Initial ComParE commit, Primate Dataset

datasets/metadata_readout.py

@@ -0,0 +1,34 @@
+import numpy as np
+import torch
+import pytorch_lightning as pl
+import librosa
+import pandas as pd
+import variables as v
+from tqdm import tqdm
+
+sr = 16000
+wavs = list((v.PRIMATES_Root/'wav').glob('*.wav'))
+
+if __name__ == '__main__':
+
+    durations = []
+
+    for wav in tqdm(wavs):
+        duration = librosa.get_duration(filename=str(wav), sr=sr)
+        durations.append(duration)
+
+    mean_duration, std_duration, min_duration, max_duration = np.mean(durations), np.std(durations), np.min(durations), np.max(durations)
+
+    print(f'Mean duration: {mean_duration:.3f}s\tstd: {std_duration:.3f}s\tmin: {min_duration:.3f}s\t max: {max_duration:.3f}s')
+
+    primates_train_csv, primates_devel_csv, primates_test_csv = \
+        [pd.read_csv(p) for p in [v.PRIMATES_Root / 'lab' / 'train.csv',
+                                  v.PRIMATES_Root / 'lab' / 'devel.csv',
+                                  v.PRIMATES_Root / 'lab' / 'test.csv']]
+
+    csv = pd.concat((primates_train_csv,
+                     primates_devel_csv,
+                     primates_test_csv))
+
+    print(csv.groupby('label').count())
+    print([np.quantile(durations, q=q) for q in [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9]])