Fix import issue
Browse files- app.py +159 -32
- requirements.txt +10 -9
- video_accent_analyzer.py +160 -62
app.py
CHANGED
@@ -1,47 +1,174 @@
|
|
1 |
import gradio as gr
|
2 |
from video_accent_analyzer import VideoAccentAnalyzer
|
|
|
|
|
3 |
|
4 |
analyzer = VideoAccentAnalyzer()
|
5 |
|
6 |
|
7 |
-
def
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
result = analyzer.analyze_local_video(video_file)
|
12 |
-
else:
|
13 |
-
return "β Please provide a video URL or upload a file"
|
14 |
|
15 |
-
if '
|
16 |
-
|
17 |
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
|
|
|
|
|
|
|
|
23 |
|
24 |
-
|
25 |
-
|
|
|
|
|
|
|
|
|
|
|
26 |
|
27 |
-
return
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
28 |
|
29 |
|
30 |
# Create Gradio interface
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
45 |
|
46 |
if __name__ == "__main__":
|
47 |
-
interface.launch()
|
|
|
1 |
import gradio as gr
|
2 |
from video_accent_analyzer import VideoAccentAnalyzer
|
3 |
+
import plotly.graph_objects as go
|
4 |
+
import pandas as pd
|
5 |
|
6 |
analyzer = VideoAccentAnalyzer()
|
7 |
|
8 |
|
9 |
+
def create_plotly_chart(probabilities):
|
10 |
+
"""Create an interactive Plotly bar chart for accent probabilities"""
|
11 |
+
accents = [analyzer.accent_display_names.get(acc, acc.title()) for acc in probabilities.keys()]
|
12 |
+
probs = list(probabilities.values())
|
|
|
|
|
|
|
13 |
|
14 |
+
colors = ['#4CAF50' if p == max(probs) else '#2196F3' if p >= 20
|
15 |
+
else '#FFC107' if p >= 10 else '#9E9E9E' for p in probs]
|
16 |
|
17 |
+
fig = go.Figure(data=[
|
18 |
+
go.Bar(
|
19 |
+
x=accents,
|
20 |
+
y=probs,
|
21 |
+
marker_color=colors,
|
22 |
+
text=[f'{p:.1f}%' for p in probs],
|
23 |
+
textposition='auto',
|
24 |
+
)
|
25 |
+
])
|
26 |
|
27 |
+
fig.update_layout(
|
28 |
+
title='Accent Probability Distribution',
|
29 |
+
xaxis_title='Accent Type',
|
30 |
+
yaxis_title='Probability (%)',
|
31 |
+
template='plotly_white',
|
32 |
+
yaxis_range=[0, 100],
|
33 |
+
)
|
34 |
|
35 |
+
return fig
|
36 |
+
|
37 |
+
|
38 |
+
def analyze_video(url=None, video_file=None, duration=30):
|
39 |
+
"""Analyze video from URL or file with enhanced output"""
|
40 |
+
try:
|
41 |
+
if not url and not video_file:
|
42 |
+
return (
|
43 |
+
"### β Error\nPlease provide either a video URL or upload a video file.",
|
44 |
+
None
|
45 |
+
)
|
46 |
+
|
47 |
+
if url:
|
48 |
+
result = analyzer.analyze_video_url(url, max_duration=duration)
|
49 |
+
else:
|
50 |
+
result = analyzer.analyze_local_video(video_file, max_duration=duration)
|
51 |
+
|
52 |
+
if 'error' in result:
|
53 |
+
return (
|
54 |
+
f"### β Error\n{result['error']}",
|
55 |
+
None
|
56 |
+
)
|
57 |
+
|
58 |
+
# Create markdown output
|
59 |
+
markdown = f"""
|
60 |
+
### π― Analysis Results
|
61 |
+
|
62 |
+
**Primary Classification:**
|
63 |
+
- π£οΈ Predicted Accent: {analyzer.accent_display_names.get(result['predicted_accent'])}
|
64 |
+
- π Confidence: {result['accent_confidence']:.1f}%
|
65 |
+
- π English Confidence: {result['english_confidence']:.1f}%
|
66 |
+
|
67 |
+
**Audio Analysis:**
|
68 |
+
- β±οΈ Duration: {result['audio_duration']:.1f} seconds
|
69 |
+
- π Quality Score: {result.get('audio_quality_score', 'N/A')}
|
70 |
+
- π΅ Chunks Analyzed: {result.get('chunks_analyzed', 1)}
|
71 |
+
|
72 |
+
**Assessment:**
|
73 |
+
- {'β
Strong English Speaker' if result['english_confidence'] >= 70 else 'β οΈ Moderate English Confidence' if result['english_confidence'] >= 50 else 'β Low English Confidence'}
|
74 |
+
- {'π― High Accent Confidence' if result['accent_confidence'] >= 70 else 'π€ Moderate Accent Confidence' if result['accent_confidence'] >= 50 else 'β Low Accent Confidence'}
|
75 |
+
"""
|
