diff --git a/experiments/ldcnn20k60.yaml b/experiments/ldcnn20k60.yaml
index 149571b..c4b2624 100644
--- a/experiments/ldcnn20k60.yaml
+++ b/experiments/ldcnn20k60.yaml
@@ -4,8 +4,8 @@ fmin: 0
 fmax: 11025
 n_fft: 1024
 hop_length: 512
-augmentations: 12
-augment: 1
+augmentations: 0
+augment: 0
 frames: 31
 batch: 400
 epochs: 100
diff --git a/microesc/features.py b/microesc/features.py
index 89d3f36..a5e6c69 100644
--- a/microesc/features.py
+++ b/microesc/features.py
@@ -6,6 +6,7 @@
 import urllib.request
 import zipfile
 import collections
+import time
 
 import pandas
 import numpy
@@ -108,6 +109,8 @@ def download_progress(count, blocksize, totalsize):
 
 def load_sample(sample, settings, feature_dir, window_frames,
                 start_time=None, augment=None, normalize='meanstd'):
+    start_t = time.time()    
+
     n_mels = settings['n_mels']
     sample_rate = settings['samplerate']
     hop_length = settings['hop_length']
@@ -121,7 +124,12 @@ def load_sample(sample, settings, feature_dir, window_frames,
     # Load precomputed features
     folder = os.path.join(feature_dir, settings_id(settings))
     path = feature_path(sample, out_folder=folder, augmentation=aug)
+
+
+    before_load = time.time()
     mels = numpy.load(path)['arr_0']
+    after_load = time.time()
+
     assert mels.shape[0] == n_mels, mels.shape
 
     if start_time is None:
@@ -161,6 +169,11 @@ def load_sample(sample, settings, feature_dir, window_frames,
 
     # add channel dimension
     data = numpy.expand_dims(padded, -1)
+
+    end_t = time.time()
+
+    #print(f'load feature {(end_t-start_t)*1000}, {(after_load-before_load)*1000}')
+
     return data
 
 
diff --git a/microesc/train.py b/microesc/train.py
index 6c74801..354f294 100644
--- a/microesc/train.py
+++ b/microesc/train.py
@@ -33,7 +33,10 @@ def dataframe_generator(X, Y, loader, batchsize=10, n_classes=10):
     while True:
         idx = numpy.random.choice(len(X), size=batchsize, replace=False)
         rows = X.iloc[idx, :].iterrows()
-        data = [ loader(d) for _, d in rows ]
+
+        #data = [ loader(d) for _, d in rows ]
+        data = numpy.zeros(shape=(400, 60, 31, 1))
+
         y = Y.iloc[idx]
         y = keras.utils.to_categorical(y, num_classes=n_classes)
         batch = (numpy.array(data), numpy.array(y))