Spaces:
Running
on
Zero
Running
on
Zero
artificialguybr
commited on
Commit
•
8d0320b
1
Parent(s):
12ce031
Update app.py
Browse files
app.py
CHANGED
@@ -1,35 +1,28 @@
|
|
1 |
-
import
|
2 |
-
import tempfile
|
3 |
-
import gradio as gr
|
4 |
-
import subprocess
|
5 |
-
import os, stat
|
6 |
import uuid
|
7 |
-
from googletrans import Translator
|
8 |
-
import edge_tts
|
9 |
import asyncio
|
10 |
-
import
|
11 |
import json
|
12 |
-
from scipy.signal import wiener
|
13 |
-
import soundfile as sf
|
14 |
-
from pydub import AudioSegment
|
15 |
-
import numpy as np
|
16 |
-
import librosa
|
17 |
from zipfile import ZipFile
|
18 |
-
import
|
|
|
19 |
import cv2
|
20 |
-
import
|
21 |
-
import
|
22 |
-
from tqdm import tqdm
|
23 |
-
from numba import jit
|
24 |
from huggingface_hub import HfApi
|
25 |
import moviepy.editor as mp
|
|
|
26 |
|
|
|
27 |
HF_TOKEN = os.environ.get("HF_TOKEN")
|
|
|
|
|
|
|
28 |
api = HfApi(token=HF_TOKEN)
|
29 |
-
|
|
|
30 |
ZipFile("ffmpeg.zip").extractall()
|
31 |
-
|
32 |
-
os.chmod('ffmpeg', st.st_mode | stat.S_IEXEC)
|
33 |
|
34 |
print("Starting the program...")
|
35 |
|
@@ -50,19 +43,16 @@ def check_for_faces(video_path):
|
|
50 |
ret, frame = cap.read()
|
51 |
if not ret:
|
52 |
break
|
53 |
-
|
54 |
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
|
55 |
-
|
56 |
-
|
57 |
-
if len(faces) > 0:
|
58 |
return True
|
59 |
-
|
60 |
return False
|
61 |
|
62 |
@spaces.GPU(duration=90)
|
63 |
def transcribe_audio(file_path):
|
64 |
print(f"Starting transcription of file: {file_path}")
|
65 |
temp_audio = None
|
|
|
66 |
if file_path.endswith(('.mp4', '.avi', '.mov', '.flv')):
|
67 |
print("Video file detected. Extracting audio...")
|
68 |
try:
|
@@ -73,10 +63,7 @@ def transcribe_audio(file_path):
|
|
73 |
except Exception as e:
|
74 |
print(f"Error extracting audio from video: {e}")
|
75 |
raise
|
76 |
-
|
77 |
-
print(f"Does the file exist? {os.path.exists(file_path)}")
|
78 |
-
print(f"File size: {os.path.getsize(file_path) if os.path.exists(file_path) else 'N/A'} bytes")
|
79 |
-
|
80 |
output_file = generate_unique_filename(".json")
|
81 |
command = [
|
82 |
"insanely-fast-whisper",
|
@@ -87,37 +74,24 @@ def transcribe_audio(file_path):
|
|
87 |
"--timestamp", "chunk",
|
88 |
"--transcript-path", output_file
|
89 |
]
|
90 |
-
|
91 |
try:
|
92 |
result = subprocess.run(command, check=True, capture_output=True, text=True)
|
93 |
-
print(f"
|
94 |
-
print(f"Error output: {result.stderr}")
|
95 |
except subprocess.CalledProcessError as e:
|
96 |
print(f"Error running insanely-fast-whisper: {e}")
|
97 |
-
print(f"Standard output: {e.stdout}")
|
98 |
-
print(f"Error output: {e.stderr}")
|
99 |
raise
|
100 |
-
|
101 |
-
print(f"Reading transcription file: {output_file}")
|
102 |
try:
|
103 |
with open(output_file, "r") as f:
|
104 |
transcription = json.load(f)
|
105 |
except json.JSONDecodeError as e:
|
106 |
print(f"Error decoding JSON: {e}")
|
107 |
-
print(f"File content: {open(output_file, 'r').read()}")
|
108 |
raise
|
|
|
|
|
109 |
|
110 |
-
|
111 |
-
result = transcription["text"]
|
112 |
-
else:
|
113 |
-
result = " ".join([chunk["text"] for chunk in transcription.get("chunks", [])])
|
114 |
-
|
115 |
-
print("Transcription completed.")