76 |
+
|
77 |
+
# Create visualization
|
78 |
+
fig = create_plotly_chart(result['all_probabilities'])
|
79 |
+
|
80 |
+
return markdown, fig
|
81 |
+
|
82 |
+
except Exception as e:
|
83 |
+
return f"### β Error\nAn unexpected error occurred: {str(e)}", None
|
84 |
|
85 |
|
86 |
# Create Gradio interface
|
87 |
+
css = """
|
88 |
+
.gradio-container {
|
89 |
+
font-family: 'IBM Plex Sans', sans-serif;
|
90 |
+
}
|
91 |
+
.gr-button {
|
92 |
+
background: linear-gradient(45deg, #4CAF50, #2196F3);
|
93 |
+
border: none;
|
94 |
+
}
|
95 |
+
.gr-button:hover {
|
96 |
+
background: linear-gradient(45deg, #2196F3, #4CAF50);
|
97 |
+
transform: scale(1.02);
|
98 |
+
}
|
99 |
+
"""
|
100 |
+
|
101 |
+
with gr.Blocks(css=css) as interface:
|
102 |
+
gr.Markdown("""
|
103 |
+
# π§ Video Accent Analyzer
|
104 |
+
|
105 |
+
Analyze English accents in videos from various sources:
|
106 |
+
- YouTube videos
|
107 |
+
- Loom recordings
|
108 |
+
- Direct video links
|
109 |
+
- Uploaded video files
|
110 |
+
|
111 |
+
### π‘ Tips
|
112 |
+
- Keep videos under 2 minutes for best results
|
113 |
+
- Ensure clear audio quality
|
114 |
+
- Multiple speakers may affect accuracy
|
115 |
+
""")
|
116 |
+
|
117 |
+
with gr.Row():
|
118 |
+
with gr.Column():
|
119 |
+
url_input = gr.Textbox(
|
120 |
+
label="Video URL",
|
121 |
+
placeholder="Enter YouTube, Loom, or direct video URL"
|
122 |
+
)
|
123 |
+
video_input = gr.File(
|
124 |
+
label="Or Upload Video",
|
125 |
+
file_types=["video"]
|
126 |
+
)
|
127 |
+
duration = gr.Slider(
|
128 |
+
minimum=10,
|
129 |
+
maximum=120,
|
130 |
+
value=30,
|
131 |
+
step=10,
|
132 |
+
label="Maximum Duration (seconds)"
|
133 |
+
)
|
134 |
+
analyze_btn = gr.Button("π Analyze Video", variant="primary")
|
135 |
+
|
136 |
+
with gr.Column():
|
137 |
+
output_text = gr.Markdown(label="Analysis Results")
|
138 |
+
output_plot = gr.Plot(label="Accent Distribution")
|
139 |
+
|
140 |
+
analyze_btn.click(
|
141 |
+
fn=analyze_video,
|
142 |
+
inputs=[url_input, video_input, duration],
|
143 |
+
outputs=[output_text, output_plot]
|
144 |
+
)
|
145 |
+
|
146 |
+
gr.Examples(
|
147 |
+
examples=[
|
148 |
+
["https://www.youtube.com/watch?v=NO5SbsvIjHE", None, 30],
|
149 |
+
["https://www.youtube.com/watch?v=YQHsXMglC9A", None, 30],
|
150 |
+
],
|
151 |
+
inputs=[url_input, video_input, duration],
|
152 |
+
outputs=[output_text, output_plot],
|
153 |
+
label="Example Videos"
|
154 |
+
)
|
155 |
+
|
156 |
+
# Add requirements.txt
|
157 |
+
requirements = """
|
158 |
+
gradio>=4.0.0
|
159 |
+
plotly>=5.0.0
|
160 |
+
yt-dlp
|
161 |
+
librosa
|
162 |
+
soundfile
|
163 |
+
transformers
|
164 |
+
torch
|
165 |
+
ffmpeg-python
|
166 |
+
matplotlib
|
167 |
+
seaborn
|
168 |
+
"""
|
169 |
+
|
170 |
+
with open("requirements.txt", "w") as f:
|
171 |
+
f.write(requirements)
|
172 |
|
173 |
if __name__ == "__main__":
|
174 |
+
interface.launch()
|
requirements.txt
CHANGED
@@ -1,9 +1,10 @@
|
|
1 |
-
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
|
|
|
1 |
+
gradio>=4.0.0
|
2 |
+
plotly>=5.0.0
|
3 |
+
yt-dlp
|
4 |
+
librosa
|
5 |
+
soundfile
|
6 |
+
transformers
|
7 |
+
torch
|
8 |
+
ffmpeg-python
|
9 |
+
matplotlib
|
10 |
+
seaborn
|
video_accent_analyzer.py
CHANGED
@@ -1,3 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
1 |
|
2 |
import os
|
3 |
import sys
|
@@ -17,7 +22,6 @@ import seaborn as sns
|
|
17 |
# Suppress warnings for cleaner output
|
18 |
warnings.filterwarnings('ignore')
|
19 |
|
20 |
-
|
21 |
def install_if_missing(packages):
|
22 |
"""Install packages if they're not already available in Kaggle"""
|
23 |
for package in packages:
|
@@ -30,7 +34,6 @@ def install_if_missing(packages):
|
|
30 |
print(f"Installing {package}...")
