add logmel spec

deeplightning/model/dcgan.py

@@ -44,6 +44,15 @@ def __init__(self, num_channels, latent_dim):
             nn.ConvTranspose2d(in_channels=32, out_channels=num_channels, kernel_size=4, stride=2, padding=1, bias=False),
             nn.Tanh(),
         )
+        """ Parametrize layer creation:
+        last_channels = 32
+        channels = [latent_dim] + [last_channels * (2 ** i) for i in range(num_layers-1)][::-1] + [num_channels]
+        layers = []
+        for i in range(num_layers):
+            layers.append(nn.ConvTranspose2d(in_channels=channels[i], out_channels=channels[i+1], kernel_size=4, stride=2, padding=0, bias=False))
+            layers.append(nn.BatchNorm2d(channels[i+1]))
+            layers.append(nn.ReLU() if i < num_layers-1 else nn.Tanh())
+        """
 
     def forward(self, x):
         return self.generator(x)
@@ -89,7 +98,16 @@ def tensor_size_debugger(self):
 
 class DCGAN(nn.Module):
     def __init__(self, batch_size: int, sample_size: int, image_size: int, latent_dim: int):
-        """
+        """DCGAN: Deep Convolutional Generative Adversarial Network.
+
+        Parameters
+        ----------
+        ???
+
+        References
+        ----------
+        >   A Radford, L Metz, S Chintala (2015) "Unsupervised representation learning with 
+            deep convolutional generative adversarial networks", arXiv:1511.06434
         """
         super(DCGAN, self).__init__()
 

deeplightning/viz/audio/spectrum.py

@@ -70,30 +70,26 @@ def spectrogram(
     scale: str = None, 
     n_fft: int = 2048, 
     hop_length: int = 512, 
-    figsize: tuple = (5,5), 
+    figsize: tuple = (8,3), 
     save_plot: str = None,
     show_plot: bool = True,
+    x_axis: str = "time",
 ):
     """Display Mel Frequency Cepstral Coefficients.
 
     Parameters
     ----------
     path : path to the audio file.
-    
     mode : the type of features to be shown in the spectrogram. 
         Can be Short Time Fourier Transform (STFT) amplitude 
-        (`stft_ampl`) or decibels (`stft_db`); or Mel Frequency 
-        Cepstral Coefficients (MFCC)
-    
+        (`stft_ampl`) or decibels (`stft_db`); Log Mel spectrogram 
+        (logmel); Mel Frequency Cepstral Coefficients (`mfcc`)
     scale : the y-axis scale. Can be `linear` or `log`. For 
         `mode == "mfcc"` the scale is chosen automatically
-     
     n_fft : 
-    
     hop_length :
-    
     figsize : the figure size
-
+    x_axis : 
     """
 
     assert scale is None or scale in ["linear", "log"]
@@ -121,12 +117,19 @@ def spectrogram(
         frequency_type = 'MFCC'
         colorbar_label = 'coefficients'  
         scale_type = ''
+    elif mode == "logmel":
+        S = librosa.feature.melspectrogram(y=signal, sr=sample_rate, n_fft=n_fft, hop_length=hop_length)
+        S = librosa.power_to_db(S, ref=np.max)
+        specshow(data = S, x_axis = x_axis, y_axis = 'mel', sr = sample_rate, hop_length=hop_length)
+        frequency_type = 'Log Mel'
+        colorbar_label = 'decibel (dB)'  
+        scale_type = ''
     else:
         raise NotImplementedError
 
     plt.xlabel("Time")
     plt.ylabel(f"Frequency (Hz)")
-    plt.colorbar(label=colorbar_label, orientation="horizontal")
+    plt.colorbar(label=colorbar_label, orientation="vertical", pad=0.02)
     plt.title("Spectrogram ({}{})".format(
         frequency_type if frequency_type is not None else '',
         scale_type if scale_type is not None else '',

deeplightning/viz/audio/wave.py

@@ -1,3 +1,4 @@
+from typing import Union, Tuple
 import matplotlib.pyplot as plt
 import numpy as np
 import librosa
@@ -9,7 +10,9 @@ def waveplot(
     x_axis: str = "time", 
     save_plot: str = None, 
     show_plot: bool = True, 
-    figsize=(8,4)
+    figsize: Tuple[int,int] = (8,3),
+    title: Union[str,None] = None,
+
 ):
     """Display waveform in the time domain.
 
@@ -33,7 +36,9 @@ def waveplot(
     # plot
     plt.figure(figsize=figsize)
     waveshow(signal, sr=sample_rate, x_axis=x_axis)
-    plt.title("Audio waveplot")
+    plt.ylabel("Amplitude")
+    if title:
+        plt.title(title)
 
     if save_plot is not None:
         plt.savefig(save_plot, bbox_inches='tight')