Spaces:
Sleeping
Sleeping
Commit
·
356f877
1
Parent(s):
91a85ec
feat(app.py): add download video and audio options
Browse files
app.py
CHANGED
@@ -9,122 +9,112 @@ from silero_vad import load_silero_vad, get_speech_timestamps
|
|
9 |
import numpy as np
|
10 |
import pydub
|
11 |
|
12 |
-
VAD_SENSITIVITY = 0.1
|
13 |
-
|
14 |
# --- Model Loading and Caching ---
|
15 |
@st.cache_resource
|
16 |
def load_transcriber(_device):
|
|
|
17 |
transcriber = pipeline(model="openai/whisper-large-v3-turbo", device=_device)
|
18 |
return transcriber
|
19 |
|
20 |
@st.cache_resource
|
21 |
def load_vad_model():
|
|
|
22 |
return load_silero_vad()
|
23 |
|
24 |
# --- Audio Processing Functions ---
|
25 |
@st.cache_resource
|
26 |
-
def download_and_convert_audio(video_url):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
status_message = st.empty()
|
28 |
status_message.text("Downloading audio...")
|
29 |
try:
|
30 |
ydl_opts = {
|
31 |
-
'format': 'bestaudio/best',
|
32 |
'postprocessors': [{
|
33 |
'key': 'FFmpegExtractAudio',
|
34 |
-
'preferredcodec':
|
35 |
-
'preferredquality': '192',
|
36 |
}],
|
37 |
'outtmpl': '%(id)s.%(ext)s',
|
|
|
|
|
38 |
}
|
39 |
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
|
40 |
info = ydl.extract_info(video_url, download=False)
|
|
|
|
|
41 |
video_id = info['id']
|
42 |
-
filename = f"{video_id}.
|
|
|
|
|
|
|
|
|
|
|
|
|
43 |
ydl.download([video_url])
|
44 |
-
status_message.text("Audio downloaded and converted.")
|
45 |
-
|
46 |
-
# Read the file and return its contents
|
47 |
with open(filename, 'rb') as audio_file:
|
48 |
audio_bytes = audio_file.read()
|
49 |
-
|
50 |
-
# Clean up the temporary file
|
51 |
os.remove(filename)
|
52 |
-
|
53 |
-
return audio_bytes, 'wav'
|
54 |
except Exception as e:
|
55 |
st.error(f"Error during download or conversion: {e}")
|
56 |
-
return None, None
|
|
|
|
|
|
|
|
|
|
|
|
|
57 |
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
|
62 |
Args:
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
|
|
68 |
|
69 |
Returns:
|
70 |
-
A list of dictionaries, where each dictionary represents an
|
71 |
-
|
72 |
"""
|
73 |
-
|
74 |
-
if not speech_timestamps:
|
75 |
-
return []
|
76 |
-
|
77 |
-
aggregated_segments = []
|
78 |
-
current_segment_start = speech_timestamps[0]['start']
|
79 |
-
current_segment_end = speech_timestamps[0]['end']
|
80 |
-
|
81 |
-
for segment in speech_timestamps[1:]:
|
82 |
-
if segment['start'] - current_segment_start >= max_duration:
|
83 |
-
# Start a new segment if the current duration exceeds max_duration
|
84 |
-
aggregated_segments.append({'start': current_segment_start, 'end': current_segment_end})
|
85 |
-
current_segment_start = segment['start']
|
86 |
-
current_segment_end = segment['end']
|
87 |
-
else:
|
88 |
-
# Extend the current segment
|
89 |
-
current_segment_end = segment['end']
|
90 |
-
|
91 |
-
# Add the last segment, checking for redundancy
|
92 |
-
last_segment = {'start': current_segment_start, 'end': current_segment_end}
|
93 |
-
if aggregated_segments:
|
94 |
-
second_last_segment = aggregated_segments[-1]
|
95 |
-
if last_segment['start'] >= second_last_segment['start'] and last_segment['end'] <= second_last_segment['end']:
|
96 |
-
# Last segment is fully contained in the second-to-last, so don't add it
|
97 |
-
pass
|
98 |
-
else:
|
99 |
-
aggregated_segments.append(last_segment)
|
100 |
-
else:
|
101 |
-
# If aggregated_segments is empty, add the last segment
|
102 |
-
aggregated_segments.append(last_segment)
|
103 |
-
|
104 |
-
return aggregated_segments
|
105 |
-
|
106 |
-
@st.cache_data
|
107 |
-
def split_audio_by_vad(audio_data: bytes, ext: str, _vad_model, sensitivity: float, return_seconds: bool = True):
|
108 |
if not audio_data:
|
109 |
st.error("No audio data received.")
