def find_matches(track, model):
track_id = str(track)
# Extract keypoints
fp = fingerprint.from_file(
track.path, model.settings['sr'], track_id, model.settings
)
# Find (approximate) nearest neighbors
distances, indices = ann.find_neighbors(
model.matcher,
fp.descriptors,
algorithm=model.settings['algorithm'],
k=2
)
# Build match objects
logger.info('Building match objects')
matches = []
for i, distance in enumerate(distances):
matches.append(Match(
fp.keypoints[i],
model.keypoints[indices[i][0]],
distance[0],
distance[1]
))
return matches, fp.spectrogram, fp.keypoints
def find_matches(track, model):
track_id = str(track)
# Extract keypoints
fp = fingerprint.from_file(track.path, model.settings['sr'], track_id,
model.settings)
# Find (approximate) nearest neighbors
distances, indices = ann.find_neighbors(
model.matcher,
fp.descriptors,
algorithm=model.settings['algorithm'],
k=20)
# Build match objects
logger.info('Building match objects')
matches = []
for i, distance in enumerate(distances):
neighbors = [
Neighbor(model.keypoints[index], dist)
for index, dist in zip(indices[i], distance)
]
matches.append(Match(
fp.keypoints[i],
neighbors,
))
return matches, fp.spectrogram, fp.keypoints
def train_keypoints(tracks, hop_length, octave_bins=24, n_octaves=7,
fmin=50, sr=22050, algorithm='lshf', dedupe=False,
save=None, **kwargs):
settings = locals().copy()
for key in settings['kwargs']:
settings[key] = settings['kwargs'][key]
del settings['kwargs']
del settings['tracks']
logger.info('Settings: {}'.format(settings))
spectrograms = {}
keypoints = []
descriptors = []
for track in tracks:
track_id = str(track)
fp = fingerprint.from_file(track.path, sr, track_id, settings)
if dedupe:
# Remove duplicate keypoints
# (important for ratio thresholding if source track
# has exact repeated segments)
fp.remove_similar_keypoints()
if save:
path = save_spectrogram(fp.spectrogram, track_id, save)
spectrograms[track_id] = path
keypoints.extend(fp.keypoints)
descriptors.extend(fp.descriptors)
descriptors = np.vstack(descriptors)
matcher = ann.train_matcher(descriptors, algorithm=algorithm)
model = Model(matcher, keypoints, settings)
model.spectrograms = spectrograms
if save:
save_model(model, save)
return model
def train_keypoints(tracks,
hop_length,
octave_bins=24,
n_octaves=7,
fmin=50,
sr=22050,
algorithm='lshf',
dedupe=False,
save=None,
save_spec=False,
**kwargs):
settings = locals().copy()
for key in settings['kwargs']:
settings[key] = settings['kwargs'][key]
del settings['kwargs']
del settings['tracks']
logger.info('Settings: {}'.format(settings))
spectrograms = {}
keypoints = []
descriptors = []
model_tracks = {}
for track in tracks:
track_id = str(track)
model_tracks[track_id] = track
fp = fingerprint.from_file(track.path, sr, track_id, settings)
if dedupe:
# Remove duplicate keypoints
# (important for ratio thresholding if source track
# has exact repeated segments)
fp.remove_similar_keypoints()
if save_spec:
path = save_spectrogram(fp.spectrogram, track_id, save)
spectrograms[track_id] = path
keypoints.extend(fp.keypoints)
descriptors.extend(fp.descriptors)
descriptors = np.vstack(descriptors)
matcher = ann.train_matcher(descriptors, algorithm=algorithm)
descriptors = None
model = Model(matcher, keypoints, settings, tracks=model_tracks)
model.spectrograms = spectrograms
if save:
save_model(model, save)
return model