|
31 |
subprocess.check_call([sys.executable, "-m", "pip", "install", package, "--quiet"])
|
32 |
|
33 |
-
|
34 |
# Required packages for Kaggle
|
35 |
required_packages = [
|
36 |
"yt-dlp",
|
@@ -52,7 +55,6 @@ import librosa
|
|
52 |
import soundfile as sf
|
53 |
import yt_dlp
|
54 |
|
55 |
-
|
56 |
class VideoAccentAnalyzer:
|
57 |
def __init__(self, model_name="dima806/multiple_accent_classification"):
|
58 |
"""Initialize the accent analyzer for Kaggle environment"""
|
@@ -101,6 +103,21 @@ class VideoAccentAnalyzer:
|
|
101 |
|
102 |
return True, url
|
103 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
104 |
def download_video(self, url, max_duration=None):
|
105 |
"""Download video using yt-dlp with improved error handling"""
|
106 |
is_valid, result = self._validate_url(url)
|
@@ -111,55 +128,106 @@ class VideoAccentAnalyzer:
|
|
111 |
url = result
|
112 |
output_path = os.path.join(self.temp_dir, "video.%(ext)s")
|
113 |
|
|
|
114 |
ydl_opts = {
|
115 |
'outtmpl': output_path,
|
116 |
-
'format': '
|
117 |
-
'quiet':
|
118 |
-
'no_warnings':
|
119 |
-
'socket_timeout':
|
120 |
-
'retries':
|
|
|
|
|
|
|
|
|
|
|
121 |
}
|
122 |
|
123 |
if max_duration:
|
124 |
-
|
125 |
-
|
126 |
|
127 |
try:
|
128 |
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
|
129 |
print(f"π₯ Downloading video from: {url}")
|
130 |
start_time = time.time()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
131 |
ydl.download([url])
|
132 |
download_time = time.time() - start_time
|
133 |
|
134 |
-
# Find downloaded file
|
|
|
135 |
for file in os.listdir(self.temp_dir):
|
136 |
-
if file.startswith("video."):
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
|
|
|
|
|
|
|
|
|
141 |
else:
|
142 |
-
print("
|
143 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
144 |
|
145 |
except Exception as e:
|
146 |
print(f"β οΈ yt-dlp failed: {e}")
|
147 |
return self._try_direct_download(url)
|
148 |
|
149 |
def _is_valid_video(self, file_path):
|
150 |
-
"""Verify video file has valid structure"""
|
151 |
try:
|
|
|
|
|
|
|
|
|
|
|
152 |
result = subprocess.run(
|
153 |
-
['ffprobe', '-v', '
|
154 |
-
capture_output=True, text=True, timeout=
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
155 |
)
|
156 |
-
|
|
|
|
|
157 |
except subprocess.TimeoutExpired:
|
158 |
-
print("β οΈ Video validation timed out")
|
159 |
-
return
|
160 |
except Exception as e:
|
161 |
-
print(f"β οΈ Video validation error: {e}")
|
162 |
-
return
|
163 |
|
164 |
def _try_direct_download(self, url):
|
165 |
"""Enhanced fallback for direct video URLs"""
|
@@ -186,7 +254,7 @@ class VideoAccentAnalyzer:
|
|
186 |
f.write(chunk)
|
187 |
file_size += len(chunk)
|
188 |
|
189 |
-
print(f"π Downloaded {file_size / (1024
|
190 |
|
191 |
if self._is_valid_video(video_path):
|
192 |
print("β
Direct download successful")
|
@@ -203,8 +271,9 @@ class VideoAccentAnalyzer:
|
|
203 |
"""Extract audio with improved error handling and progress"""
|
204 |
audio_path = os.path.join(self.temp_dir, "audio.wav")
|
205 |
|
|
|
206 |
cmd = ['ffmpeg', '-i', video_path, '-vn', '-acodec', 'pcm_s16le',
|
207 |
-
'-ar', '16000', '-ac', '1', '-y', '-
|
208 |
|
209 |
if max_duration:
|
210 |
cmd.extend(['-t', str(max_duration)])
|
@@ -213,15 +282,43 @@ class VideoAccentAnalyzer:
|
|
213 |
try:
|
214 |
print(f"π΅ Extracting audio (max {max_duration}s)...")
