*/*.wav',
+ 'wavpath': 'data/wolves/*/*.wav',
'FS': 16000,
'nfft': 1024,
'downsample':1,
'step': 1/8
},
- # 'SynthBirds':{
- # 'wavpath': 'Synth_birds_database/1_original_files/*/*.wav',
- # 'FS': 44100,
- # 'nfft': 1024,
- # 'downsample':1,
- # 'step': 1/16
- # },
'dclde':{
- 'wavpath': 'dclmmpa2011/cut_no_overlap/*.wav',
+ 'wavpath': 'data/dclmmpa2011/cut_no_overlap/*.wav',
'FS': 192000,
'nfft':1536, # according to silbido paper, 8ms windows and 125Hz resolution
'downsample':20,
'step': 1/8
},
'hyenas':{
- 'wavpath': 'hyenas/lehmann hyena whoop traces/*.wav',
+ 'wavpath': 'data/hyenas/lehmann hyena whoop traces/*.wav',
'FS': 8000,
'nfft': 2048,
'downsample':1,
'step': 1/8
},
'orangs':{
- 'wavpath': 'orangs/*.wav',
+ 'wavpath': 'data/orangs/*.wav',
'FS': 44100,
'nfft':2048,
'downsample':1,
'step': 1/8
},
'mice':{
- 'wavpath': 'LiuLabData/cut_no_overlap/*.wav',
+ 'wavpath': 'data/LiuLabData/cut_no_overlap/*.wav',
'FS': 250000,
'nfft':512,
'downsample':50,
'step': 1/8
},
'lions':{
- 'wavpath': 'Lion Roar Data/Biologger Roars Expanded (500ms)/*.WAV',
+ 'wavpath': 'data/Lion Roar Data/Biologger Roars Expanded (500ms)/*.WAV',
'FS':16000,
'nfft':2048, # was zero-padded to 4 times 2048 in the original
'downsample':0.5,
'step': 1/8
},
'parakeets':{
- 'wavpath': 'monk parakeet/pre-processed_calls/*.WAV',
+ 'wavpath': 'data/monk parakeet/pre-processed_calls/*.WAV',
'FS':44100,
'nfft':512,
'downsample':3,
'step': 1/16
},
'palmae':{
- 'wavpath': 'FCPalmae/cut/*.wav',
+ 'wavpath': 'data/FCPalmae/cut/*.wav',
'FS':44100,
'nfft':1024,
'downsample':5,
'step': 1/16
},
'little_owl':{
- 'wavpath': 'little_owl/cut/*.wav',
+ 'wavpath': 'data/little_owl/cut/*.wav',
'FS':4000,
'nfft':512,
'downsample':1,
'step': 1/8
},
'white_eye':{
- 'wavpath': 'white_eye/cut/*.wav',
+ 'wavpath': 'data/white_eye/cut/*.wav',
'FS':44100,
'nfft':1024,
'downsample':5,
'step': 1/16
},
'long_billed':{
- 'wavpath':'marcelo/long_billed_hermit_songs/*.wav',
+ 'wavpath':'data/marcelo/long_billed_hermit_songs/*.wav',
'FS':44100,
'nfft':512,
'downsample':5,
'step': 1/16
+ },
+ 'hummingbird':{
+ 'wavpath':'data/marcelo/hummingbird_songs/*.wav',
+ 'FS':44100,
+ 'nfft':512,
+ 'downsample':5,
+ 'step':1/16
+ },
+ 'bats':{
+ 'wavpath':'data/marcelo/spixs_disc_winged_bat_*/*.wav',
+ 'FS':400000,
+ 'nfft':512,
+ 'downsample':20,
+ 'step':1/16
}
}
diff --git a/print_annot.py b/print_annot.py
index 45ac930c132e139e05ec50d792d5cb1c939eaa30..97aed3e86e8dbc3625968a65c800bdfeaf7f476f 100644
--- a/print_annot.py
+++ b/print_annot.py
@@ -14,13 +14,12 @@ args = parser.parse_args()
for specie in species if args.specie is None else [args.specie]:
