def extract_features(track, fctr=400, fsd=1.0, type=1):
"""Computes beat-synchronous chroma features.
Uses Dan Ellis' chrombeatftrs Matlab function (via the mlabwrap
module, which is included with this feature extractor).
See http://labrosa.ee.columbia.edu/projects/coversongs
for more details.
Parameters
----------
track : gordon Track instance
fctr : float
Center frequency (in Hz) for chromagram weighting.
fsd : float
Standard deviation (in octaves) for chromagram weighting.
type : int
Selects chroma calculation type; 1 (default) uses IF; 2 uses
all FFT bins, 3 uses only local peaks (a bit like Emilia).
Returns
-------
chroma : array, shape (n, 12)
Feature vector of beat-level chroma features (n time
step rows x 12 columns)
beats : array, length n
Beat times in seconds for each frame of chroma.
"""
x,fs,svals = track.audio()
feats,beats = mlab.chrombeatftrs(x.mean(1)[:,np.newaxis], fs, fctr, fsd,
type, nout=2)
return feats.T, beats.flatten()
def extract_features(wavfilename, fctr=400, fsd=1.0, type=1):
"""Computes beat-synchronous chroma features from the given wave file
Calls Dan Ellis' chrombeatftrs Matlab function.
"""
x, fs = mlab.wavread(wavfilename, nout=2)
feats, beats = mlab.chrombeatftrs(x, fs, fctr, fsd, type, nout=2)
return feats, beats.flatten()
def extract_features(wavfilename, fctr=400, fsd=1.0, type=1):
"""Computes beat-synchronous chroma features from the given wave file
Calls Dan Ellis' chrombeatftrs Matlab function.
"""
x, fs = mlab.wavread(wavfilename, nout=2)
feats, beats = mlab.chrombeatftrs(x, fs, fctr, fsd, type, nout=2)
return feats, beats.flatten()
def extract_features(wavfilename, fctr=400, fsd=1.0, type=1):
"""Computes beat-synchronous chroma features from the given wave file
Calls Dan Ellis' chrombeatftrs Matlab function.
"""
logger.info('Extracting beat-synchronous chroma features from %s',
wavfilename)
x,fs = mlab.wavread(wavfilename, nout=2)
feats,beats = mlab.chrombeatftrs(x.mean(1)[:,np.newaxis], fs, fctr, fsd,
type, nout=2)
songlen = x.shape[0] / fs
return feats, beats.flatten(), songlen
def extract_features(wavfilename, fctr=400, fsd=1.0, type=1):
"""Computes beat-synchronous chroma features from the given wave file
Calls Dan Ellis' chrombeatftrs Matlab function.
"""
logger.info('Extracting beat-synchronous chroma features from %s',
wavfilename)
x, fs = mlab.wavread(wavfilename, nout=2)
feats, beats = mlab.chrombeatftrs(x.mean(1)[:, np.newaxis],
fs,
fctr,
fsd,
type,
nout=2)
songlen = x.shape[0] / fs
return feats, beats.flatten(), songlen
def extract_features(wavfilename, fctr=400, fsd=1.0, type=1):
"""Computes beat-synchronous chroma features from the given wave file
Calls Dan Ellis' chrombeatftrs Matlab function.
"""
if lower(wavfilename[-4:]) == '.csv':
logger.info('CSV filename reading preprocessed features from %s',
wavfilename)
csvr = csv.reader(open(wavfilename, 'rb'), delimiter=',')
feats = np.array([[float(x) for x in row] for row in csvr])
beats = np.arange(len(feats)) * 0.0058 # simple fake frame locations
songlen = len(feats) * 0.0058
else:
logger.info('Extracting beat-synchronous chroma features from %s',
wavfilename)
x,fs = mlab.wavread(wavfilename, nout=2)
feats,beats = mlab.chrombeatftrs(x.mean(1)[:,np.newaxis], fs, fctr, fsd,
type, nout=2)
songlen = x.shape[0] / fs
return feats, beats.flatten(), songlen