|
215 |
start_time = time.time()
|
216 |
-
|
|
|
|
|
217 |
extraction_time = time.time() - start_time
|
218 |
|
219 |
-
if result.returncode == 0 and os.path.exists(audio_path):
|
220 |
-
file_size = os.path.getsize(audio_path) / (1024
|
221 |
print(f"β
Audio extracted successfully ({extraction_time:.1f}s, {file_size:.1f}MB)")
|
222 |
return audio_path
|
223 |
else:
|
224 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
225 |
|
226 |
except subprocess.TimeoutExpired:
|
227 |
print("β Audio extraction timed out")
|
@@ -366,7 +463,7 @@ class VideoAccentAnalyzer:
|
|
366 |
return {"error": f"File not found: {file_path}"}
|
367 |
|
368 |
# Check file size
|
369 |
-
file_size = os.path.getsize(file_path) / (1024
|
370 |
print(f"π File size: {file_size:.1f} MB")
|
371 |
|
372 |
video_filename = os.path.basename(file_path)
|
@@ -394,8 +491,7 @@ class VideoAccentAnalyzer:
|
|
394 |
def display_results(self, results):
|
395 |
"""Enhanced results display with visualizations"""
|
396 |
if 'error' in results:
|
397 |
-
display(HTML(
|
398 |
-
f"<div style='color: red; font-size: 16px; padding: 10px; border: 1px solid red; border-radius: 5px;'>β {results['error']}</div>"))
|
399 |
return
|
400 |
|
401 |
accent = results['predicted_accent']
|
@@ -411,20 +507,20 @@ class VideoAccentAnalyzer:
|
|
411 |
html = f"""
|
412 |
<div style='border: 2px solid #4CAF50; border-radius: 10px; padding: 20px; margin: 10px 0; background: linear-gradient(135deg, #f5f7fa 0%, #c3cfe2 100%);'>
|
413 |
<h2 style='color: #2E7D32; margin-top: 0; text-align: center;'>π― Accent Analysis Results</h2>
|
414 |
-
|
415 |
<div style='display: flex; flex-wrap: wrap; gap: 20px; margin-bottom: 20px;'>
|
416 |
<div style='flex: 1; min-width: 200px; background: white; padding: 15px; border-radius: 8px; box-shadow: 0 2px 4px rgba(0,0,0,0.1);'>
|
417 |
<h3 style='color: #1976D2; margin-top: 0;'>π Primary Classification</h3>
|
418 |
<p style='font-size: 20px; margin: 5px 0; font-weight: bold;'>{accent_display}</p>
|
419 |
<p style='margin: 5px 0;'>Confidence: <strong style='color: {"#4CAF50" if confidence >= 70 else "#FF9800" if confidence >= 50 else "#F44336"};'>{confidence:.1f}%</strong></p>
|
420 |
</div>
|
421 |
-
|
422 |
<div style='flex: 1; min-width: 200px; background: white; padding: 15px; border-radius: 8px; box-shadow: 0 2px 4px rgba(0,0,0,0.1);'>
|
423 |
<h3 style='color: #1976D2; margin-top: 0;'>π English Proficiency</h3>
|
424 |
<p style='font-size: 18px; margin: 5px 0;'><strong style='color: {"#4CAF50" if english_conf >= 70 else "#FF9800" if english_conf >= 50 else "#F44336"};'>{english_conf:.1f}%</strong></p>
|
425 |
<p style='margin: 5px 0;'>Audio Quality: <strong>{quality_score:.0f}/100</strong></p>
|
426 |
</div>
|
427 |
-
|
428 |
<div style='flex: 1; min-width: 200px; background: white; padding: 15px; border-radius: 8px; box-shadow: 0 2px 4px rgba(0,0,0,0.1);'>
|
429 |
<h3 style='color: #1976D2; margin-top: 0;'>β±οΈ Processing Info</h3>
|
430 |
<p style='margin: 5px 0;'>Duration: <strong>{duration:.1f}s</strong></p>
|
@@ -432,7 +528,7 @@ class VideoAccentAnalyzer:
|
|
432 |
<p style='margin: 5px 0;'>Chunks: <strong>{results.get("chunks_analyzed", 1)}</strong></p>
|
433 |
</div>
|
434 |
</div>
|
435 |
-
|
436 |
<div style='background: white; padding: 15px; border-radius: 8px; box-shadow: 0 2px 4px rgba(0,0,0,0.1);'>
|
437 |
<h3 style='color: #1976D2; margin-top: 0;'>π Assessment</h3>
|
438 |
<div style='display: flex; flex-wrap: wrap; gap: 10px;'>
|
@@ -477,7 +573,7 @@ class VideoAccentAnalyzer:
|
|
477 |
|
478 |
# Create color map
|
479 |
colors = ['#4CAF50' if p == max(probs) else '#2196F3' if p >= 20 else '#FFC107' if p >= 10 else '#9E9E9E'
|
480 |
-
|
481 |
|
482 |
bars = plt.bar(accents, probs, color=colors, alpha=0.8, edgecolor='black', linewidth=0.5)
|
483 |
|
@@ -490,8 +586,8 @@ class VideoAccentAnalyzer:
|
|
490 |
# Add value labels on bars
|
491 |
for bar, prob in zip(bars, probs):
|
492 |
height = bar.get_height()
|
493 |
-
plt.text(bar.get_x() + bar.get_width()
|
494 |
-
|
495 |
|
496 |
plt.tight_layout()
|
497 |
plt.show()
|
@@ -507,7 +603,7 @@ class VideoAccentAnalyzer:
|
|
507 |
print(f"π Starting batch analysis of {len(urls)} videos")
