targetDrummers = ['drummer1', 'drummer2', 'drummer3']
parentFolder = '../../unlabeledDrumDataset/evaluation_enst/STFT/'
saveParentFolder = '../../unlabeledDrumDataset/evaluation_enst/Activations/'
# parentFolder = '/Volumes/CW_MBP15/Datasets/unlabeledDrumDataset/evaluation_enst/STFT/'
# saveParentFolder = '/Volumes/CW_MBP15/Datasets/unlabeledDrumDataset/evaluation_enst/Activations/'
modelpath = './models/dnn_model_1nn_lstm_rc.h5'
'''
==== File IO + testing
'''
model = load_model(modelpath)
for drummer in targetDrummers:
#==== get STFT
stftpath = parentFolder + drummer + '/'
stftFilePathList = getFilePathList(stftpath, 'mat')
saveFolder = saveParentFolder + drummer + '/'
for i in range(0, len(stftFilePathList)):
print 'test on file %f' % i
filename = stftFilePathList[i][-11:-4]
savepath = saveFolder + filename
tmp = loadmat(stftFilePathList[i])
X = np.ndarray.transpose(tmp['X'])
X = np.reshape(X, (X.shape[0], 1, X.shape[1]))
'''
==== Testing
'''
Y = model.predict(X, batch_size=64)
all = [Y[:, 0], Y[:, 1], Y[:, 2]]
np.save(savepath, all)
targetDrummers = ['drummer1', 'drummer2', 'drummer3']
parentFolder = '../../unlabeledDrumDataset/evaluation_enst/CQT/'
saveParentFolder = '../../unlabeledDrumDataset/evaluation_enst/Activations/'
# parentFolder = '/Volumes/CW_MBP15/Datasets/unlabeledDrumDataset/evaluation_enst/STFT/'
# saveParentFolder = '/Volumes/CW_MBP15/Datasets/unlabeledDrumDataset/evaluation_enst/Activations/'
modelpath = './models/dnn_model_cqt_rms.h5'
'''
==== File IO + testing
'''
model = load_model(modelpath)
for drummer in targetDrummers:
#==== get STFT
cqtpath = parentFolder + drummer + '/'
cqtFilePathList = getFilePathList(cqtpath, 'mat')
saveFolder = saveParentFolder + drummer + '/'
for i in range(0, len(cqtFilePathList)):
print 'test on file %f' % i
filename = cqtFilePathList[i][-11:-4]
savepath = saveFolder + filename
tmp = loadmat(cqtFilePathList[i])
X = np.ndarray.transpose(tmp['Xcqt'])
'''
==== Testing
'''
X_diff = np.diff(X, axis=0)
finalRow = np.zeros((1, np.size(X, 1)))
X_diff = np.concatenate((X_diff, finalRow), axis=0)
X_all = np.concatenate((X, X_diff), axis=1)
histogram_freq=0,
write_graph=True,
write_images=True)
# earlyStopCallBack = EarlyStopping(monitor='loss', min_delta=1e-6, patience=5)
'''
==== File IO + file concatenation
'''
X = np.ndarray((0, 121))
Y = np.ndarray((0, 3))
for method in targetPseudoLabels:
for genre in targetGenres:
#==== get STFT & pseudo label
cqtpath = parentFolder + 'CQT/' + genre + '/'
cqtFilePathList = getFilePathList(cqtpath, 'mat')
pseudoLabelPath = parentFolder + method + '/' + genre + '/'
pseudoLabelFilePathList = getFilePathList(pseudoLabelPath, 'mat')
for i in range(0, 200): #len(stftFilePathList)):
tmp = loadmat(cqtFilePathList[i])
X_song = np.ndarray.transpose(tmp['Xcqt'])
tmp = loadmat(pseudoLabelFilePathList[i])
Y_song = np.ndarray.transpose(tmp['HD'])
assert (len(X_song) == len(Y_song)
), 'dimensionality mismatch between CQT and Pseudo-Labels!'
