def main():
name = 't3_train'
directory = "./bin/labelled"
audio, fs = fileutils.load_audio(name, audio_dir=directory)
audio_labels = fileutils.load_labels(name, label_dir=directory)
# audio_hashes, audio_pairs, _ = shazam.fingerprint(audio)
# _, matches, segments = thumbnail_shazam(audio, fs, audio_pairs)
# ax = vis.plot_similarity_curve(matches, segments, labels=audio_labels)
# ax.set_title('Similarity Curve for {}'.format(name))
thumb, similarity, segments, sim_matrix = thumbnail(audio,
fs,
seg_method='onset',
length=2)
ax = vis.plot_similarity_matrix(sim_matrix)
ax.set_title('Regular Segmentation Similarity Matrix')
ax = vis.plot_similarity_curve(similarity,
segment_times=segments,
labels=audio_labels)
ax.set_title('Regular Segmentation Similarity')
ax.legend()
# ax = vis.plot_window_overlap(segments, np.ones(segments.shape) * 2, tick_step=3)
# ax.set_title('Regular Segmentation Overlap')
# ax.grid()
vis.show()
return
def main():
name = 't20'
in_dir = './bin/'
out_dir = './bin/results'
audio, fs = fileutils.load_audio(name, audio_dir=in_dir)
# audio_labels = fileutils.load_labels(name, label_dir=in_dir)
# Should be sensitive to the length of the track, as well as k
# Perhaps length should be extended as song goes longer than 30 seconds;
# 3 second = 30 seconds, 18 seconds = 3 min
# length = tune_length_with_audio(audio, fs)
length = cfg.SEGMENT_LENGTH
# explots.draw_results_reference(audio, fs, audio_labels, name=name,
# show_plot=('motif',))
# segmentation_analysis(audio, fs, length, num_motifs=3, name=name,
# show_plot=('arc',))
# k_means_analysis(audio, fs, length, name=name, k_clusters=(5, 25, 50),
# show_plot=('motif',))
thresh = 11
audio_sample = audio
results, G_set = threshold_analysis(audio_sample,
fs,
length,
name=name,
show_plot=('motif', 'matrix'),
threshold=thresh)
write_motifs(audio_sample, fs, name, out_dir, results[thresh])
visutils.show()
return
def main():
name = 'genre_test_3'
directory = "./bin/labelled"
audio, fs = fileutils.load_audio(name, audio_dir=directory)
pairs_hash, pairs, peaks = fingerprint(audio)
sxx = stft(audio,
n_fft=cfg.WINDOW_SIZE,
win_length=cfg.WINDOW_SIZE,
hop_length=int(cfg.WINDOW_SIZE * cfg.OVERLAP_RATIO),
window='hann')
sxx = np.abs(sxx)
sxx = 10 * np.log10(sxx)
sxx[sxx == -np.inf] = 0 # replace infs with zeros
seg_hash_one, _, _ = fingerprint(audio[0 * fs:3 * fs])
seg_hash_two, _, _ = fingerprint(audio[0 * fs:3 * fs])
print(hash_search(seg_hash_one, seg_hash_two))
vis.plot_stft(sxx, fs=fs, frames=False)
vis.plot_peaks(peaks)
vis.plot_pairs(peaks, pairs)
vis.plot_stft_with_pairs(sxx, peaks, pairs)
vis.show()
return None
def setUp(self):
# Read in the audio
name = 't1'
directory = "../bin/"
audio, fs = fileutils.load_audio(name, audio_dir=directory)
# Identify beats in the audio
tempo, beats = lb.beat.beat_track(y=audio, sr=fs, units='time')
beats = np.concatenate([[0], beats, [audio.shape[0]]])
# Construct graph from beats
G = nx.DiGraph(tempo=tempo)
for i in range(0, beats.shape[0] - 1):
G.add_node(i,
start=beats[i],
end=beats[i + 1],
length=(beats[i + 1] - beats[i]))
# Create randomly-weighted edges for segments with exactly the same length
for i in range(0, G.number_of_nodes()):
for j in range(0, G.number_of_nodes()):
if i <= j: # No self loops
continue