|
508 |
|
509 |
for i, url in enumerate(urls, 1):
|
510 |
-
print(f"\n{'='
|
511 |
print(f"Processing video {i}/{len(urls)}")
|
512 |
|
513 |
result = self.analyze_video_url(url, max_duration)
|
@@ -553,7 +649,6 @@ class VideoAccentAnalyzer:
|
|
553 |
except Exception as e:
|
554 |
print(f"β οΈ Cleanup warning: {e}")
|
555 |
|
556 |
-
|
557 |
# Helper Functions
|
558 |
def show_examples():
|
559 |
"""Show usage examples"""
|
@@ -574,36 +669,40 @@ def show_examples():
|
|
574 |
print(" β’ Multiple speakers may affect accuracy")
|
575 |
print(" β’ Model works best with sustained speech")
|
576 |
|
577 |
-
|
578 |
-
|
579 |
-
"
|
580 |
-
print("
|
581 |
-
print("
|
582 |
-
print(" /
|
583 |
-
print(" /
|
584 |
-
|
585 |
-
|
586 |
-
|
587 |
-
|
|
|
588 |
return None
|
589 |
|
590 |
-
|
591 |
-
|
592 |
-
|
|
|
|
|
|
|
593 |
|
594 |
analyzer = VideoAccentAnalyzer()
|
595 |
try:
|
596 |
-
|
|
|
597 |
analyzer.display_results(results)
|
598 |
return results
|
599 |
finally:
|
600 |
analyzer.cleanup()
|
601 |
|
602 |
-
|
603 |
def demo_analysis():
|
604 |
"""Demo function with example usage"""
|
605 |
print("π¬ Video Accent Analyzer Demo")
|
606 |
-
print("="
|
607 |
|
608 |
# Initialize analyzer
|
609 |
analyzer = VideoAccentAnalyzer()
|
@@ -623,6 +722,5 @@ def demo_analysis():
|
|
623 |
print("3. analyzer.display_results(results)")
|
624 |
print("4. analyzer.cleanup() # Clean up temporary files")
|
625 |
|
626 |
-
|
627 |
# Show examples on import
|
628 |
show_examples()
|
|
|
1 |
+
"""
|
2 |
+
Enhanced Video Accent Analyzer
|
3 |
+
Supports YouTube, Loom, direct MP4 links, and local video files with improved error handling and features.
|
4 |
+
|
5 |
+
"""
|
6 |
|
7 |
import os
|
8 |
import sys
|
|
|
22 |
# Suppress warnings for cleaner output
|
23 |
warnings.filterwarnings('ignore')
|
24 |
|
|
|
25 |
def install_if_missing(packages):
|
26 |
"""Install packages if they're not already available in Kaggle"""
|
27 |
for package in packages:
|
|
|
34 |
print(f"Installing {package}...")
|
35 |
subprocess.check_call([sys.executable, "-m", "pip", "install", package, "--quiet"])
|
36 |
|
|
|
37 |
# Required packages for Kaggle
|
38 |
required_packages = [
|
39 |
"yt-dlp",
|
|
|
55 |
import soundfile as sf
|
56 |
import yt_dlp
|
57 |
|
|
|
58 |
class VideoAccentAnalyzer:
|
59 |
def __init__(self, model_name="dima806/multiple_accent_classification"):
|
60 |
"""Initialize the accent analyzer for Kaggle environment"""
|
|
|
103 |
|
104 |
return True, url
|
105 |
|
106 |
+
|
107 |
+
def trim_video(self, input_path, output_path, duration):
|
108 |
+
try:
|
109 |
+
cmd = ['ffmpeg', '-i', input_path, '-t', str(duration), '-c', 'copy', output_path, '-y']
|
110 |
+
result = subprocess.run(cmd, capture_output=True, text=True, timeout=60)
|
111 |
+
if result.returncode == 0:
|
112 |
+
print(f"βοΈ Trimmed video to {duration} seconds")
|
113 |
+
return output_path
|
114 |
+
else:
|
115 |
+
print(f"β Trimming failed: {result.stderr}")
|
116 |
+
return input_path # fallback to original
|
117 |
+
except Exception as e:
|
118 |
+
print(f"β Trimming exception: {e}")
|
119 |
+
return input_path
|
120 |
+
|
121 |
def download_video(self, url, max_duration=None):
|
122 |
"""Download video using yt-dlp with improved error handling"""
|
123 |
is_valid, result = self._validate_url(url)
|
|
|
128 |
url = result
|
129 |
output_path = os.path.join(self.temp_dir, "video.%(ext)s")
|
130 |
|
131 |
+
# Enhanced yt-dlp options for better compatibility
|
132 |
ydl_opts = {
|
133 |
'outtmpl': output_path,
|
134 |
+
'format': 'worst[ext=mp4]/worst', # Use worst quality for faster download and better compatibility
|
135 |
+
'quiet': False, # Show some output for debugging
|
136 |
+
'no_warnings': False,
|
137 |
+
'socket_timeout': 60,
|
138 |
+
'retries': 5,
|
139 |
+
'fragment_retries': 5,
|
140 |
+
'extract_flat': False,
|
141 |
+
'writeinfojson': False,
|
142 |
+
'writethumbnail': False,
|
143 |
+
'writesubtitles': False,
|
144 |
}
|
145 |
|
146 |
if max_duration:
|
147 |
+
# More generous time limit for download
|
148 |
+
ydl_opts['match_filter'] = lambda info: None if info.get('duration', 0) <= 200000 else "Video too long"
|
149 |
|
150 |
try:
|
151 |
