steffen/ml_lib
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

speed aug fixed

Browse Source
This commit is contained in:
Si11ium
2020-05-21 14:42:35 +02:00
parent 235743b225
commit 8cec323286
1 changed files with 11 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
audio_toolset/audio_augmentation.py
View File

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