fix dependencies with librosa

CHANGELOG.md

@@ -1,3 +1,7 @@
+## [1.1.5] - 2022-10-17
+### Changed
+- Fixed dependencies with `librosa` library
+
 ## [1.1.4] - 2022-09-29
 ### Changed
 - Improoved `mltu.torch.dataProvider.DataProvider` to hangle `multiprocessing` when it doesn't work to switch to `multithreading`

mltu/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "1.1.4"
+__version__ = "1.1.5"
 
 from .annotations.images import Image
 from .annotations.images import CVImage

mltu/augmentors.py

@@ -581,11 +581,6 @@ class RandomAudioPitchShift(Augmentor):
         augment_annotation (bool): Whether to augment the annotation. Defaults to False.
         max_n_steps (int): Maximum number of steps to shift audio. Defaults to 5.
     """
-    try:
-        import librosa
-        # samplerate
-    except ImportError:
-        raise ImportError("librosa is required to augment Audio. Please install it with `pip install librosa`.")
     def __init__(
             self, 
             random_chance: float = 0.5,
@@ -596,6 +591,12 @@ def __init__(
         super(RandomAudioPitchShift, self).__init__(random_chance, log_level, augment_annotation)
         self.max_n_steps = max_n_steps
 
+        try:
+            import librosa
+            # samplerate
+        except ImportError:
+            raise ImportError("librosa is required to augment Audio. Please install it with `pip install librosa`.")
+
     def augment(self, audio: Audio) -> Audio:
         random_n_steps = np.random.randint(-self.max_n_steps, self.max_n_steps)
         # changing default res_type "kaiser_best" to "linear" for speed and memory efficiency
@@ -617,10 +618,6 @@ class RandomAudioTimeStretch(Augmentor):
         min_rate (float): Minimum rate to stretch audio. Defaults to 0.8.
         max_rate (float): Maximum rate to stretch audio. Defaults to 1.2.
     """
-    try:
-        import librosa
-    except ImportError:
-        raise ImportError("librosa is required to augment Audio. Please install it with `pip install librosa`.")
     def __init__(
             self, 
             random_chance: float = 0.5,
@@ -633,6 +630,11 @@ def __init__(
         self.min_rate = min_rate
         self.max_rate = max_rate
 
+        try:
+            import librosa
+        except ImportError:
+            raise ImportError("librosa is required to augment Audio. Please install it with `pip install librosa`.")
+
     def augment(self, audio: Audio) -> Audio:
         random_rate = np.random.uniform(self.min_rate, self.max_rate)
         stretch_audio = self.librosa.effects.time_stretch(audio.numpy(), rate=random_rate)

mltu/preprocessors.py

@@ -57,12 +57,6 @@ class AudioReader:
         sample_rate (int): Sample rate. Defaults to None.
         log_level (int): Log level. Defaults to logging.INFO.
     """
-    try:
-        import librosa
-        librosa.__version__
-    except AttributeError:
-        raise ImportError("librosa is required to read WAV files. Please install it with `pip install librosa`.")
-
     def __init__(
             self, 
             sample_rate = None,
@@ -72,6 +66,12 @@ def __init__(
         self.logger = logging.getLogger(self.__class__.__name__)
         self.logger.setLevel(log_level)
 
+        try:
+            import librosa
+            librosa.__version__
+        except AttributeError:
+            raise ImportError("librosa is required to read WAV files. Please install it with `pip install librosa`.")
+
     def __call__(self, audio_path: str, label: typing.Any) -> typing.Tuple[np.ndarray, typing.Any]:
         """ Read audio from path and return audio and label
         
@@ -105,12 +105,6 @@ class WavReader:
         frame_step (int): Step size between frames in samples.
         fft_length (int): Number of FFT components.
     """
-    # Check if librosa is installed
-    try:
-        import librosa
-        librosa.__version__
-    except AttributeError:
-        raise ImportError("librosa is required to read WAV files. Please install it with `pip install librosa`.")
 
     def __init__(
             self,
@@ -124,6 +118,12 @@ def __init__(
         self.fft_length = fft_length
 
         matplotlib.interactive(False)
+        # Check if librosa is installed
+        try:
+            import librosa
+            librosa.__version__
+        except AttributeError:
+            raise ImportError("librosa is required to read WAV files. Please install it with `pip install librosa`.")
 
     @staticmethod
     def get_spectrogram(wav_path: str, frame_length: int, frame_step: int, fft_length: int) -> np.ndarray:

mltu/tensorflow/callbacks.py

@@ -5,15 +5,6 @@
 import logging
 
 class Model2onnx(Callback):
-    try:
-        import tf2onnx
-    except:
-        raise ImportError("tf2onnx is not installed. Please install it using 'pip install tf2onnx'")
-
-    try:
-        import onnx
-    except:
-        raise ImportError("onnx is not installed. Please install it using 'pip install onnx'")
     """ Converts the model to onnx format after training is finished. """
     def __init__(
         self, 
@@ -32,6 +23,16 @@ def __init__(
         self.metadata = metadata
         self.save_on_epoch_end = save_on_epoch_end
 
+        try:
+            import tf2onnx
+        except:
+            raise ImportError("tf2onnx is not installed. Please install it using 'pip install tf2onnx'")
+
+        try:
+            import onnx
+        except:
+            raise ImportError("onnx is not installed. Please install it using 'pip install onnx'")
+
     @staticmethod
     def model2onnx(model: tf.keras.Model, onnx_model_path: str):
         try:

mltu/transformers.py

@@ -252,11 +252,6 @@ class AudioToSpectrogram(Transformer):
         fft_length (int): Number of FFT components.
         log_level (int): Logging level (default: logging.INFO)
     """
-    try:
-        import librosa
-    except ImportError:
-        raise ImportError("librosa is required to transform Audio. Please install it with `pip install librosa`.")
-
     def __init__(
             self,
             frame_length: int = 256,
@@ -269,6 +264,11 @@ def __init__(
         self.frame_step = frame_step
         self.fft_length = fft_length
 
+        try:
+            import librosa
+        except ImportError:
+            raise ImportError("librosa is required to transform Audio. Please install it with `pip install librosa`.")
+
     def __call__(self, audio: Audio, label: typing.Any):
         """Compute the spectrogram of a WAV file