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
|
152 |
print(f"π₯ Downloading video from: {url}")
|
153 |
start_time = time.time()
|
154 |
+
|
155 |
+
# Get video info first
|
156 |
+
try:
|
157 |
+
info = ydl.extract_info(url, download=False)
|
158 |
+
print(f"πΊ Found video: {info.get('title', 'Unknown')} ({info.get('duration', 0)}s)")
|
159 |
+
except Exception as e:
|
160 |
+
print(f"β οΈ Could not extract video info: {e}")
|
161 |
+
|
162 |
+
# Download the video
|
163 |
ydl.download([url])
|
164 |
download_time = time.time() - start_time
|
165 |
|
166 |
+
# Find downloaded file (try multiple patterns)
|
167 |
+
video_path = None
|
168 |
for file in os.listdir(self.temp_dir):
|
169 |
+
if file.startswith("video.") and os.path.getsize(os.path.join(self.temp_dir, file)) > 1000: # At least 1KB
|
170 |
+
potential_path = os.path.join(self.temp_dir, file)
|
171 |
+
print(f"π Found downloaded file: {file} ({os.path.getsize(potential_path)/1024:.1f}KB)")
|
172 |
+
|
173 |
+
# Try basic validation - if ffprobe fails, still try to extract audio
|
174 |
+
if self._is_valid_video(potential_path):
|
175 |
+
print(f"β
Video validation passed: {file}")
|
176 |
+
video_path = potential_path
|
177 |
+
break
|
178 |
else:
|
179 |
+
print(f"β οΈ Video validation failed, but continuing with: {file}")
|
180 |
+
video_path = potential_path # Still try to use it
|
181 |
+
break
|
182 |
+
|
183 |
+
if video_path:
|
184 |
+
print(f"β
Downloaded video: {os.path.basename(video_path)} ({download_time:.1f}s)")
|
185 |
+
return video_path
|
186 |
+
else:
|
187 |
+
print("β No video files found after download")
|
188 |
+
return None
|
189 |
|
190 |
except Exception as e:
|
191 |
print(f"β οΈ yt-dlp failed: {e}")
|
192 |
return self._try_direct_download(url)
|
193 |
|
194 |
def _is_valid_video(self, file_path):
|
195 |
+
"""Verify video file has valid structure (more lenient)"""
|
196 |
try:
|
197 |
+
# First check if file exists and has reasonable size
|
198 |
+
if not os.path.exists(file_path) or os.path.getsize(file_path) < 1000:
|
199 |
+
return False
|
200 |
+
|
201 |
+
# Try ffprobe with more lenient settings
|
202 |
result = subprocess.run(
|
203 |
+
['ffprobe', '-v', 'quiet', '-print_format', 'json', '-show_format', file_path],
|
204 |
+
capture_output=True, text=True, timeout=15
|
205 |
+
)
|
206 |
+
|
207 |
+
if result.returncode == 0:
|
208 |
+
try:
|
209 |
+
# Try to parse the JSON output
|
210 |
+
info = json.loads(result.stdout)
|
211 |
+
# Check if we have format information
|
212 |
+
if 'format' in info and 'duration' in info['format']:
|
213 |
+
return True
|
214 |
+
except json.JSONDecodeError:
|
215 |
+
pass
|
216 |
+
|
217 |
+
# If ffprobe fails, try a simpler check - just see if ffmpeg can read it
|
218 |
+
result2 = subprocess.run(
|
219 |
+
['ffmpeg', '-i', file_path, '-t', '1', '-f', 'null', '-', '-v', 'quiet'],
|
220 |
+
capture_output=True, text=True, timeout=15
|
221 |
)
|
222 |
+
|
223 |
+
return result2.returncode == 0
|
224 |
+
|
225 |
except subprocess.TimeoutExpired:
|
226 |
+
print("β οΈ Video validation timed out, assuming valid")
|
227 |
+
return True # If validation times out, assume it's valid
|
228 |
except Exception as e:
|
229 |
+
print(f"β οΈ Video validation error: {e}, assuming valid")
|
230 |
+
return True # If validation fails, assume it's valid and let audio extraction handle it
|
231 |
|
232 |
def _try_direct_download(self, url):
|
233 |
"""Enhanced fallback for direct video URLs"""
|
|
|
254 |
f.write(chunk)
|
255 |
file_size += len(chunk)
|
256 |
|
257 |
+
print(f"π Downloaded {file_size / (1024*1024):.1f} MB")
|
258 |
|
259 |
if self._is_valid_video(video_path):
|
260 |
print("β
Direct download successful")
|
|
|
271 |
"""Extract audio with improved error handling and progress"""
|
272 |
audio_path = os.path.join(self.temp_dir, "audio.wav")
|
273 |
|
274 |
+
# Enhanced ffmpeg command with better error handling
|
275 |
cmd = ['ffmpeg', '-i', video_path, '-vn', '-acodec', 'pcm_s16le',
|
276 |
+
'-ar', '16000', '-ac', '1', '-y', '-v', 'warning']
|
277 |
|
278 |
if max_duration:
|
279 |
cmd.extend(['-t', str(max_duration)])
|
|
|
282 |
try:
|
283 |
print(f"π΅ Extracting audio (max {max_duration}s)...")
