-
Notifications
You must be signed in to change notification settings - Fork 0
/
update_embeddings.py
179 lines (146 loc) · 6.49 KB
/
update_embeddings.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
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
from encoder import inference as encoder
from pathlib import Path
import os
import pickle
from tqdm import tqdm
from scipy.io.wavfile import read
import tensorflow_hub as hub
import numpy as np
import sys
import tensorflow.compat.v2 as tf
tf.enable_v2_behavior()
assert tf.executing_eagerly()
EMBEDDING_TYPE = "trill" # "ge2e" for GE2E, "trill" for trill
def create_embedding(path):
if EMBEDDING_TYPE == "ge2e":
in_fpath = Path(path.replace("\"", "").replace("\'", ""))
preprocessed_wav = encoder.preprocess_wav(in_fpath)
return encoder.embed_utterance(preprocessed_wav)
else:
# in_fpath = Path(path.replace("\"", "").replace("\'", ""))
# preprocessed_wav = encoder.preprocess_wav(in_fpath)
samplerate, data = read(path.replace("\"", "").replace("\'", ""))
if samplerate != 16000:
sys.exit(f"{path} does not have sample rate of 16000")
emb = module(samples=np.array(data, dtype=float), sample_rate=16000)['embedding']
emb.shape.assert_is_compatible_with([None, 2048])
return emb.numpy()
if EMBEDDING_TYPE not in {"ge2e", "trill"}:
sys.exit("Embedding type has to be 'ge2e' or 'trill'.")
encoder.load_model(Path("encoder/saved_models/pretrained.pt"))
if EMBEDDING_TYPE == "trill":
module = hub.load('https://tfhub.dev/google/nonsemantic-speech-benchmark/trill-distilled/3')
dir_to_update = "wavs"
wavs_total = 0
for root, dirs, files in os.walk(dir_to_update):
wavs_total += len(files)
print()
print(f"Updating {EMBEDDING_TYPE} type embeddings in directory embeddings-{EMBEDDING_TYPE}.")
print("🔊 There are {0} .wav files inside {1}.".format(wavs_total, dir_to_update))
print("Starting update ...")
print()
embeddings_not_modified = 0
embeddings_deleted = 0
embeddings_created = 0
# Remove outdated embeddings with no matching source in wavs
for root, dirs, files in os.walk(f"embeddings-{EMBEDDING_TYPE}", topdown=False):
for f in files:
embedding = os.path.join(root, f)
wav = embedding.replace(f"embeddings-{EMBEDDING_TYPE}/", "wavs/").replace(".pickle", ".wav")
if not os.path.exists(wav):
os.remove(embedding)
for d in dirs:
embedding_dir = os.path.join(root, d)
wav_dir = embedding_dir.replace(f"embeddings-{EMBEDDING_TYPE}/", "wavs/")
if not os.path.exists(wav_dir):
os.rmdir(embedding_dir)
# Then create new embeddings for all .wav files that don't already have one
progressbar = tqdm(total=wavs_total)
for root, dirs, files in os.walk(dir_to_update):
for d in dirs:
directory = os.path.join(root, d).replace("wavs/", f"embeddings-{EMBEDDING_TYPE}/")
if not os.path.isdir(directory):
os.makedirs(directory)
for f in files:
wav_path = os.path.join(root, f)
if wav_path.endswith(".wav"):
embedding_path = wav_path.replace("wavs/", f"embeddings-{EMBEDDING_TYPE}/").replace(".wav", ".pickle")
if os.path.isfile(embedding_path):
embeddings_not_modified += 1
else:
embedding = create_embedding(wav_path)
pickle.dump(embedding, open(embedding_path, "wb"))
embeddings_created += 1
progressbar.update(1)
progressbar.close()
print()
print("✔️ Done")
print("{0} embeddings not modified, {1} deleted, {2} created"
.format(embeddings_not_modified, embeddings_deleted, embeddings_created))
EMBEDDING_TYPE = "ge2e" # "ge2e" for GE2E, "trill" for trill
def create_embedding(path):
if EMBEDDING_TYPE == "ge2e":
in_fpath = Path(path.replace("\"", "").replace("\'", ""))
preprocessed_wav = encoder.preprocess_wav(in_fpath)
return encoder.embed_utterance(preprocessed_wav)
else:
# in_fpath = Path(path.replace("\"", "").replace("\'", ""))
# preprocessed_wav = encoder.preprocess_wav(in_fpath)
samplerate, data = read(path.replace("\"", "").replace("\'", ""))
if samplerate != 16000:
sys.exit(f"{path} does not have sample rate of 16000")
emb = module(samples=np.array(data, dtype=float), sample_rate=16000)['embedding']
emb.shape.assert_is_compatible_with([None, 2048])
return emb.numpy()
if EMBEDDING_TYPE not in {"ge2e", "trill"}:
sys.exit("Embedding type has to be 'ge2e' or 'trill'.")
encoder.load_model(Path("encoder/saved_models/pretrained.pt"))
if EMBEDDING_TYPE == "trill":
module = hub.load('https://tfhub.dev/google/nonsemantic-speech-benchmark/trill-distilled/3')
dir_to_update = "wavs"
wavs_total = 0
for root, dirs, files in os.walk(dir_to_update):
wavs_total += len(files)
print()
print(f"Updating {EMBEDDING_TYPE} type embeddings in directory embeddings-{EMBEDDING_TYPE}.")
print("🔊 There are {0} .wav files inside {1}.".format(wavs_total, dir_to_update))
print("Starting update ...")
print()
embeddings_not_modified = 0
embeddings_deleted = 0
embeddings_created = 0
# Remove outdated embeddings with no matching source in wavs
for root, dirs, files in os.walk(f"embeddings-{EMBEDDING_TYPE}", topdown=False):
for f in files:
embedding = os.path.join(root, f)
wav = embedding.replace(f"embeddings-{EMBEDDING_TYPE}/", "wavs/").replace(".pickle", ".wav")
if not os.path.exists(wav):
os.remove(embedding)
for d in dirs:
embedding_dir = os.path.join(root, d)
wav_dir = embedding_dir.replace(f"embeddings-{EMBEDDING_TYPE}/", "wavs/")
if not os.path.exists(wav_dir):
os.rmdir(embedding_dir)
# Then create new embeddings for all .wav files that don't already have one
progressbar = tqdm(total=wavs_total)
for root, dirs, files in os.walk(dir_to_update):
for d in dirs:
directory = os.path.join(root, d).replace("wavs/", f"embeddings-{EMBEDDING_TYPE}/")
if not os.path.isdir(directory):
os.makedirs(directory)
for f in files:
wav_path = os.path.join(root, f)
if wav_path.endswith(".wav"):
embedding_path = wav_path.replace("wavs/", f"embeddings-{EMBEDDING_TYPE}/").replace(".wav", ".pickle")
if os.path.isfile(embedding_path):
embeddings_not_modified += 1
else:
embedding = create_embedding(wav_path)
pickle.dump(embedding, open(embedding_path, "wb"))
embeddings_created += 1
progressbar.update(1)
progressbar.close()
print()
print("✔️ Done")
print("{0} embeddings not modified, {1} deleted, {2} created"
.format(embeddings_not_modified, embeddings_deleted, embeddings_created))