Audio Dataset

audio_toolset/audio_augmentation.py

@@ -15,6 +15,9 @@ class Speed(object):
         # noinspection PyTypeChecker
         self.max_amount = min(max(0, max_amount), 1)
 
+    def __repr__(self):
+        return f'{self.__class__.__name__}({self.__dict__})'
+
     def __call__(self, x):
         if self.speed_min == 1 and self.speed_max == 1:
             return x

audio_toolset/audio_io.py

@@ -37,6 +37,9 @@ class MFCC(object):
     def __init__(self, **kwargs):
         self.__dict__.update(kwargs)
 
+    def __repr__(self):
+        return f'{self.__class__.__name__}({self.__dict__})'
+
     def __call__(self, y):
         mfcc = librosa.feature.mfcc(y, **self.__dict__)
         return mfcc
@@ -47,6 +50,9 @@ class NormalizeLocal(object):
         self.cache: np.ndarray
         pass
 
+    def __repr__(self):
+        return f'{self.__class__.__name__}({self.__dict__})'
+
     def __call__(self, x: np.ndarray):
         mean = x.mean()
         std = x.std() + 0.0001
@@ -65,6 +71,9 @@ class NormalizeMelband(object):
         self.cache: np.ndarray
         pass
 
+    def __repr__(self):
+        return f'{self.__class__.__name__}({self.__dict__})'
+
     def __call__(self, x: np.ndarray):
         mean = x.mean(-1).unsqueeze(-1)
         std = x.std(-1).unsqueeze(-1)
@@ -98,6 +107,9 @@ class PowerToDB(object):
     def __init__(self, running_max=False):
         self.running_max = 0 if running_max else None
 
+    def __repr__(self):
+        return f'{self.__class__.__name__}({self.__dict__})'
+
     def __call__(self, x):
         if self.running_max is not None:
             self.running_max = max(np.max(x), self.running_max)
@@ -109,6 +121,9 @@ class LowPass(object):
     def __init__(self, sr=16000):
         self.sr = sr
 
+    def __repr__(self):
+        return f'{self.__class__.__name__}({self.__dict__})'
+
     def __call__(self, x):
         return butter_lowpass_filter(x, 1000, 1)
 
@@ -117,6 +132,9 @@ class MelToImage(object):
     def __init__(self):
         pass
 
+    def __repr__(self):
+        return f'{self.__class__.__name__}({self.__dict__})'
+
     def __call__(self, x):
         # Source to Solution: https://stackoverflow.com/a/57204349
         mels = np.log(x + 1e-9)  # add small number to avoid log(0)

audio_toolset/mel_dataset.py

@@ -0,0 +1,30 @@
+from pathlib import Path
+
+import numpy as np
+from torch.utils.data import Dataset
+
+
+class TorchMelDataset(Dataset):
+    def __init__(self, identifier, mel_path, segment_len, hop_len, label, padding=0, transform=None):
+        self.padding = padding
+        self.path = next(iter(Path(mel_path).glob(f'{identifier}_*')))
+        self.segment_len = segment_len
+        self.m, self.n = str(self.path).split('_')[-2:]  # get spectrogram dimensions
+        self.n = int(self.n.split('.', 1)[0])  # remove .npy
+        self.m, self.n = (int(i) for i in (self.m, self.n))
+        self.offsets = list(range(0, self.n - segment_len, hop_len))
+        self.label = label
+        self.transform = transform
+
+    def __getitem__(self, item):
+        start = self.offsets[item]
+        mel_spec = np.load(str(self.path), allow_pickle=True)
+        if self.padding > 0:
+            mel_spec = np.pad(mel_spec, pad_width=[(0, 0), (self.padding // 2, self.padding // 2)], mode='mean')
+        snippet = mel_spec[:, start: start + self.segment_len]
+        if self.transform:
+            snippet = self.transform(snippet)
+        return snippet, self.label
+
+    def __len__(self):
+        return len(self.offsets)

audio_toolset/mel_transforms.py

@@ -8,6 +8,9 @@ class Normalize(object):
     def __init__(self, min_db_level: Union[int, float]):
         self.min_db_level = min_db_level
 
+    def __repr__(self):
+        return f'{self.__class__.__name__}({self.__dict__})'
+
     def __call__(self, s: np.ndarray) -> np.ndarray:
         return np.clip((s - self.min_db_level) / -self.min_db_level, 0, 1)
 
@@ -17,5 +20,8 @@ class DeNormalize(object):
     def __init__(self, min_db_level: Union[int, float]):
         self.min_db_level = min_db_level
 
+    def __repr__(self):
+        return f'{self.__class__.__name__}({self.__dict__})'
+
     def __call__(self, s: np.ndarray) -> np.ndarray:
         return (np.clip(s, 0, 1) * -self.min_db_level) + self.min_db_level