|
284 |
start_time = time.time()
|
285 |
+
|
286 |
+
# Run ffmpeg with more detailed output for debugging
|
287 |
+
result = subprocess.run(cmd, capture_output=True, text=True, timeout=180)
|
288 |
extraction_time = time.time() - start_time
|
289 |
|
290 |
+
if result.returncode == 0 and os.path.exists(audio_path) and os.path.getsize(audio_path) > 1000:
|
291 |
+
file_size = os.path.getsize(audio_path) / (1024*1024)
|
292 |
print(f"β
Audio extracted successfully ({extraction_time:.1f}s, {file_size:.1f}MB)")
|
293 |
return audio_path
|
294 |
else:
|
295 |
+
print(f"β FFmpeg stderr: {result.stderr}")
|
296 |
+
print(f"β FFmpeg stdout: {result.stdout}")
|
297 |
+
|
298 |
+
# Try alternative extraction method
|
299 |
+
print("π Trying alternative audio extraction...")
|
300 |
+
cmd_alt = ['ffmpeg', '-i', video_path, '-vn', '-acodec', 'libmp3lame',
|
301 |
+
'-ar', '16000', '-ac', '1', '-y', '-v', 'warning']
|
302 |
+
if max_duration:
|
303 |
+
cmd_alt.extend(['-t', str(max_duration)])
|
304 |
+
|
305 |
+
audio_path_alt = os.path.join(self.temp_dir, "audio.mp3")
|
306 |
+
cmd_alt.append(audio_path_alt)
|
307 |
+
|
308 |
+
result_alt = subprocess.run(cmd_alt, capture_output=True, text=True, timeout=180)
|
309 |
+
|
310 |
+
if result_alt.returncode == 0 and os.path.exists(audio_path_alt):
|
311 |
+
# Convert mp3 to wav
|
312 |
+
cmd_convert = ['ffmpeg', '-i', audio_path_alt, '-ar', '16000', '-ac', '1',
|
313 |
+
audio_path, '-y', '-v', 'quiet']
|
314 |
+
result_convert = subprocess.run(cmd_convert, capture_output=True, text=True, timeout=60)
|
315 |
+
|
316 |
+
if result_convert.returncode == 0 and os.path.exists(audio_path):
|
317 |
+
file_size = os.path.getsize(audio_path) / (1024*1024)
|
318 |
+
print(f"β
Alternative extraction successful ({file_size:.1f}MB)")
|
319 |
+
return audio_path
|
320 |
+
|
321 |
+
raise Exception(f"Both extraction methods failed. FFmpeg error: {result.stderr}")
|
322 |
|
323 |
except subprocess.TimeoutExpired:
|
324 |
print("β Audio extraction timed out")
|
|
|
463 |
return {"error": f"File not found: {file_path}"}
|
464 |
|
465 |
# Check file size
|
466 |
+
file_size = os.path.getsize(file_path) / (1024*1024) # MB
|
467 |
print(f"π File size: {file_size:.1f} MB")
|
468 |
|
469 |
video_filename = os.path.basename(file_path)
|
|
|
491 |
def display_results(self, results):
|
492 |
"""Enhanced results display with visualizations"""
|
493 |
if 'error' in results:
|
494 |
+
display(HTML(f"<div style='color: red; font-size: 16px; padding: 10px; border: 1px solid red; border-radius: 5px;'>β {results['error']}</div>"))
|
|
|
495 |
return
|
496 |
|
497 |
accent = results['predicted_accent']
|
|
|
507 |
html = f"""
|
508 |
<div style='border: 2px solid #4CAF50; border-radius: 10px; padding: 20px; margin: 10px 0; background: linear-gradient(135deg, #f5f7fa 0%, #c3cfe2 100%);'>
|
509 |
<h2 style='color: #2E7D32; margin-top: 0; text-align: center;'>π― Accent Analysis Results</h2>
|
510 |
+
|
511 |
<div style='display: flex; flex-wrap: wrap; gap: 20px; margin-bottom: 20px;'>
|
512 |
<div style='flex: 1; min-width: 200px; background: white; padding: 15px; border-radius: 8px; box-shadow: 0 2px 4px rgba(0,0,0,0.1);'>
|
513 |
<h3 style='color: #1976D2; margin-top: 0;'>π Primary Classification</h3>
|
514 |
<p style='font-size: 20px; margin: 5px 0; font-weight: bold;'>{accent_display}</p>
|
515 |
<p style='margin: 5px 0;'>Confidence: <strong style='color: {"#4CAF50" if confidence >= 70 else "#FF9800" if confidence >= 50 else "#F44336"};'>{confidence:.1f}%</strong></p>
|
516 |
</div>
|
517 |
+
|
518 |
<div style='flex: 1; min-width: 200px; background: white; padding: 15px; border-radius: 8px; box-shadow: 0 2px 4px rgba(0,0,0,0.1);'>
|
519 |
<h3 style='color: #1976D2; margin-top: 0;'>π English Proficiency</h3>
|
520 |
