Merge pull request #3378 from plakal/master

Cosmetic changes to spectrogram_to_mel_matrix()

research/audioset/mel_features.py

@@ -150,7 +150,7 @@ def spectrogram_to_mel_matrix(num_mel_bins=20,
     An np.array with shape (num_spectrogram_bins, num_mel_bins).
 
   Raises:
-    ValueError: if frequency edges are incorrectly ordered.
+    ValueError: if frequency edges are incorrectly ordered or out of range.
   """
   nyquist_hertz = audio_sample_rate / 2.
   if lower_edge_hertz < 0.0:
@@ -160,7 +160,7 @@ def spectrogram_to_mel_matrix(num_mel_bins=20,
                      (lower_edge_hertz, upper_edge_hertz))
   if upper_edge_hertz > nyquist_hertz:
     raise ValueError("upper_edge_hertz %.1f is greater than Nyquist %.1f" %
-                     (upper_edge_hertz, nyquist_hertz))  
+                     (upper_edge_hertz, nyquist_hertz))
   spectrogram_bins_hertz = np.linspace(0.0, nyquist_hertz, num_spectrogram_bins)
   spectrogram_bins_mel = hertz_to_mel(spectrogram_bins_hertz)
   # The i'th mel band (starting from i=1) has center frequency