|
116 |
-
|
117 |
-
# Cleanup
|
118 |
-
cleanup_files(output_file)
|
119 |
-
if temp_audio:
|
120 |
-
cleanup_files(temp_audio)
|
121 |
|
122 |
return result
|
123 |
|
@@ -143,22 +117,16 @@ def process_video(radio, video, target_language, has_closeup_face):
|
|
143 |
video_info = ffmpeg.probe(video_path)
|
144 |
video_duration = float(video_info['streams'][0]['duration'])
|
145 |
|
146 |
-
if video_duration >
|
147 |
-
|
148 |
-
raise ValueError("Video duration exceeds
|
149 |
|
150 |
ffmpeg.input(video_path).output(f"{run_uuid}_output_audio.wav", acodec='pcm_s24le', ar=48000, map='a').run()
|
151 |
|
152 |
-
|
153 |
-
subprocess.run([item for item in shell_command], capture_output=False, text=True, check=True)
|
154 |
|
155 |
-
|
156 |
-
|
157 |
-
whisper_text = transcribe_audio(f"{run_uuid}_output_audio_final.wav")
|
158 |
-
print(f"Transcription successful: {whisper_text}")
|
159 |
-
except Exception as e:
|
160 |
-
print(f"Error encountered during transcription: {str(e)}")
|
161 |
-
raise
|
162 |
|
163 |
language_mapping = {
|
164 |
'English': ('en', 'en-US-EricNeural'),
|
@@ -189,61 +157,36 @@ def process_video(radio, video, target_language, has_closeup_face):
|
|
189 |
|
190 |
asyncio.run(text_to_speech(translated_text, voice, f"{run_uuid}_output_synth.wav"))
|
191 |
|
192 |
-
|
193 |
-
pad_bottom = 15
|
194 |
-
pad_left = 0
|
195 |
-
pad_right = 0
|
196 |
-
rescaleFactor = 1
|
197 |
-
|
198 |
-
video_path_fix = video_path
|
199 |
-
|
200 |
-
if has_closeup_face:
|
201 |
-
has_face = True
|
202 |
-
else:
|
203 |
-
has_face = check_for_faces(video_path)
|
204 |
-
|
205 |
-
if has_closeup_face:
|
206 |
try:
|
207 |
-
|
208 |
-
|
209 |
-
|
210 |
-
|
211 |
-
gr.Warning("Wav2lip didn't detect a face. Please try again with the option disabled.")
|
212 |
-
cmd = f"ffmpeg -i {video_path} -i {run_uuid}_output_synth.wav -c:v copy -c:a aac -strict experimental -map 0:v:0 -map 1:a:0 {run_uuid}_output_video.mp4"
|
213 |
-
subprocess.run(cmd, shell=True)
|
214 |
else:
|
215 |
-
|
216 |
-
subprocess.run(cmd, shell=True)
|
217 |
-
|
218 |
-
if not os.path.exists(f"{run_uuid}_output_video.mp4"):
|
219 |
-
raise FileNotFoundError(f"Error: {run_uuid}_output_video.mp4 was not generated.")
|
220 |
|
221 |
output_video_path = f"{run_uuid}_output_video.mp4"
|
|
|
|
|
222 |
|
223 |
-
|
224 |
f"{run_uuid}_resized_video.mp4",
|
225 |
f"{run_uuid}_output_audio.wav",
|
226 |
f"{run_uuid}_output_audio_final.wav",
|
227 |
f"{run_uuid}_output_synth.wav"
|
228 |
-
|
229 |
-
for file in files_to_delete:
|
230 |
-
try:
|
231 |
-
os.remove(file)
|
232 |
-
except FileNotFoundError:
|
233 |
-
print(f"File {file} not found for deletion.")