if G.node[i]['length'] == G.node[j]['length']:
e_weight = np.random.randn()
G.add_edge(i, j, value=e_weight)
self.G = G
def main():
name = 't1'
directory = "./bin/"
audio, fs = fileutils.load_audio(name, audio_dir=directory)
length = 2
overlap = 0.5
print(segment_onset(audio, fs, length=length, overlap=overlap))
print(segment_onset(audio, fs, length=length, overlap=overlap, method='beat'))
def main():
# Run example with regular segmentation
name = 't1'
directory = "./bin/test"
audio, fs = fileutils.load_audio(name, audio_dir=directory)
audio_labels = fileutils.load_labels(name, label_dir=directory)
thumb, similarity, segments, sim_matrix = thumbnail(audio, fs,
seg_method='regular',
length=2)
ax = vis.plot_similarity_matrix(sim_matrix)
ax.set_title('Regular Segmentation Similarity Matrix')
ax = vis.plot_similarity_curve(similarity, segment_times=segments, labels=audio_labels)
ax.set_title('Regular Segmentation Similarity')
ax.legend()
vis.show()
def main():
name = 'Repeat'
in_dir = './bin/test'
audio, fs = fileutils.load_audio(name, audio_dir=in_dir)
length = cfg.SEGMENT_LENGTH
#
# fig = draw_segmentation_evolution(audio, fs)
# vis.save_fig(fig, './bin/graphs/', 'SEG_{audio_name}'.format(audio_name=name))
#
# thresh = 0.95
thresh = 0.985
starts, ends, labels, G = analyzer.analyze(audio,
fs,
seg_length=length,
threshold=thresh,
seg_method='beat')
#
# fig = vis.get_fig()
# fig.suptitle('Arc Graph Clustering')
# ax = draw_super_axis(fig)
# ax = draw_simple_arc(G, with_labels=True, with_color=True, ax=ax)
# vis.save_fig(fig, './bin/graphs/', 'ARC_{audio_name}_clustered'.format(audio_name=name))
fig = draw_matrix_arc_chord_demo(G, name, with_chord=False)
vis.save_fig(fig, './bin/graphs/',
'SSM2ARC_{audio_name}'.format(audio_name=name))
#
# name = 't3'
# in_dir = './bin/test'
# audio, fs = fileutils.load_audio(name, audio_dir=in_dir)
#
# fig = draw_matrix_evolution(audio, fs, length, name)
# vis.save_fig(fig, './bin/graphs/', 'SSM_{audio_name}'.format(audio_name=name))
# results = {}
# methods = ('With Clustering', 'With Join')
#
# name = 'Avril'
# in_dir = "./bin/test"
# audio, fs = fileutils.load_audio(name, audio_dir=in_dir)
# length = cfg.SEGMENT_LENGTH
#
# audio_labels = fileutils.load_labels(name, label_dir=in_dir)
# ref_starts, ref_ends, ref_labels = motif.df_to_motif(audio_labels)
# fig = vis.get_fig()
# ax = fig.add_subplot(1, 1, 1)
# ax = vis.plot_motif_segmentation(audio, fs, ref_starts, ref_ends, ref_labels, ax=ax)
# fig.suptitle('Hand-Labelled Description of {}'.format(name))
# vis.save_fig(fig, './bin/graphs/', 'IDEAL_{audio_name}'.format(audio_name=name))
# starts, ends, labels, G = analyzer.analyze(audio, fs, seg_length=length, with_join=False)
# this_result = motif.pack_motif(starts, ends, labels)
# results['With Merging'] = this_result
#
# starts, ends, labels, G = analyzer.analyze(audio, fs, seg_length=length, with_join=True)
# this_result = motif.pack_motif(starts, ends, labels)
# results['With Join'] = this_result
#
# fig = draw_motif_group(audio, fs, results, methods=methods, title='Joining Motif Segments', subplots=(2, 1))
# vis.save_fig(fig, './bin/graphs/', 'MOTIF_{audio_name}'.format(audio_name=name))
# name = 't1'
# in_dir = "./bin/"
# audio, fs = fileutils.load_audio(name, audio_dir=in_dir)
