def compute_feature_map(input_file, annotations_file, delimiter,
downbeat_label, n_tatums):
'''Accentuation feature map computation
:parameters:
- input_file : str
path to input audio file (wav, mp3, m4a, flac, etc.)
- annotations_file : str
path to the annotations file (txt, csv, etc)
- delimiter: str
delimiter string to process the annotations file
- downbeat_label: str
string to look for in the label data to select downbeats
- n_tatums: int
number of tatums (subdivisions) per tactus beat
- n_clusters: int
number of clusters for rhythmic patterns clustering
'''
# 1. load the wav file
print('Loading audio file ...', input_file)
y, sr = audio.load(input_file, sr=None)
# 2. load beat and downbeat annotations
print('Loading beat and downbeat annotations ...', annotations_file)
beats, _ = annotations.load_beats(annotations_file, delimiter=delimiter)
downbeats, _ = annotations.load_downbeats(annotations_file,
delimiter=delimiter,
downbeat_label=downbeat_label)
# number of beats per bar
n_beats = int(round(beats.size / downbeats.size))
# 3. compute accentuation feature
print('Computing accentuation feature ...')
acce, times, _ = features.accentuation_feature(y,
sr,
minfreq=20,
maxfreq=200)
# 4. compute feature map
print('Computing feature map ...')
map_acce, _, _, _ = features.feature_map(acce,
times,
beats,
downbeats,
n_beats=n_beats,
n_tatums=n_tatums)
# 5. plot feature map
plt.figure()
ax1 = plt.subplot(211)
display.map_show(map_acce, ax=ax1, n_tatums=n_tatums)
plt.show()
# We compute the feature map of rhythmic patterns and we
# learn a manifold in a low--dimensional space.
# The patterns are they shown in the low--dimensional space
# before and after being grouped into clusters.
#
# First, we'll load one of the audio files included in `carat`.
audio_path = util.example("ansina_audio")
y, sr = audio.load(audio_path)
##############################################
# Next, we'll load the annotations provided for the example audio file.
annotations_path = util.example("ansina_beats")
beats, beat_labs = annotations.load_beats(annotations_path)
downbeats, downbeat_labs = annotations.load_downbeats(annotations_path)
##############################################
# Then, we'll compute the accentuation feature.
#
# **Note:** This example is tailored towards the rhythmic patterns of the lowest
# sounding of the three drum types taking part in the recording, so the analysis
# focuses on the low frequencies (20 to 200 Hz).
acce, times, _ = features.accentuation_feature(y, sr, minfreq=20, maxfreq=200)
##############################################
# Next, we'll compute the feature map.
n_beats = int(round(beats.size / downbeats.size))
n_tatums = 4
map_acce, _, _, _ = features.feature_map(acce,
def rhythmic_patterns_analysis(input_file, annotations_file, delimiter,
downbeat_label, n_tatums, n_clusters):
'''Rhythmic patterns analysis
:parameters:
- input_file : str
path to input audio file (wav, mp3, m4a, flac, etc.)
- annotations_file : str
path to the annotations file (txt, csv, etc)
- delimiter: str
delimiter string to process the annotations file
- downbeat_label: str
string to look for in the label data to select downbeats
- n_tatums: int
number of tatums (subdivisions) per tactus beat
- n_clusters: int
number of clusters for rhythmic patterns clustering
'''
# 1. load the wav file
print('Loading audio file ...', input_file)
y, sr = audio.load(input_file, sr=None)
# 2. load beat and downbeat annotations
print('Loading beat and downbeat annotations ...', annotations_file)
beats, _ = annotations.load_beats(annotations_file, delimiter=delimiter)
downbeats, _ = annotations.load_downbeats(annotations_file,
delimiter=delimiter,
downbeat_label=downbeat_label)
# number of beats per bar
n_beats = int(round(beats.size / downbeats.size))
# 3. compute accentuation feature
print('Computing accentuation feature ...')
acce, times, _ = features.accentuation_feature(y,
sr,
minfreq=20,
maxfreq=200)
# 4. compute feature map
print('Computing feature map ...')
map_acce, _, _, _ = features.feature_map(acce,
times,
beats,
downbeats,
n_beats=n_beats,
n_tatums=n_tatums)
# 5. cluster rhythmic patterns
print('Clustering rhythmic patterns ...')
cluster_labs, centroids, _ = clustering.rhythmic_patterns(
map_acce, n_clusters=n_clusters)
# 6. manifold learning
print('Dimensionality reduction ...')
map_emb = clustering.manifold_learning(map_acce)
# 7. plot everything
plt.figure()
ax1 = plt.subplot(211)
# plot feature map
display.map_show(map_acce, ax=ax1, n_tatums=n_tatums)
ax2 = plt.subplot(212)
# plot feature map with clusters in colors
display.map_show(map_acce,
ax=ax2,
n_tatums=n_tatums,
clusters=cluster_labs)
fig = plt.figure()
ax3 = fig.add_subplot(111, projection='3d')
# plot low-dimensional embedding of feature data
display.embedding_plot(map_emb, ax=ax3, clusters=cluster_labs)
fig = plt.figure()
display.centroids_plot(centroids, n_tatums=n_tatums)
plt.show()