|
234 |
|
235 |
-
return output_video_path, ""
|
236 |
|
237 |
except Exception as e:
|
238 |
print(f"Error in process_video: {str(e)}")
|
239 |
-
return None, f"Error: {str(e)}"
|
240 |
|
241 |
def swap(radio):
|
242 |
-
if
|
243 |
-
|
244 |
-
|
245 |
-
return gr.update(source="webcam")
|
246 |
-
|
247 |
video = gr.Video()
|
248 |
radio = gr.Radio(["Upload", "Record"], value="Upload", show_label=False)
|
249 |
iface = gr.Interface(
|
@@ -251,11 +194,8 @@ iface = gr.Interface(
|
|
251 |
inputs=[
|
252 |
radio,
|
253 |
video,
|
254 |
-
gr.Dropdown(choices=
|
255 |
-
gr.Checkbox(
|
256 |
-
label="Video has a close-up face. Use Wav2lip.",
|
257 |
-
value=False,
|
258 |
-
info="Say if video have close-up face. For Wav2lip. Will not work if checked wrongly.")
|
259 |
],
|
260 |
outputs=[
|
261 |
gr.Video(label="Processed Video"),
|
|
|
1 |
+
import os
|
|
|
|
|
|
|
|
|
2 |
import uuid
|
|
|
|
|
3 |
import asyncio
|
4 |
+
import subprocess
|
5 |
import json
|
|
|
|
|
|
|
|
|
|
|
6 |
from zipfile import ZipFile
|
7 |
+
import gradio as gr
|
8 |
+
import ffmpeg
|
9 |
import cv2
|
10 |
+
import edge_tts
|
11 |
+
from googletrans import Translator
|
|
|
|
|
12 |
from huggingface_hub import HfApi
|
13 |
import moviepy.editor as mp
|
14 |
+
import spaces
|
15 |
|
16 |
+
# Constants and initialization
|
17 |
HF_TOKEN = os.environ.get("HF_TOKEN")
|
18 |
+
REPO_ID = "artificialguybr/video-dubbing"
|
19 |
+
MAX_VIDEO_DURATION = 60 # seconds
|
20 |
+
|
21 |
api = HfApi(token=HF_TOKEN)
|
22 |
+
|
23 |
+
# Extract and set permissions for ffmpeg
|
24 |
ZipFile("ffmpeg.zip").extractall()
|
25 |
+
os.chmod('ffmpeg', os.stat('ffmpeg').st_mode | os.stat.S_IEXEC)
|
|
|
26 |
|
27 |
print("Starting the program...")
|
28 |
|
|
|
43 |
ret, frame = cap.read()
|
44 |
if not ret:
|
45 |
break
|
|
|
46 |
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
|
47 |
+
if face_cascade.detectMultiScale(gray, 1.1, 4):
|
|
|
|
|
48 |
return True
|
|
|
49 |
return False
|
50 |
|
51 |
@spaces.GPU(duration=90)
|
52 |
def transcribe_audio(file_path):
|
53 |
print(f"Starting transcription of file: {file_path}")
|
54 |
temp_audio = None
|
55 |
+
|
56 |
if file_path.endswith(('.mp4', '.avi', '.mov', '.flv')):
|
57 |
print("Video file detected. Extracting audio...")
|
58 |
try:
|
|
|
63 |
except Exception as e:
|
64 |
print(f"Error extracting audio from video: {e}")
|
65 |
raise
|
66 |
+
|
|
|
|
|
|
|
67 |
output_file = generate_unique_filename(".json")
|
68 |
command = [
|
69 |
"insanely-fast-whisper",
|
|
|
74 |
"--timestamp", "chunk",
|
75 |
"--transcript-path", output_file
|
76 |
]
|
77 |
+
|
78 |
try:
|
79 |
result = subprocess.run(command, check=True, capture_output=True, text=True)
|
80 |
+
print(f"Transcription output: {result.stdout}")
|
|
|
81 |
except subprocess.CalledProcessError as e:
|
82 |
print(f"Error running insanely-fast-whisper: {e}")
|
|
|
|
|
83 |
raise
|
84 |
+
|
|
|
85 |
try:
|
86 |
with open(output_file, "r") as f:
|
87 |
transcription = json.load(f)
|
88 |
except json.JSONDecodeError as e:
|
89 |
print(f"Error decoding JSON: {e}")
|
|
|
90 |
raise
|
91 |
+
|
92 |
+
result = transcription.get("text", " ".join([chunk["text"] for chunk in transcription.get("chunks", [])]))
|
93 |
|
94 |
+
cleanup_files(output_file, temp_audio)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
95 |
|
96 |
return result
|
97 |
|
|
|
117 |
video_info = ffmpeg.probe(video_path)
|
118 |
video_duration = float(video_info['streams'][0]['duration'])
|
119 |
|
120 |
+
if video_duration > MAX_VIDEO_DURATION:
|
121 |
+
cleanup_files(video_path)
|
122 |
+
raise ValueError(f"Video duration exceeds {MAX_VIDEO_DURATION} seconds. Please upload a shorter video.")