|
110 |
return []
|
111 |
-
|
112 |
try:
|
113 |
audio = pydub.AudioSegment.from_file(io.BytesIO(audio_data), format=ext)
|
114 |
-
|
115 |
-
# VAD parameters
|
116 |
rate = audio.frame_rate
|
|
|
|
|
|
|
|
|
|
|
|
|
117 |
window_size_samples = int(512 + (1536 - 512) * (1 - sensitivity))
|
118 |
speech_threshold = 0.5 + (0.95 - 0.5) * sensitivity
|
119 |
|
120 |
-
# Convert audio to numpy array for VAD
|
121 |
samples = np.array(audio.get_array_of_samples())
|
122 |
|
123 |
-
# Get speech timestamps
|
124 |
speech_timestamps = get_speech_timestamps(
|
125 |
-
samples,
|
126 |
_vad_model,
|
127 |
-
sampling_rate=rate,
|
128 |
return_seconds=return_seconds,
|
129 |
window_size_samples=window_size_samples,
|
130 |
threshold=speech_threshold,
|
@@ -134,43 +124,45 @@ def split_audio_by_vad(audio_data: bytes, ext: str, _vad_model, sensitivity: flo
|
|
134 |
st.warning("No speech segments detected.")
|
135 |
return []
|
136 |
|
137 |
-
# rectify timestamps
|
138 |
speech_timestamps[0]["start"] = 0.
|
139 |
speech_timestamps[-1]['end'] = audio.duration_seconds
|
140 |
for i, chunk in enumerate(speech_timestamps[1:], start=1):
|
141 |
-
chunk["start"] = speech_timestamps[i-1]['end']
|
142 |
-
|
143 |
-
# Aggregate segments into ~30 second chunks
|
144 |
-
aggregated_segments = aggregate_speech_segments(speech_timestamps, max_duration=30)
|
145 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
146 |
if not aggregated_segments:
|
147 |
return []
|
148 |
|
149 |
-
# Create audio chunks based on timestamps
|
150 |
chunks = []
|
151 |
for segment in aggregated_segments:
|
152 |
start_ms = int(segment['start'] * 1000)
|
153 |
end_ms = int(segment['end'] * 1000)
|
154 |
chunk = audio[start_ms:end_ms]
|
155 |
-
|
156 |
-
# Export chunk to bytes
|
157 |
chunk_io = io.BytesIO()
|
158 |
chunk.export(chunk_io, format=ext)
|
159 |
-
chunk_data = chunk_io.getvalue() # Get bytes directly
|
160 |
-
|
161 |
chunks.append({
|
162 |
-
'data':
|
163 |
'start': segment['start'],
|
164 |
'end': segment['end']
|
165 |
})
|
166 |
-
chunk_io.close()
|
167 |
-
|
168 |
return chunks
|
169 |
except Exception as e:
|
170 |
st.error(f"Error processing audio in split_audio_by_vad: {str(e)}")
|
171 |
return []
|
172 |
finally:
|
173 |
-
# Explicitly release pydub resources to prevent memory issues
|
174 |
if 'audio' in locals():
|
175 |
del audio
|
176 |
if 'samples' in locals():
|
@@ -178,18 +170,28 @@ def split_audio_by_vad(audio_data: bytes, ext: str, _vad_model, sensitivity: flo
|
|
178 |
|
179 |
@st.cache_data
|
180 |
def transcribe_batch(batch, _transcriber, language=None):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
181 |
transcriptions = []
|
182 |
for i, chunk_data in enumerate(batch):
|
183 |
try:
|
184 |
generate_kwargs = {
|
185 |
"task": "transcribe",
|
186 |
-
"return_timestamps": True
|
|
|
187 |
}
|
188 |
-
|
189 |
-
generate_kwargs["language"] = language
|
190 |
-
|
191 |
transcription = _transcriber(
|
192 |
-
chunk_data['data'],
|
193 |
generate_kwargs=generate_kwargs
|
194 |
)
|
195 |
transcriptions.append({
|
@@ -204,47 +206,93 @@ def transcribe_batch(batch, _transcriber, language=None):
|
|
204 |
|
205 |
# --- Streamlit App ---
|
206 |
def setup_ui():
|
|
|
207 |
st.title("YouTube Video Transcriber")
|
208 |
-
|
209 |
-
|
210 |
-
|
211 |
-
|
212 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
213 |
|
214 |
@st.cache_resource
|
215 |
def initialize_models():
|
|
|
216 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
217 |
transcriber = load_transcriber(device)
|
218 |
vad_model = load_vad_model()
|
219 |
return transcriber, vad_model
|
220 |
|
221 |
-
def process_transcription(video_url, vad_sensitivity, batch_size, transcriber, vad_model, language=None):
|
222 |
-
|
223 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
224 |
if not audio_data:
|
225 |
-
return
|
226 |
-
|
227 |
-
chunks = split_audio_by_vad(audio_data,
|
228 |
if not chunks:
|
229 |
-
return
|
230 |
|
231 |
total_chunks = len(chunks)
|
232 |
transcriptions = []
|
|
|
233 |
for i in range(0, total_chunks, batch_size):
|
234 |
batch = chunks[i:i + batch_size]
|
235 |
batch_transcriptions = transcribe_batch(batch, transcriber, language)
|
236 |
transcriptions.extend(batch_transcriptions)
|
237 |
-
|
238 |
|
|
|
239 |
st.success("Transcription complete!")
