Spaces:
Runtime error
Runtime error
Upload app.py
Browse files
app.py
ADDED
@@ -0,0 +1,77 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import io
|
2 |
+
import os
|
3 |
+
|
4 |
+
#os.system("wget -P hubert/ https://huggingface.co/spaces/innnky/nanami/resolve/main/checkpoint_best_legacy_500.pt")
|
5 |
+
import gradio as gr
|
6 |
+
import librosa
|
7 |
+
import numpy as np
|
8 |
+
import soundfile
|
9 |
+
from inference.infer_tool import Svc
|
10 |
+
import logging
|
11 |
+
|
12 |
+
logging.getLogger('numba').setLevel(logging.WARNING)
|
13 |
+
logging.getLogger('markdown_it').setLevel(logging.WARNING)
|
14 |
+
logging.getLogger('urllib3').setLevel(logging.WARNING)
|
15 |
+
logging.getLogger('matplotlib').setLevel(logging.WARNING)
|
16 |
+
|
17 |
+
model = Svc("logs/44k/06_21_27_28/G_61000.pth", "logs/44k/06_21_27_28/config.json", cluster_model_path="logs/44k/06_21_27_28/kmeans_10000.pt")
|
18 |
+
|
19 |
+
|
20 |
+
|
21 |
+
def vc_fn(sid, input_audio, vc_transform, auto_f0,cluster_ratio, noise_scale):
|
22 |
+
if input_audio is None:
|
23 |
+
return "You need to upload an audio", None
|
24 |
+
sampling_rate, audio = input_audio
|
25 |
+
# print(audio.shape,sampling_rate)
|
26 |
+
duration = audio.shape[0] / sampling_rate
|
27 |
+
if duration > 45:
|
28 |
+
return "请上传小于45s的音频,需要转换长音频请本地进行转换", None
|
29 |
+
audio = (audio / np.iinfo(audio.dtype).max).astype(np.float32)
|
30 |
+
if len(audio.shape) > 1:
|
31 |
+
audio = librosa.to_mono(audio.transpose(1, 0))
|
32 |
+
if sampling_rate != 16000:
|
33 |
+
audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=16000)
|
34 |
+
print(audio.shape)
|
35 |
+
out_wav_path = "temp.wav"
|
36 |
+
soundfile.write(out_wav_path, audio, 16000, format="wav")
|
37 |
+
print( cluster_ratio, auto_f0, noise_scale)
|
38 |
+
out_audio, out_sr = model.infer(sid, vc_transform, out_wav_path,
|
39 |
+
cluster_infer_ratio=cluster_ratio,
|
40 |
+
auto_predict_f0=auto_f0,
|
41 |
+
noice_scale=noise_scale
|
42 |
+
)
|
43 |
+
return "Success", (44100, out_audio.cpu().numpy())
|
44 |
+
|
45 |
+
|
46 |
+
app = gr.Blocks()
|
47 |
+
with app:
|
48 |
+
with gr.Tabs():
|
49 |
+
with gr.TabItem("Basic"):
|
50 |
+
gr.Markdown(value="""
|
51 |
+
datealive角色语音合成 so-vits-svc-4.0 在线试用
|
52 |
+
使用凛绪轮回中的语音数据训练而成。
|
53 |
+
目前可用角色为凛祢、凛绪、鞠亚、鞠奈。
|
54 |
+
其它角色的模型已训练完成,包含游戏中出现的大部分角色。
|
55 |
+
可以在models中下载其它角色的模型以试用,文件夹名称对应角色编号。
|
56 |
+
""")
|
57 |
+
spks = list(model.spk2id.keys())
|
58 |
+
sid = gr.Dropdown(label="音色", choices=spks, value=spks[0])
|
59 |
+
vc_input3 = gr.Audio(label="上传音频(长度小于45秒)")
|
60 |
+
vc_transform = gr.Number(label="变调(整数,可以正负,半音数量,升高八度就是12),使用变调请务必同时将伴奏变调相同半音数!", value=0)
|
61 |
+
cluster_ratio = gr.Number(label="聚类模型混合比例,0-1之间,默认为0不启用聚类,能提升音色相似度,但会导致咬字下降(如果使用建议0.5左右)", value=0)
|
62 |
+
auto_f0 = gr.Checkbox(label="自动f0预测,配合聚类模型f0预测效果更好,会导致变调功能失效(仅限转换语音,歌声不要勾选此项会究极跑调)", value=False)
|
63 |
+
noise_scale = gr.Number(label="noise_scale 建议不要动,会影响音质,玄学参数", value=0.4)
|
64 |
+
vc_submit = gr.Button("转换", variant="primary")
|
65 |
+
vc_output1 = gr.Textbox(label="Output Message")
|
66 |
+
vc_output2 = gr.Audio(label="Output Audio")
|
67 |
+
vc_submit.click(vc_fn, [sid, vc_input3, vc_transform,auto_f0,cluster_ratio, noise_scale], [vc_output1, vc_output2])
|
68 |
+
|
69 |
+
gr.HTML("""
|
70 |
+
<div style="text-align:center">
|
71 |
+
仅供学习交流试用,不可用于商业或非法用途。
|
72 |
+
<br/>
|
73 |
+
由此项目产出的任何结果公开发表需指明输入源,注明原作者及<a href="https://github.com/svc-develop-team/so-vits-svc" target="_blank">代码来源</a>。
|
74 |
+
</div>
|
75 |
+
""")
|
76 |
+
|
77 |
+
app.launch()
|