speed aug fixed

audio_toolset/audio_augmentation.py

@@ -11,18 +11,24 @@ class Speed(object):
 
     def __init__(self, max_amount=0.3, speed_min=1, speed_max=1):
         self.speed_max = speed_max
-        self.speed_min = speed_min
-        self.max_amount = max_amount
+        self.speed_min = speed_min if speed_min else 1
+        self.max_amount = speed_max if speed_max else 1
 
     def __call__(self, x):
-        assert all([self.speed_min, self.speed_max, self.max_amount])
+        if self.speed_min == 1 and self.speed_max == 1:
+            return x
         start = int(np.random.randint(low=0, high=x.shape[-1], size=1))
         width = np.random.uniform(low=0, high=self.max_amount, size=1) * x.shape[-1]
         end = int(width + start)
         end = min(end, x.shape[-1])
         try:
             speed_factor = float(np.random.uniform(low=self.speed_min, high=self.speed_max, size=1))
-            aug_data = librosa.effects.time_stretch(x[start:end], speed_factor)
-            return np.concatenate((x[:start], aug_data, x[end:]), axis=0)[:x.shape[-1]]
+            aug_data = librosa.effects.time_stretch(y=x[start:end], rate=speed_factor)
+            x_aug = np.concatenate((x[:start], aug_data, x[end:]), axis=0)[:x.shape[-1]]
+            if speed_factor > 1:
+                embedding = np.zeros_like(x)
+                embedding[:x_aug.shape[0]] = x_aug
+                x_aug = embedding
+            return x_aug
         except ValueError:
             return x