'''
==== Concatenating matrices
'''
X = np.concatenate((X, X_song), axis=0)
Y = np.concatenate((Y, Y_song), axis=0)
'drummer2',
'drummer3']
parentFolder = '../../unlabeledDrumDataset/evaluation_enst/200drums/'
parentFolder2 = '../../unlabeledDrumDataset/evaluation_enst/enst/'
parentFolder3 = '../../unlabeledDrumDataset/evaluation_enst/smt/'
saveParentFolder = '../../unlabeledDrumDataset/evaluation_enst/Activations/'
# parentFolder = '/Volumes/CW_MBP15/Datasets/unlabeledDrumDataset/evaluation_enst/STFT/'
# saveParentFolder = '/Volumes/CW_MBP15/Datasets/unlabeledDrumDataset/evaluation_enst/Activations/'
'''
==== File IO + testing
'''
for drummer in targetDrummers:
#==== get STFT
stftpath = parentFolder + drummer + '/'
stftFilePathList = getFilePathList(stftpath, 'mat')
stftpath2 = parentFolder2 + drummer + '/'
stftFilePathList2 = getFilePathList(stftpath2, 'mat')
stftpath3 = parentFolder3 + drummer + '/'
stftFilePathList3 = getFilePathList(stftpath3, 'mat')
saveFolder = saveParentFolder + drummer + '/'
for i in range(0, len(stftFilePathList)):
print 'test on file %f' % i
filename = stftFilePathList[i][-11:-4]
savepath = saveFolder + filename
tmp = loadmat(stftFilePathList[i])
Y1 = np.ndarray.transpose(tmp['HD'])
filename2 = stftFilePathList2[i][-11:-4]
model.add(Dropout(0.25))
model.add(Dense(units = 3))
model.add(Activation('sigmoid'))
model.compile(optimizer = 'rmsprop', loss='mse', metrics = ['mae'])
return model
model = createModel()
'''
==== File IO + training
'''
for method in targetPseudoLabels:
for genre in targetGenres:
#==== get STFT & pseudo label
stftpath = parentFolder + 'STFT/' + genre + '/'
stftFilePathList = getFilePathList(stftpath, 'mat')
pseudoLabelPath = parentFolder + method + '/' + genre + '/'
pseudoLabelFilePathList = getFilePathList(pseudoLabelPath, 'mat')
for i in range(0, len(stftFilePathList)):
tmp = loadmat(stftFilePathList[i])
X = np.ndarray.transpose(tmp['X'])
tmp = loadmat(pseudoLabelFilePathList[i])
Y = np.ndarray.transpose(tmp['HD'])
assert (len(X) == len(Y)), 'dimensionality mismatch between STFT and Pseudo-Labels!'
'''
==== Training
'''
[y_hh_scaled, dump, dump] = scaleData(Y[:, 0])
[y_kd_scaled, dump, dump] = scaleData(Y[:, 1])
hopSize = 512.00
fs = 44100.00
order = round(0.1 / (hopSize / fs))
offset = 0.12
'''
==== File IO + testing
'''
hh_all_results = []
bd_all_results = []
sd_all_results = []
for drummer in targetDrummers:
#==== get STFT
activpath = parentFolder + drummer + '/'
annpath = annotationFolder + drummer + '/'
activFilePathList = getFilePathList(activpath, 'npy')
annFilePathList = getFilePathList(annpath, 'mat')
for i in range(0, len(activFilePathList)):
print 'evaluate file %f' % i
filename = activFilePathList[i]
all = np.load(filename)
filename_ann = annFilePathList[i]
ann = loadmat(filename_ann)
drums_ann = ann['drums']
onsets_ann = ann['onsets']
ref_hh, ref_bd, ref_sd = getIndividualOnset(onsets_ann, drums_ann)
#==== load activation functions
activ_hh = all[0]
activ_bd = all[1]
activ_sd = all[2]