# pairs_hash, pairs, peaks = shazam.fingerprint(audio)
#
# sxx = shazam.stft(audio,
# n_fft=cfg.WINDOW_SIZE,
# win_length=cfg.WINDOW_SIZE,
# hop_length=int(cfg.WINDOW_SIZE * cfg.OVERLAP_RATIO),
# window='hann')
# sxx = np.abs(sxx)
#
# fig = vis.get_fig()
# ax = fig.add_subplot(1, 1, 1)
# ax = vis.plot_stft_with_pairs(sxx, peaks, pairs, ax=ax)
# fig.suptitle('Spectrogram With Peaks and Pairs')
#
# vis.save_fig(fig, './bin/graphs/', 'SHAZAM_{audio_name}'.format(audio_name=name))
vis.show()
return
def _experiment(exp_name,
audio_name,
in_dir,
out_dir,
methods,
write_motifs=False,
show_plot=()):
audio, fs = fileutils.load_audio(audio_name, audio_dir=in_dir)
audio_labels = fileutils.load_labels(audio_name, label_dir=in_dir)
ref_starts, ref_ends, ref_labels = motif.df_to_motif(audio_labels)
ref_motifs = motif.pack_motif(ref_starts, ref_ends, ref_labels)
length = cfg.SEGMENT_LENGTH
if exp_name == 'Segmentation':
results, _ = exputils.segmentation_analysis(audio,
fs,
length,
audio_name,
methods=methods,
k=cfg.N_ClUSTERS,
show_plot=show_plot)
elif exp_name == 'Similarity':
results, _ = exputils.similarity_analysis(audio,
fs,
length,
audio_name,
methods=methods,
k=cfg.N_ClUSTERS,
show_plot=show_plot)
elif exp_name == 'K-Means':
results, _ = exputils.k_means_analysis(audio,
fs,
length,
audio_name,
k_clusters=methods,
show_plot=show_plot)
elif exp_name == 'Clustering':
results, _ = exputils.clustering_analysis(audio,
fs,
length,
audio_name,
methods=methods,
k=cfg.N_ClUSTERS,
show_plot=show_plot)
else:
print("Unrecognized experiment name: {exp_name}".format(
exp_name=exp_name))
return
metric_dict = results_to_metrics(results, methods, ref_motifs)
# Output Plots
if exp_name == 'K-Means':
lp = 'k='
else:
lp = ''
# Plot the recall, precision, f-measure, boundary measure, and edit distance as bar plots.
if 'bar' in show_plot:
fig = vis.get_fig()
ax = fig.add_subplot(1, 1, 1)
ax = explots.draw_results_rpf(methods,
metric_dict,
label_prefix=lp,
ax=ax)
fig.suptitle('{exp_name} Performance for {audio_name}'.format(
exp_name=exp_name, audio_name=audio_name))
vis.save_fig(fig, './bin/graphs/',
'RPF_{}_{}'.format(audio_name, exp_name))
fig = vis.get_fig()
explots.draw_results_bed(methods,
metric_dict,
audio_name,
exp_name,
fig=fig)
fig.suptitle("{exp_name} Accuracy on {audio_name}".format(
exp_name=exp_name, audio_name=audio_name),
fontsize=24)
if exp_name == 'K-Means':
ax = fig.get_axes()[0]
ax.set_xlabel('Number of clusters')
ax = fig.get_axes()[1]
ax.set_xlabel('Number of clusters')
vis.save_fig(fig, './bin/graphs/',
'BED_{}_{}'.format(audio_name, exp_name))
# Plot the motif segmentations as subplots in a larger figure
if 'group' in show_plot:
label_key = 'Ideal'
methods_grp = (label_key, ) + methods
results[label_key] = ref_motifs
fig = visualizations.draw_motif_group(audio,
fs,
results,
methods_grp,
title='',
subplots=(2, 2),
label_prefix=lp)
fig.suptitle('{exp_name} Motifs on {audio_name}'.format(
exp_name=exp_name, audio_name=audio_name))
vis.save_fig(fig, './bin/graphs/',
'GRP_{}_{}'.format(audio_name, exp_name))
if exp_name == 'K-Means':
ax = fig.get_axes()[1]
ax.set_title(label_key, fontsize=18)
if write_motifs:
exputils.write_results(audio, fs, audio_name, out_dir, methods,
results)
return metric_dict