|
123 |
|
124 |
ffmpeg.input(video_path).output(f"{run_uuid}_output_audio.wav", acodec='pcm_s24le', ar=48000, map='a').run()
|
125 |
|
126 |
+
subprocess.run(f"ffmpeg -y -i {run_uuid}_output_audio.wav -af lowpass=3000,highpass=100 {run_uuid}_output_audio_final.wav", shell=True, check=True)
|
|
|
127 |
|
128 |
+
whisper_text = transcribe_audio(f"{run_uuid}_output_audio_final.wav")
|
129 |
+
print(f"Transcription successful: {whisper_text}")
|
|
|
|
|
|
|
|
|
|
|
130 |
|
131 |
language_mapping = {
|
132 |
'English': ('en', 'en-US-EricNeural'),
|
|
|
157 |
|
158 |
asyncio.run(text_to_speech(translated_text, voice, f"{run_uuid}_output_synth.wav"))
|
159 |
|
160 |
+
if has_closeup_face or check_for_faces(video_path):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
161 |
try:
|
162 |
+
subprocess.run(f"python Wav2Lip/inference.py --checkpoint_path 'Wav2Lip/checkpoints/wav2lip_gan.pth' --face '{video_path}' --audio '{run_uuid}_output_synth.wav' --pads 0 15 0 0 --resize_factor 1 --nosmooth --outfile '{run_uuid}_output_video.mp4'", shell=True, check=True)
|
163 |
+
except subprocess.CalledProcessError:
|
164 |
+
gr.Warning("Wav2lip didn't detect a face. Please try again with the option disabled.")
|
165 |
+
subprocess.run(f"ffmpeg -i {video_path} -i {run_uuid}_output_synth.wav -c:v copy -c:a aac -strict experimental -map 0:v:0 -map 1:a:0 {run_uuid}_output_video.mp4", shell=True)
|
|
|
|
|
|
|
166 |
else:
|
167 |
+
subprocess.run(f"ffmpeg -i {video_path} -i {run_uuid}_output_synth.wav -c:v copy -c:a aac -strict experimental -map 0:v:0 -map 1:a:0 {run_uuid}_output_video.mp4", shell=True)
|
|
|
|
|
|
|
|
|
168 |
|
169 |
output_video_path = f"{run_uuid}_output_video.mp4"
|
170 |
+
if not os.path.exists(output_video_path):
|
171 |
+
raise FileNotFoundError(f"Error: {output_video_path} was not generated.")
|
172 |
|
173 |
+
cleanup_files(
|
174 |
f"{run_uuid}_resized_video.mp4",
|
175 |
f"{run_uuid}_output_audio.wav",
|
176 |
f"{run_uuid}_output_audio_final.wav",
|
177 |
f"{run_uuid}_output_synth.wav"
|
178 |
+
)
|
|
|
|
|
|
|
|
|
|
|
179 |
|
180 |
+
return output_video_path, ""
|
181 |
|
182 |
except Exception as e:
|
183 |
print(f"Error in process_video: {str(e)}")
|
184 |
+
return None, f"Error: {str(e)}"
|
185 |
|
186 |
def swap(radio):
|
187 |
+
return gr.update(source="upload" if radio == "Upload" else "webcam")
|
188 |
+
|
189 |
+
# Gradio interface setup
|
|
|
|
|
190 |
video = gr.Video()
|
191 |
radio = gr.Radio(["Upload", "Record"], value="Upload", show_label=False)
|
192 |
iface = gr.Interface(
|
|
|
194 |
inputs=[
|
195 |
radio,
|
196 |
video,
|
197 |
+
gr.Dropdown(choices=list(language_mapping.keys()), label="Target Language for Dubbing", value="Spanish"),
|
198 |
+
gr.Checkbox(label="Video has a close-up face. Use Wav2lip.", value=False, info="Say if video have close-up face. For Wav2lip. Will not work if checked wrongly.")
|
|
|
|
|
|
|
199 |
],
|
200 |
outputs=[
|
201 |
gr.Video(label="Processed Video"),
|