<p style='font-size: 18px; margin: 5px 0;'><strong style='color: {"#4CAF50" if english_conf >= 70 else "#FF9800" if english_conf >= 50 else "#F44336"};'>{english_conf:.1f}%</strong></p>
|
521 |
<p style='margin: 5px 0;'>Audio Quality: <strong>{quality_score:.0f}/100</strong></p>
|
522 |
</div>
|
523 |
+
|
524 |
<div style='flex: 1; min-width: 200px; background: white; padding: 15px; border-radius: 8px; box-shadow: 0 2px 4px rgba(0,0,0,0.1);'>
|
525 |
<h3 style='color: #1976D2; margin-top: 0;'>β±οΈ Processing Info</h3>
|
526 |
<p style='margin: 5px 0;'>Duration: <strong>{duration:.1f}s</strong></p>
|
|
|
528 |
<p style='margin: 5px 0;'>Chunks: <strong>{results.get("chunks_analyzed", 1)}</strong></p>
|
529 |
</div>
|
530 |
</div>
|
531 |
+
|
532 |
<div style='background: white; padding: 15px; border-radius: 8px; box-shadow: 0 2px 4px rgba(0,0,0,0.1);'>
|
533 |
<h3 style='color: #1976D2; margin-top: 0;'>π Assessment</h3>
|
534 |
<div style='display: flex; flex-wrap: wrap; gap: 10px;'>
|
|
|
573 |
|
574 |
# Create color map
|
575 |
colors = ['#4CAF50' if p == max(probs) else '#2196F3' if p >= 20 else '#FFC107' if p >= 10 else '#9E9E9E'
|
576 |
+
for p in probs]
|
577 |
|
578 |
bars = plt.bar(accents, probs, color=colors, alpha=0.8, edgecolor='black', linewidth=0.5)
|
579 |
|
|
|
586 |
# Add value labels on bars
|
587 |
for bar, prob in zip(bars, probs):
|
588 |
height = bar.get_height()
|
589 |
+
plt.text(bar.get_x() + bar.get_width()/2., height + 0.5,
|
590 |
+
f'{prob:.1f}%', ha='center', va='bottom', fontweight='bold')
|
591 |
|
592 |
plt.tight_layout()
|
593 |
plt.show()
|
|
|
603 |
print(f"π Starting batch analysis of {len(urls)} videos")
|
604 |
|
605 |
for i, url in enumerate(urls, 1):
|
606 |
+
print(f"\n{'='*60}")
|
607 |
print(f"Processing video {i}/{len(urls)}")
|
608 |
|
609 |
result = self.analyze_video_url(url, max_duration)
|
|
|
649 |
except Exception as e:
|
650 |
print(f"β οΈ Cleanup warning: {e}")
|
651 |
|
|
|
652 |
# Helper Functions
|
653 |
def show_examples():
|
654 |
"""Show usage examples"""
|
|
|
669 |
print(" β’ Multiple speakers may affect accuracy")
|
670 |
print(" β’ Model works best with sustained speech")
|
671 |
|
672 |
+
def quick_test_url():
|
673 |
+
"""Interactive test for video URLs"""
|
674 |
+
print("π Quick Test Mode for Video URLs")
|
675 |
+
print("π― Supported: YouTube, Loom, Direct MP4 links")
|
676 |
+
print("π‘ Examples:")
|
677 |
+
print(" YouTube: https://youtube.com/watch?v=VIDEO_ID")
|
678 |
+
print(" Loom: https://www.loom.com/share/VIDEO_ID")
|
679 |
+
print(" Direct: https://example.com/video.mp4")
|
680 |
+
|
681 |
+
url = input("\nπ Enter your video URL (Loom, YouTube, etc.): ").strip()
|
682 |
+
if not url:
|
683 |
+
print("β No URL provided.")
|
684 |
return None
|
685 |
|
686 |
+
max_duration = input("β±οΈ Max duration in seconds (default 20): ").strip()
|
687 |
+
try:
|
688 |
+
max_duration = int(max_duration) if max_duration else 20
|
689 |
+
except ValueError:
|
690 |
+
max_duration = 20
|
691 |
+
print(f"β οΈ Invalid duration, using {max_duration} seconds")
|
692 |
|
693 |
analyzer = VideoAccentAnalyzer()
|
694 |
try:
|
695 |
+
print(f"\nπ Starting analysis...")
|
696 |
+
results = analyzer.analyze_video_url(url, max_duration=max_duration)
|
697 |
analyzer.display_results(results)
|
698 |
return results
|
699 |
finally:
|
700 |
analyzer.cleanup()
|
701 |
|
|
|
702 |
def demo_analysis():
|
703 |
"""Demo function with example usage"""
|
704 |
print("π¬ Video Accent Analyzer Demo")
|
705 |
+
print("="*50)
|
706 |
|
707 |
# Initialize analyzer
|
708 |
analyzer = VideoAccentAnalyzer()
|
|
|
722 |
print("3. analyzer.display_results(results)")
|
723 |
print("4. analyzer.cleanup() # Clean up temporary files")
|
724 |
|
|
|
725 |
# Show examples on import
|
726 |
show_examples()
|