def get_features(features, datasets_dir, pca=False):
timestamps_gtzan, feature_vectors_gtzan = Util.read_merged_features(
datasets_dir + "/gtzan/gtzan_combined.wav", features)
labels_gtzan = Util.read_audacity_labels(datasets_dir +
"/gtzan/gtzan_combined.txt")
X_gtzan, Y_gtzan, lbls_gtzan = Util.get_annotated_data_x_y(
timestamps_gtzan, feature_vectors_gtzan, labels_gtzan)
timestamps_labrosa, feature_vectors_labrosa = Util.read_merged_features(
datasets_dir + "/labrosa/labrosa_combined.wav", features)
labels_labrosa = Util.read_audacity_labels(datasets_dir +
"/labrosa/labrosa_combined.txt")
X_labrosa, Y_labrosa, lbls_labrosa = Util.get_annotated_data_x_y(
timestamps_labrosa, feature_vectors_labrosa, labels_labrosa)
timestamps_mirex, feature_vectors_mirex = Util.read_merged_features(
datasets_dir + "/mirex/mirex_combined.wav", features)
labels_mirex = Util.read_audacity_labels(datasets_dir +
"/mirex/mirex_combined.txt")
X_mirex, Y_mirex, lbls_mirex = Util.get_annotated_data_x_y(
timestamps_mirex, feature_vectors_mirex, labels_mirex)
scaler = StandardScaler()
scaler.fit(np.concatenate((X_labrosa, X_gtzan, X_mirex)))
with open("pickled/scaler.pickle", 'w') as f:
pickle.dump(scaler, f)
X_gtzan = scaler.transform(X_gtzan)
X_labrosa = scaler.transform(X_labrosa)
X_mirex = scaler.transform(X_mirex)
if pca:
pca = PCA(n_components=20)
pca.fit(np.concatenate((X_labrosa, X_gtzan, X_mirex)))
X_gtzan = pca.transform(X_gtzan)
X_labrosa = pca.transform(X_labrosa)
X_mirex = pca.transform(X_mirex)
data = {
"x_gtzan": X_gtzan,
"y_gtzan": Y_gtzan,
"labels_gtzan": labels_gtzan,
"x_labrosa": X_labrosa,
"y_labrosa": Y_labrosa,
"labels_labrosa": labels_labrosa,
"x_mirex": X_mirex,
"y_mirex": Y_mirex,
"labels_mirex": labels_mirex,
"timestamps_gtzan": timestamps_gtzan,
"timestamps_labrosa": timestamps_labrosa,
"timestamps_mirex": timestamps_mirex
}
return data
def read_features(features, wavfile, scale=False):
timestamps, feature_vectors = Util.read_merged_features(wavfile, features)
if scale:
with open("/opt/speech-music-discrimination/pickled/scaler.pickle",
'r') as f:
scaler = pickle.load(f)
feature_vectors = scaler.transform(feature_vectors)
return timestamps, feature_vectors