wavpath, FS, nfft, downsample, step = species[specie].values()
dt = nfft * step / FS # winsize / 8
- Hz2bin = lambda f: np.round(f/FS*nfft).astype(int)
# for fn in glob(wavpath):
def fun(fn):
# if os.path.isfile(f'annot_pngs/{fn[:-4]}.png'):
# return
- if not os.path.isdir(f'annot_pngs/{fn.rsplit("/",1)[0]}'):
- os.mkdir(f'annot_pngs/{fn.rsplit("/",1)[0]}')
+ if not os.path.isdir(f'annot_pngs/{fn[5:].rsplit("/",1)[0]}'):
+ os.makedirs(f'annot_pngs/{fn[5:].rsplit("/",1)[0]}', exist_ok=True)
# load signal and compute spetrogram
sig, fs = librosa.load(fn, sr=FS)
df = pd.read_csv(f'{fn[:-4]}_preds.csv').dropna(subset='annot')
@@ -30,12 +29,13 @@ for specie in species if args.specie is None else [args.specie]:
if 'salience' in df.columns:
plt.scatter(df.time, df.annot, c=df.salience, s=1 if specie in {'dclde','mice'} else None, cmap='jet', vmin=0, vmax=1)
plt.colorbar()
- plt.title(f'Salience {df.salience.quantile(.25):.2f}, SHR {df.SHR.quantile(.25):.2f}')
+ plt.title(f'Salience {df.salience.quantile(.25):.2f}, SHR {df.SHR.quantile(.25):.2f}, Harmonicity {df.harmonicity.quantile(.25):.2f}')
else:
plt.scatter(df.time, df.annot, alpha=.2, s=1 if specie in {'dclde','mice'} else None)
plt.ylim(0, df.annot.max()*1.5)
plt.tight_layout()
- plt.savefig(f'annot_pngs/{fn[:-4]}.png')
+ plt.savefig(f'annot_pngs/{fn[5:-4]}.png')
plt.close()
- p_umap(fun, glob(wavpath), desc=specie)
+ files = pd.Series(glob(wavpath))
+ p_umap(fun, files, desc=specie)
diff --git a/print_preds.py b/print_preds.py
index 312e766554f498a10394f7d1ff7b32a1f0131e4b..5f724210429d3087241f79713d9f38da8d088e2a 100644
--- a/print_preds.py
+++ b/print_preds.py
@@ -7,15 +7,15 @@ import os, argparse
from metadata import species
np.seterr(divide = 'ignore')
-algos = ['pyin', 'praat', 'crepe', 'tcrepe', 'tcrepe_ft', 'basic', 'pesto']
+algos = ['praat', 'tcrepe_ft']
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()
+ wavpath, FS, nfft, downsample, step = species[specie].values()
thrs = pd.read_csv(f'scores/{specie}_scores.csv', index_col=0).threshold
- dt = nfft / 8 / FS # winsize / 8
+ dt = nfft * step / FS # winsize / 8
# for fn in glob(wavpath):
def fun(fn):
# if os.path.isfile(f'annot_pngs/{fn[:-4]}.png'):
@@ -25,7 +25,7 @@ for specie in species if args.specie is None else [args.specie]:
# load signal and compute spetrogram
sig, fs = librosa.load(fn, sr=FS)
df = pd.read_csv(f'{fn[:-4]}_preds.csv')
- df.time /= downsample
+
S, freqs, times, ax = plt.specgram(sig, Fs=FS, NFFT=nfft, noverlap=int(nfft-dt*fs))
S = 10*np.log10(S+1e-10)
plt.scatter(df.dropna(subset='annot').time, df.dropna(subset='annot').annot, c='k', alpha=.2, label='annot')