perceptual_weighting moved into core

librosa/core.py

@@ -621,6 +621,37 @@ def logamplitude(S, ref_power=1.0, amin=1e-10, top_db=80.0):
 
     return log_spec
 
+def perceptual_weighting(S, frequencies, **kwargs):
+    '''Perceptual weighting of a power spectrogram:
+
+    ``S_p[f] = A_weighting(f) + 10*log(S[f] / ref_power)``
+
+    :usage:
+        >>> # Re-weight a CQT representation, using peak power as reference
+        >>> CQT             = librosa.cqt(y, sr, fmin=55, fmax=440)
+        >>> freqs           = librosa.cqt_frequencies(CQT.shape[0], fmin=55)
+        >>> percept_CQT     = librosa.perceptual_weighting(CQT, freqs,
+                                                            ref_power=np.max)
+
+    :parameters:
+      - S : np.ndarray, shape=(d,t)
+          Power spectrogram
+
+      - frequencies : np.ndarray, shape=(d,)
+          Center frequency for each row of ``S``
+
+      - *kwargs*
+          Additional keyword arguments to pass to ``librosa.logamplitude``.
+
+    :returns:
+      - S_p : np.ndarray, shape=(d,t)
+          perceptually weighted version of ``S``
+    '''
+
+    offset = A_weighting(frequencies).reshape((-1, 1))
+
+    return offset + logamplitude(S, **kwargs)
+
 def magphase(D):
     """Separate a complex-valued spectrogram D into its magnitude (S)
     and phase (P) components, so that ``D = S * P``.

librosa/feature.py

@@ -538,7 +538,7 @@ def piptrack(y=None, sr=22050, S=None, n_fft=4096, fmin=150.0, fmax=4000.0, thre
     return pitches, mags
 
 #-- Mel spectrogram and MFCCs --#
-def mfcc(S=None, y=None, sr=22050, n_mfcc=20):
+def mfcc(y=None, sr=22050, S=None, n_mfcc=20, **kwargs):
     """Mel-frequency cepstral coefficients
 
     :usage:
@@ -553,18 +553,22 @@ def mfcc(S=None, y=None, sr=22050, n_mfcc=20):
         >>> mfccs = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=40)
 
     :parameters:
-      - S     : np.ndarray or None
-          log-power Mel spectrogram
-
       - y     : np.ndarray or None
           audio time series
 
       - sr    : int > 0
           sampling rate of ``y``
 
+      - S     : np.ndarray or None
+          log-power Mel spectrogram
+
       - n_mfcc: int > 0
           number of MFCCs to return
 
+      - *kwargs*
+          Additional keyword arguments for ``librosa.feature.melspectrogram``, if
+          operating on time series data
+
     .. note::
         One of ``S`` or ``y, sr`` must be provided.
 
@@ -577,7 +581,7 @@ def mfcc(S=None, y=None, sr=22050, n_mfcc=20):
     """
 
     if S is None:
-        S = librosa.logamplitude(melspectrogram(y=y, sr=sr))
+        S = librosa.logamplitude(melspectrogram(y=y, sr=sr, **kwargs))
 
     return np.dot(librosa.filters.dct(n_mfcc, S.shape[0]), S)