Spaces:
Running
Running
Update rvc/lib/utils.py
Browse files- rvc/lib/utils.py +66 -64
rvc/lib/utils.py
CHANGED
@@ -1,64 +1,66 @@
|
|
1 |
-
import os, sys
|
2 |
-
import ffmpeg
|
3 |
-
import numpy as np
|
4 |
-
import re
|
5 |
-
import unicodedata
|
6 |
-
from fairseq import checkpoint_utils
|
7 |
-
|
8 |
-
import logging
|
9 |
-
|
10 |
-
logging.getLogger("fairseq").setLevel(logging.WARNING)
|
11 |
-
|
12 |
-
now_dir = os.getcwd()
|
13 |
-
sys.path.append(now_dir)
|
14 |
-
|
15 |
-
|
16 |
-
def load_audio(file, sampling_rate):
|
17 |
-
try:
|
18 |
-
file = file.strip(" ").strip('"').strip("\n").strip('"').strip(" ")
|
19 |
-
out, _ = (
|
20 |
-
ffmpeg.input(file, threads=0)
|
21 |
-
.output("-", format="f32le", acodec="pcm_f32le", ac=1, ar=sampling_rate)
|
22 |
-
.run(cmd=["ffmpeg", "-nostdin"], capture_stdout=True, capture_stderr=True)
|
23 |
-
)
|
24 |
-
except Exception as error:
|
25 |
-
raise RuntimeError(f"Failed to load audio: {error}")
|
26 |
-
|
27 |
-
return np.frombuffer(out, np.float32).flatten()
|
28 |
-
|
29 |
-
|
30 |
-
def format_title(title):
|
31 |
-
formatted_title = (
|
32 |
-
unicodedata.normalize("NFKD", title).encode("ascii", "ignore").decode("utf-8")
|
33 |
-
)
|
34 |
-
formatted_title = re.sub(r"[\u2500-\u257F]+", "", formatted_title)
|
35 |
-
formatted_title = re.sub(r"[^\w\s.-]", "", formatted_title)
|
36 |
-
formatted_title = re.sub(r"\s+", "_", formatted_title)
|
37 |
-
return formatted_title
|
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 |
-
|
|
|
|
|
|
1 |
+
import os, sys
|
2 |
+
import ffmpeg
|
3 |
+
import numpy as np
|
4 |
+
import re
|
5 |
+
import unicodedata
|
6 |
+
from fairseq import checkpoint_utils
|
7 |
+
|
8 |
+
import logging
|
9 |
+
|
10 |
+
logging.getLogger("fairseq").setLevel(logging.WARNING)
|
11 |
+
|
12 |
+
now_dir = os.getcwd()
|
13 |
+
sys.path.append(now_dir)
|
14 |
+
|
15 |
+
|
16 |
+
def load_audio(file, sampling_rate):
|
17 |
+
try:
|
18 |
+
file = file.strip(" ").strip('"').strip("\n").strip('"').strip(" ")
|
19 |
+
out, _ = (
|
20 |
+
ffmpeg.input(file, threads=0)
|
21 |
+
.output("-", format="f32le", acodec="pcm_f32le", ac=1, ar=sampling_rate)
|
22 |
+
.run(cmd=["ffmpeg", "-nostdin"], capture_stdout=True, capture_stderr=True)
|
23 |
+
)
|
24 |
+
except Exception as error:
|
25 |
+
raise RuntimeError(f"Failed to load audio: {error}")
|
26 |
+
|
27 |
+
return np.frombuffer(out, np.float32).flatten()
|
28 |
+
|
29 |
+
|
30 |
+
def format_title(title):
|
31 |
+
formatted_title = (
|
32 |
+
unicodedata.normalize("NFKD", title).encode("ascii", "ignore").decode("utf-8")
|
33 |
+
)
|
34 |
+
formatted_title = re.sub(r"[\u2500-\u257F]+", "", formatted_title)
|
35 |
+
formatted_title = re.sub(r"[^\w\s.-]", "", formatted_title)
|
36 |
+
formatted_title = re.sub(r"\s+", "_", formatted_title)
|
37 |
+
return formatted_title
|
38 |
+
|
39 |
+
import spaces
|
40 |
+
|
41 |
+
@spaces.GPU
|
42 |
+
def load_embedding(embedder_model, custom_embedder=None):
|
43 |
+
embedder_root = os.path.join(now_dir, "rvc", "embedders")
|
44 |
+
embedding_list = {
|
45 |
+
"contentvec": os.path.join(embedder_root, "contentvec_base.pt"),
|
46 |
+
"hubert": os.path.join(embedder_root, "hubert_base.pt"),
|
47 |
+
}
|
48 |
+
|
49 |
+
if embedder_model == "custom":
|
50 |
+
model_path = custom_embedder
|
51 |
+
if not custom_embedder and os.path.exists(custom_embedder):
|
52 |
+
print("Custom embedder not found. Using the default embedder.")
|
53 |
+
model_path = embedding_list["hubert"]
|
54 |
+
else:
|
55 |
+
model_path = embedding_list[embedder_model]
|
56 |
+
if not os.path.exists(model_path):
|
57 |
+
print("Custom embedder not found. Using the default embedder.")
|
58 |
+
model_path = embedding_list["hubert"]
|
59 |
+
|
60 |
+
models = checkpoint_utils.load_model_ensemble_and_task(
|
61 |
+
[model_path],
|
62 |
+
suffix="",
|
63 |
+
)
|
64 |
+
|
65 |
+
print(f"Embedding model {embedder_model} loaded successfully.")
|
66 |
+
return models
|