-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathspeaker_embedding2.py
163 lines (145 loc) · 5.35 KB
/
speaker_embedding2.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
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
from resemblyzer import preprocess_wav, VoiceEncoder
import numpy as np
import pickle
from sklearn.decomposition import PCA
from sklearn.manifold import TSNE
import matplotlib.pyplot as plt
import csv
from tqdm import tqdm
SPEAKERS = ['p364', 'p256', 'p299', 'p313', 'p306', 'p261', 'p244', 'p304', 'p240', 'p305', 'p250', 'p278', 'p255',
'p239', 'p293', 'p288', 'p225', 'p316', 'p270', 'p286', 'p334', 'p271', 'p279', 'p336', 'p343', 'p295',
'p285', 'p312', 'p265', 's5', 'p326', 'p262', 'p361', 'p253', 'p302', 'p226', 'p269', 'p341', 'p300',
'p259', 'p252', 'p317', 'p376', 'p268', 'p247', 'p318', 'p287', 'p251', 'p330', 'p301', 'p283', 'p335',
'p329', 'p275', 'p314', 'p238', 'p297', 'p311', 'p345', 'p254', 'p363', 'p236', 'p284', 'p282', 'p245',
'p294', 'p276', 'p281', 'p234', 'p258', 'p351', 'p241', 'p340', 'p232', 'p237', 'p228', 'p339', 'p310',
'p267', 'p264', 'p323', 'p233', 'p298', 'p227', 'p230', 'p229', 'p333', 'p260', 'p308', 'p231', 'p263',
'p362', 'p292', 'p257', 'p360', 'p249', 'p246', 'p248', 'p374', 'p266', 'p274', 'p272', 'p307', 'p277',
'p273', 'p243', 'p303', 'p347']
"""## Compute the embeddings
encoder = VoiceEncoder().cuda()
embed_list = []
speaker_list = []
with open("data/VCTK-Corpus-0.92/splits/train.txt") as split_f:
csv_reader = csv.reader(split_f, delimiter="|")
file_count = 0
next(csv_reader, None)
for row in tqdm(csv_reader):
path_to_wav = row[0]
speaker = row[-1]
speaker = int(speaker)
wav = preprocess_wav(path_to_wav)
embedding = encoder.embed_utterance(wav)
embed_list.append(embedding)
speaker_list.append(speaker)
with open('embed_list.pickle', 'wb') as handle:
pickle.dump(embed_list, handle, protocol=pickle.HIGHEST_PROTOCOL)
with open('speaker_list.pickle', 'wb') as handle:
pickle.dump(speaker_list, handle, protocol=pickle.HIGHEST_PROTOCOL)
"""
with open('embed_list.pickle', 'rb') as handle:
embed_list = pickle.load(handle)
with open('speaker_list.pickle', 'rb') as handle:
speaker_list = pickle.load(handle)
gender_dict = {}
with open("data/VCTK-Corpus-0.92/speaker-info.txt") as f:
lines = f.readlines()
for text in tqdm(lines[1:]):
try:
row = text.split()
speaker = SPEAKERS.index(row[0])
gender = row[2]
accent = row[3]
gender_dict[speaker] = [gender, accent]
except:
continue
## Project the embeddings in 2D space
"""plot_projections(utterance_embeds, speakers, title="Embedding projections")
plt.show()"""
tsne = TSNE(n_components=2)
pca = PCA(n_components=2)
fig, ax = plt.subplots()
y = pca.fit_transform(np.array(embed_list))
x_pl = []
y_pl = []
for speaker in range(len(SPEAKERS)):
x_pl.append([i for (i, s) in zip(y[:, 0], speaker_list) if s==speaker])
y_pl.append([i for (i, s) in zip(y[:, 1], speaker_list) if s==speaker])
if gender_dict[speaker][0] == "F":
marker = "."
else:
marker = "x"
if gender_dict[speaker][1] == "American":
c = "blue"
elif gender_dict[speaker][1] == "English":
c = "red"
elif gender_dict[speaker][1] == "Canadian":
c = "purple"
elif gender_dict[speaker][1] == "Indian":
c = "yellow"
elif gender_dict[speaker][1] == "Scottish":
c = "green"
elif "Irish" in gender_dict[speaker][1]:
c = "pink"
else:
c = "black"
ax.scatter(x_pl[speaker], y_pl[speaker], marker=marker, c=c)
#plt.legend(SPEAKERS)
plt.savefig("fixed_encoding_pca_vctk.png")
plt.show()
fig, ax = plt.subplots()
for speaker in range(len(SPEAKERS)):
if gender_dict[speaker][0] == "F":
marker = "."
else:
marker = "x"
if gender_dict[speaker][1] == "American":
c = "blue"
elif gender_dict[speaker][1] == "English":
c = "red"
elif gender_dict[speaker][1] == "Canadian":
c = "purple"
elif gender_dict[speaker][1] == "Indian":
c = "yellow"
elif gender_dict[speaker][1] == "Scottish":
c = "green"
elif "Irish" in gender_dict[speaker][1]:
c = "pink"
else:
c = "black"
ax.scatter(np.mean(x_pl[speaker]), np.mean(y_pl[speaker]), marker=marker, c=c)
#plt.legend(SPEAKERS)
plt.savefig("fixed_encoding_pca_vctk.png")
plt.show()
fig, ax = plt.subplots()
y = tsne.fit_transform(np.array(embed_list))
x_pl = []
y_pl = []
for speaker in range(len(SPEAKERS)):
x_pl.append([i for (i, s) in zip(y[:, 0], speaker_list) if s==speaker])
y_pl.append([i for (i, s) in zip(y[:, 1], speaker_list) if s==speaker])
ax.scatter(x_pl[speaker], y_pl[speaker])
#plt.legend(SPEAKERS)
plt.savefig("fixed_encoding_tsne_vctk.png")
plt.show()
fig, ax = plt.subplots()
for speaker in range(len(SPEAKERS)):
if gender_dict[speaker][0] == "F":
marker = "."
else:
marker = "x"
if gender_dict[speaker][1] == "American":
c = "blue"
elif gender_dict[speaker][1] == "English":
c = "red"
elif gender_dict[speaker][1] == "Canadian":
c = "purple"
elif gender_dict[speaker][1] == "Indian":
c = "yellow"
elif gender_dict[speaker][1] == "Scottish":
c = "green"
else:
c = "black"
ax.scatter(np.mean(x_pl[speaker]), np.mean(y_pl[speaker]), marker=marker, c=c)
#plt.legend(SPEAKERS)
plt.savefig("fixed_encoding_tsne_vctk.png")
plt.show()