small stuff

compute_salience_SHR.py

@@ -35,7 +35,7 @@ for specie in species if args.specie == 'all' else [args.specie]:
             if FS*r.time < nfft//2 or FS*r.time > len(sig) - nfft//2 or (sig[int(FS*r.time) - nfft//2 : int(FS*r.time) + nfft//2] == 0).all():
                 continue
             spec = get_spectrum(sig[int(FS*r.time) - nfft//2 : int(FS*r.time) + nfft//2])
-            spec = np.clip(spec-noise, 0, 10)
+            spec = np.clip(spec-noise, 1e-5, 10)
             f0 = r.annot
 
             df.loc[i, 'salience'] = sum(spec[Hz2bin(f0*2**(-1/12)):Hz2bin(f0*2**(1/12))+1]) / sum(spec[Hz2bin(f0*2**(-6/12)):Hz2bin(f0*2**(6/12))+1])

eval_all.py

@@ -31,7 +31,7 @@ for specie in species if args.specie=='all' else args.specie.split(' '):
         confs['label'].extend(df.annot>0)
     thrs = {}
     for algo in list(algos):
-        if np.isnan(confs[algo]).all():
+        if np.isnan(confs[algo]).all() or len(confs[algo]) < len(confs['label']):
             algos -= {algo}
             continue
         fpr, tpr, thr = skmetrics.roc_curve(np.array(confs['label'])[~np.isnan(confs[algo])], np.array(confs[algo])[~np.isnan(confs[algo])])
@@ -49,7 +49,7 @@ for specie in species if args.specie=='all' else args.specie.split(' '):
                 return pd.DataFrame()
         out = pd.DataFrame(columns=metrics)
         for algo in algos:
-            if df[algo+'_f0'].isna().all():
+            if not algo+'_f0' in df.columns or df[algo+'_f0'].isna().all():
                 continue
             out.loc[algo, ['Recall', 'False alarm']] = mir_eval.melody.voicing_measures(df.annot > 0, df[algo+'_conf'] > thrs[algo])
             df[algo+'_f0'] = mir_eval.melody.hz2cents(df[algo+'_f0'])
@@ -63,12 +63,14 @@ for specie in species if args.specie=='all' else args.specie.split(' '):
 
     fig, ax = plt.subplots(ncols=3, figsize=(12, 4), sharex=True)
     for i, algo in enumerate(algos):
+        if algo in df.index:
             ax[0].violinplot(np.concatenate(df.loc[algo, 'diff_distrib']), positions=[i])
             ax[1].violinplot(df.loc[algo, 'Pitch acc'], positions=[i])
             ax[2].violinplot(df.loc[algo, 'Chroma acc'], positions=[i])
  
     ax[0].set_yscale('log')
     ax[0].set_title('Distrib of errors in cents')
+    ax[0].hlines(1200, 0, len(algos), linestyle='dashed', color='k')
     ax[1].set_title('Distrib of pitch acc per vocs in % ')
     ax[2].set_title('Distrib of chroma acc per vocs in % ')
     plt.xticks(np.arange(len(algos)), algos, rotation=45)

plot_freq_distrib.py

@@ -7,9 +7,9 @@ from metadata import species
 np.seterr(divide = 'ignore')
 
 fig, ax = plt.subplots(ncols=3, sharey=True, figsize=(10, 3.5))
-ax[0].set_title('Frequency (Hz)')
-ax[1].set_title('# Voiced time bins')
-ax[2].set_title('Modulation rate (Hz/sec)')
+ax[0].set_xlabel('Frequency (Hz)')
+ax[1].set_xlabel('# Voiced time bins')
+ax[2].set_xlabel('Modulation rate (Hz/sec)')
 for i in range(3):
     ax[i].set_xscale('log')
     ax[i].grid()
@@ -25,10 +25,15 @@ for i, specie in enumerate(species):
         fdistrib.extend(f0s[mask2.astype(bool)])
         tdistrib.append(mask2.sum())
         moddistrib.extend(abs(np.diff(f0s[mask2.astype(bool)]))/dt)
-    ax[0].violinplot(fdistrib, positions=[i], vert=False)
-    ax[1].violinplot(tdistrib, positions=[i], vert=False)
-    ax[2].violinplot(moddistrib, positions=[i], vert=False)
+    p = ax[0].violinplot(fdistrib, points=1000, positions=[i], vert=False)
+    p['bodies'][-1].set_alpha(1)
+    p = ax[1].violinplot(tdistrib, points=1000, positions=[i], vert=False)
+    p['bodies'][-1].set_alpha(1)
+    p = ax[2].violinplot(moddistrib, points=1000, positions=[i], vert=False)
+    p['bodies'][-1].set_alpha(1)
 
+
+ax[0].set_xticks(10**np.arange(1, 6))
 plt.yticks(np.arange(len(species)), [s.replace('_',' ') for s in species])
 plt.tight_layout()
 plt.savefig('freq_distrib.pdf')

plot_scores_bars.py

@@ -3,7 +3,7 @@ import pandas as pd, numpy as np
 from metadata import species
 import argparse
 
-algos = ['praat', 'tcrepe_ftsp', 'tcrepe_ftoth', 'basic', 'pesto', 'pesto_ft']
+algos = ['praat', 'basic', 'pesto', 'pesto_ft', 'tcrepe_ftoth', 'tcrepe_ftsp']
 metrics = ['Pitch acc', 'Chroma acc', 'Recall', 'Specificity']
 
 parser = argparse.ArgumentParser()

run_all.py

@@ -33,6 +33,7 @@ parser.add_argument('--quick', type=bool, help="Skip pyin and crepe to make thin
 args = parser.parse_args()
 
 algos = ['praat_f0','pyin_f0','crepe_f0','tcrepe_f0','tcrepe_ft_f0','tcrepe_ftsp_f0','tcrepe_ftoth_f0', 'basic_f0','pesto_f0', 'pesto_ft_f0']
+quick_algos = ['praat_f0','tcrepe_f0','tcrepe_ft_f0','tcrepe_ftsp_f0','tcrepe_ftoth_f0', 'basic_f0','pesto_f0', 'pesto_ft_f0']
 
 if args.overwrite:
     print('Overwriting previous results')
@@ -57,12 +58,9 @@ for specie in species if args.specie =='all' else args.specie.split(' '):
             mask = ((out.time > annot.Time.min())&(out.time < annot.Time.max()))
             out.loc[mask, 'annot'] = mir_eval.melody.resample_melody_series(annot.Time, annot.Freq, annot.Freq>0, out.loc[mask, 'time'], verbose=False)[0]
         else:
-            out = pd.read_csv(f'{fn[:-4]}_preds.csv')
+            out = pd.read_csv(f'{fn[:-4]}_preds.csv').dropna(axis=1, how='all') # drop a column if all values are None
             # check if everything has already been computed, and if yes skip the file
-            for algo in algos:
-                if algo in out.columns and out[algo].isna().all():
-                    out.drop(algo, axis=1, inplace=True)
-            if pd.Series(algos).isin(out.columns).all():
+            if pd.Series(algos).isin(out.columns).all() or (args.quick and pd.Series(quick_algos).isin(out.columns).all()):
                 continue
             sig, fs = librosa.load(fn, sr=FS)