Merge 93a99cd into 707ffd1

librosa/filters.py

@@ -154,7 +154,8 @@ def dct(n_filters, n_input):
 
 
 @cache(level=10)
-def mel(sr, n_fft, n_mels=128, fmin=0.0, fmax=None, htk=False):
+def mel(sr, n_fft, n_mels=128, fmin=0.0, fmax=None, htk=False,
+        norm=1):
     """Create a Filterbank matrix to combine FFT bins into Mel-frequency bins
 
     Parameters
@@ -178,6 +179,11 @@ def mel(sr, n_fft, n_mels=128, fmin=0.0, fmax=None, htk=False):
     htk       : bool [scalar]
         use HTK formula instead of Slaney
 
+    norm : {None, 1, np.inf} [scalar]
+        if 1, divide the triangular mel weights by the width of the mel band 
+        (area normalization).  Otherwise, leave all the triangles aiming for 
+        a peak value of 1.0
+
     Returns
     -------
     M         : np.ndarray [shape=(n_mels, 1 + n_fft/2)]
@@ -220,6 +226,9 @@ def mel(sr, n_fft, n_mels=128, fmin=0.0, fmax=None, htk=False):
     if fmax is None:
         fmax = float(sr) / 2
 
+    if norm is not None and norm != 1 and norm != np.inf:
+        raise ParameterError('Unsupported norm: {}'.format(repr(norm)))
+
     # Initialize the weights
     n_mels = int(n_mels)
     weights = np.zeros((n_mels, int(1 + n_fft // 2)))
@@ -241,9 +250,10 @@ def mel(sr, n_fft, n_mels=128, fmin=0.0, fmax=None, htk=False):
         # .. then intersect them with each other and zero
         weights[i] = np.maximum(0, np.minimum(lower, upper))
 
-    # Slaney-style mel is scaled to be approx constant energy per channel
-    enorm = 2.0 / (mel_f[2:n_mels+2] - mel_f[:n_mels])
-    weights *= enorm[:, np.newaxis]
+    if norm == 1:
+        # Slaney-style mel is scaled to be approx constant energy per channel
+        enorm = 2.0 / (mel_f[2:n_mels+2] - mel_f[:n_mels])
+        weights *= enorm[:, np.newaxis]
 
     # Only check weights if f_mel[0] is positive
     if not np.all((mel_f[:-2] == 0) | (weights.max(axis=1) > 0)):

tests/test_filters.py

@@ -87,15 +87,15 @@ def __test_to_octs(infile):
 
 def test_melfb():
 
-    def __test(infile):
+    def __test_default_norm(infile):
         DATA = load(infile)
 
-        wts = librosa.filters.mel(DATA['sr'][0],
-                                  DATA['nfft'][0],
-                                  n_mels=DATA['nfilts'][0],
-                                  fmin=DATA['fmin'][0],
-                                  fmax=DATA['fmax'][0],
-                                  htk=DATA['htk'][0])
+        wts = librosa.filters.mel(DATA['sr'][0, 0],
+                                  DATA['nfft'][0, 0],
+                                  n_mels=DATA['nfilts'][0, 0],
+                                  fmin=DATA['fmin'][0, 0],
+                                  fmax=DATA['fmax'][0, 0],
+                                  htk=DATA['htk'][0, 0])
 
         # Our version only returns the real-valued part.
         # Pad out.
@@ -107,7 +107,32 @@ def __test(infile):
         assert np.allclose(wts, DATA['wts'])
 
     for infile in files('data/feature-melfb-*.mat'):
-        yield (__test, infile)
+        yield (__test_default_norm, infile)
+
+    def __test_with_norm(infile):
+        DATA = load(infile)
+        # if DATA['norm'] is empty, pass None.
+        if DATA['norm'].shape[-1] == 0:
+            norm = None
+        else:
+            norm = DATA['norm'][0, 0]
+        wts = librosa.filters.mel(DATA['sr'][0, 0],
+                                  DATA['nfft'][0, 0],
+                                  n_mels=DATA['nfilts'][0, 0],
+                                  fmin=DATA['fmin'][0, 0],
+                                  fmax=DATA['fmax'][0, 0],
+                                  htk=DATA['htk'][0, 0], 
+                                  norm=norm)
+        # Pad out.
+        wts = np.pad(wts, [(0, 0),
+                           (0, int(DATA['nfft'][0]//2 - 1))],
+                     mode='constant')
+
+        eq_(wts.shape, DATA['wts'].shape)
+        assert np.allclose(wts, DATA['wts'])
+
+    for infile in files('data/feature-melfbnorm-*.mat'):
+        yield (__test_with_norm, infile)
 
 
 def test_mel_gap():