-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathhandfeatures.py
142 lines (115 loc) · 4.72 KB
/
handfeatures.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
import pandas as pd
from pitch_utils import ToriDataset
import math
import numpy as np
from collections import Counter
from sklearn.ensemble import RandomForestClassifier
import random
DIR = "contour_csv/"
def frequency_to_midi(frequency):
# Convert frequency to MIDI note
return 69 + 12 * math.log2(frequency / 440)
def get_midi_contour_from_csv(csv_fn, low_pitch=40, high_pitch=72):
df = pd.read_csv(csv_fn)
frequency = df['frequency'].values
confidence = df['confidence'].values
threshold = 0.8
frequency[confidence < threshold] = np.nan
pitch_in_midi = [frequency_to_midi(freq) for freq in frequency]
return [x for x in pitch_in_midi if low_pitch <= x < high_pitch]
def get_norm_pitch_histogram(midi_contour, compensate_tuning=True):
'''
midi_contour: list of midi pitch as a contour
'''
edge = [i+(-12.5) for i in range(26)]
max_appearance = 0
comp = 0
final_common_midi = 0
if compensate_tuning:
for compensate in (0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9):
intmidi = [round(midi+compensate) for midi in midi_contour]
common_pitch, num_appearance = Counter(intmidi).most_common()[0]
if max_appearance <= num_appearance:
max_appearance = num_appearance
comp = compensate
final_common_midi = common_pitch
else:
intmidi = [round(midi) for midi in midi_contour]
common_pitch, num_appearance = Counter(intmidi).most_common()[0]
final_common_midi = common_pitch
norm_midi = [midi-float(final_common_midi-comp) for midi in midi_contour]
mono_hist = np.histogram(norm_midi, bins = edge, density=True)
return mono_hist[0]
class HistogramMaker:
def __init__(self, resolution=0.2, compensate_tuning=True):
self.resolution = resolution
self.edges, self.bins = self.make_edges(resolution)
self.comp_tuning = compensate_tuning
def make_edges(self, resolution):
edges = [ ]
lowest_interval = -12.5
highest_interval = 12.5
lowest_edge = lowest_interval - resolution / 2
highest_edge = highest_interval + resolution / 2
for i in range(int((highest_edge - lowest_edge) / resolution)):
edges.append(lowest_edge + resolution * i)
bins = np.array(edges) + resolution / 2
bins = bins[:-1]
return edges, bins
def get_tonic(self, midi_contour, compensate_tuning):
max_appearance = 0
comp = 0
final_common_midi = 0
if compensate_tuning:
for compensate in (0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9):
intmidi = [round(midi+compensate) for midi in midi_contour]
common_pitch, num_appearance = Counter(intmidi).most_common()[0]
if max_appearance <= num_appearance:
max_appearance = num_appearance
comp = compensate
final_common_midi = common_pitch
else:
intmidi = [round(midi) for midi in midi_contour]
common_pitch, num_appearance = Counter(intmidi).most_common()[0]
final_common_midi = common_pitch
final_common_midi = final_common_midi - comp
return final_common_midi
def get_norm_pitch_histogram(self, midi_contour, compensate_tuning=True, edge=None):
'''
midi_contour: list of midi pitch as a contour
'''
if edge is None :
edge = [i+(-12.5) for i in range(26)]
tonic = self.get_tonic(midi_contour, compensate_tuning)
# print(f"Most frequently appearing: {final_common_midi}")
norm_midi = [midi-float(tonic) for midi in midi_contour]
mono_hist = np.histogram(norm_midi, bins = edge, density=True)
return mono_hist[0]
def __call__(self, contour):
return self.get_norm_pitch_histogram(contour, self.comp_tuning, self.edges)
def main():
# read meta
meta = pd.read_csv("metadata.csv")
dataset = ToriDataset(meta, DIR, load_all_csv=False)
csv_fns = dataset.splitted_csv_fn_list
midi_contours = [get_midi_contour_from_csv(x) for x in csv_fns]
labels = [dataset.t_meta[dataset.id_with_tori[idx]]['tori'] for idx in range(len(csv_fns))]
norm_hists = [get_norm_pitch_histogram(midi_contour, compensate_tuning=False) for midi_contour in midi_contours]
num_train_samples = 150
accs = []
for i in range(10):
random.seed(i)
rand_train_idx = random.sample(range(len(norm_hists)), num_train_samples)
train_hists = [norm_hists[idx] for idx in rand_train_idx]
train_labels = [labels[idx] for idx in rand_train_idx]
test_hists = [norm_hists[idx] for idx in range(len(norm_hists)) if idx not in rand_train_idx]
test_labels = [labels[idx] for idx in range(len(norm_hists)) if idx not in rand_train_idx]
rf_classifier = RandomForestClassifier()
rf_classifier.fit(train_hists, train_labels)
score = rf_classifier.score(test_hists, test_labels)
accs.append(score)
print(accs)
print(np.mean(accs), np.std(accs))
return
if __name__ == "__main__":
main()