Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,21 +1,94 @@
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import re
|
3 |
+
import random
|
4 |
+
from scipy.io.wavfile import write, read
|
5 |
+
import numpy as np
|
6 |
import gradio as gr
|
7 |
+
import yt_dlp
|
8 |
+
|
9 |
+
# Model dictionaries and lists
|
10 |
+
roformer_models = {
|
11 |
+
'BS-Roformer-Viperx-1297.ckpt': 'model_bs_roformer_ep_317_sdr_12.9755.ckpt',
|
12 |
+
'BS-Roformer-Viperx-1296.ckpt': 'model_bs_roformer_ep_368_sdr_12.9628.ckpt',
|
13 |
+
'BS-Roformer-Viperx-1053.ckpt': 'model_bs_roformer_ep_937_sdr_10.5309.ckpt',
|
14 |
+
'Mel-Roformer-Viperx-1143.ckpt': 'model_mel_band_roformer_ep_3005_sdr_11.4360.ckpt'
|
15 |
+
}
|
16 |
+
|
17 |
+
mdx23c_models = [
|
18 |
+
'MDX23C_D1581.ckpt',
|
19 |
+
'MDX23C-8KFFT-InstVoc_HQ.ckpt',
|
20 |
+
'MDX23C-8KFFT-InstVoc_HQ_2.ckpt',
|
21 |
+
]
|
22 |
+
|
23 |
+
# More model lists...
|
24 |
+
|
25 |
+
output_format = ['wav', 'flac', 'mp3']
|
26 |
+
mdxnet_overlap_values = ['0.25', '0.5', '0.75', '0.99']
|
27 |
+
vrarch_window_size_values = ['320', '512', '1024']
|
28 |
+
demucs_overlap_values = ['0.25', '0.50', '0.75', '0.99']
|
29 |
+
|
30 |
+
# Function to download audio
|
31 |
+
def download_audio(url):
|
32 |
+
ydl_opts = {
|
33 |
+
'format': 'bestaudio/best',
|
34 |
+
'outtmpl': 'ytdl/%(title)s.%(ext)s',
|
35 |
+
'postprocessors': [{
|
36 |
+
'key': 'FFmpegExtractAudio',
|
37 |
+
'preferredcodec': 'wav',
|
38 |
+
'preferredquality': '192',
|
39 |
+
}],
|
40 |
+
}
|
41 |
+
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
|
42 |
+
info_dict = ydl.extract_info(url, download=True)
|
43 |
+
file_path = ydl.prepare_filename(info_dict).rsplit('.', 1)[0] + '.wav'
|
44 |
+
sample_rate, audio_data = read(file_path)
|
45 |
+
audio_array = np.asarray(audio_data, dtype=np.int16)
|
46 |
+
return sample_rate, audio_array
|
47 |
+
|
48 |
+
# Function to separate audio using Roformer
|
49 |
+
def roformer_separator(audio, model, output_format, overlap, segment_size, denoise):
|
50 |
+
directory = "./outputs"
|
51 |
+
random_id = str(random.randint(10000, 99999))
|
52 |
+
os.makedirs("outputs", exist_ok=True)
|
53 |
+
write(f'{random_id}.wav', audio[0], audio[1])
|
54 |
+
full_roformer_model = roformer_models[model]
|
55 |
+
|
56 |
+
prompt = f"audio-separator {random_id}.wav --model_filename {full_roformer_model} --output_dir=./outputs --output_format={output_format} --normalization=0.9 --mdxc_overlap={overlap} --mdxc_segment_size={segment_size}"
|
57 |
+
if denoise:
|
58 |
+
prompt += " --mdx_enable_denoise"
|
59 |
+
|
60 |
+
os.system(prompt)
|
61 |
+
|
62 |
+
files_list = [os.path.join(directory, file) for file in os.listdir(directory) if re.search(random_id, file)]
|
63 |
+
stem1_file, stem2_file, stem3_file = files_list[:3] # Assuming the files are in the correct order
|
64 |
+
return stem1_file, stem2_file, stem3_file
|
65 |
+
|
66 |
+
# Gradio interface
|
67 |
+
def process_audio(url, model, output_format, overlap, segment_size, denoise):
|
68 |
+
sample_rate, audio_array = download_audio(url)
|
69 |
+
stems = roformer_separator((sample_rate, audio_array), model, output_format, overlap, segment_size, denoise)
|
70 |
+
return stems
|
71 |
+
|
72 |
+
# Gradio UI
|
73 |
+
with gr.Blocks() as demo:
|
74 |
+
gr.Markdown("# Hex Audio Separator")
|
75 |
+
with gr.Row():
|
76 |
+
url_input = gr.Textbox(label="YouTube URL")
|
77 |
+
model_input = gr.Dropdown(choices=list(roformer_models.keys()), label="Roformer Model")
|
78 |
+
format_input = gr.Dropdown(choices=output_format, label="Output Format")
|
79 |
+
overlap_input = gr.Dropdown(choices=mdxnet_overlap_values, label="Overlap")
|
80 |
+
segment_input = gr.Slider(0, 100, label="Segment Size")
|
81 |
+
denoise_input = gr.Checkbox(label="Enable Denoise")
|
82 |
+
|
83 |
+
output1 = gr.Audio(label="Vocals")
|
84 |
+
output2 = gr.Audio(label="Instrumental")
|
85 |
+
output3 = gr.Audio(label="Backing Vocals")
|
86 |
+
|
87 |
+
submit_button = gr.Button("Process")
|
88 |
+
submit_button.click(
|
89 |
+
process_audio,
|
90 |
+
inputs=[url_input, model_input, format_input, overlap_input, segment_input, denoise_input],
|
91 |
+
outputs=[output1, output2, output3]
|
92 |
+
)
|
93 |
+
|
94 |
+
demo.launch()
|