|
240 |
|
241 |
-
def display_transcription(transcriptions, output_area):
|
242 |
full_transcription = ""
|
243 |
for chunk in transcriptions:
|
244 |
start_time = format_seconds(chunk['start'])
|
245 |
end_time = format_seconds(chunk['end'])
|
246 |
full_transcription += f"[{start_time} - {end_time}]: {chunk['text'].strip()}\n\n"
|
247 |
-
|
|
|
248 |
|
249 |
def format_seconds(seconds):
|
250 |
"""Formats seconds into HH:MM:SS string."""
|
@@ -252,14 +300,123 @@ def format_seconds(seconds):
|
|
252 |
hours, minutes = divmod(minutes, 60)
|
253 |
return f"{int(hours):02}:{int(minutes):02}:{int(seconds):02}"
|
254 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
255 |
def main():
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
256 |
transcriber, vad_model = initialize_models()
|
257 |
-
|
258 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
259 |
if not video_url:
|
260 |
st.error("Please enter a YouTube video link.")
|
261 |
return
|
262 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
263 |
|
264 |
if __name__ == "__main__":
|
265 |
main()
|
|
|
9 |
import numpy as np
|
10 |
import pydub
|
11 |
|
|
|
|
|
12 |
# --- Model Loading and Caching ---
|
13 |
@st.cache_resource
|
14 |
def load_transcriber(_device):
|
15 |
+
"""Loads the Whisper transcription model."""
|
16 |
transcriber = pipeline(model="openai/whisper-large-v3-turbo", device=_device)
|
17 |
return transcriber
|
18 |
|
19 |
@st.cache_resource
|
20 |
def load_vad_model():
|
21 |
+
"""Loads the Silero VAD model."""
|
22 |
return load_silero_vad()
|
23 |
|
24 |
# --- Audio Processing Functions ---
|
25 |
@st.cache_resource
|
26 |
+
def download_and_convert_audio(video_url, audio_format="wav"):
|
27 |
+
"""Downloads and converts audio from a YouTube video.
|
28 |
+
|
29 |
+
Args:
|
30 |
+
video_url (str): The URL of the YouTube video.
|
31 |
+
audio_format (str): The desired audio format (e.g., "wav", "mp3").
|
32 |
+
|
33 |
+
Returns:
|
34 |
+
tuple: (audio_bytes, audio_format, info_dict) or (None, None, None) on error.
|
35 |
+
"""
|
36 |
status_message = st.empty()
|
37 |
status_message.text("Downloading audio...")
|
38 |
try:
|
39 |
ydl_opts = {
|
40 |
+
'format': f'bestaudio/best',
|
41 |
'postprocessors': [{
|
42 |
'key': 'FFmpegExtractAudio',
|
43 |
+
'preferredcodec': audio_format,
|
|
|
44 |
}],
|
45 |
'outtmpl': '%(id)s.%(ext)s',
|
46 |
+
'noplaylist': True,
|
47 |
+
'progress_hooks': [lambda d: update_download_progress(d, status_message)],
|
48 |
}
|
49 |
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
|
50 |
info = ydl.extract_info(video_url, download=False)
|
51 |
+
if 'entries' in info:
|
52 |
+
info = info['entries'][0]
|
53 |
video_id = info['id']
|
54 |
+
filename = f"{video_id}.{audio_format}"
|
55 |
+
|
56 |
+
audio_formats = [f for f in info.get('formats', []) if f.get('acodec') != 'none' and f.get('vcodec') == 'none']
|
57 |
+
if not audio_formats:
|
58 |
+
st.warning(f"No audio-only format found. Downloading and converting from best video format to {audio_format}.")
|
59 |
+
ydl_opts['format'] = 'best'
|
60 |
+
|
61 |
ydl.download([video_url])
|
62 |
+
status_message.text(f"Audio downloaded and converted to {audio_format}.")
|
63 |
+
|
|
|
64 |
with open(filename, 'rb') as audio_file:
|
65 |
audio_bytes = audio_file.read()
|
66 |
+
|
|
|
67 |
os.remove(filename)
|
68 |
+
return audio_bytes, audio_format, info
|
|
|
69 |
except Exception as e:
|
70 |
st.error(f"Error during download or conversion: {e}")
|
71 |
+
return None, None, None
|
72 |
+
|
73 |
+
def update_download_progress(d, status_message):
|
74 |
+
"""Updates the download progress in the Streamlit UI."""
|
75 |
+
if d['status'] == 'downloading':
|
76 |
+
p = round(d['downloaded_bytes'] / d['total_bytes'] * 100)
|
77 |
+
status_message.text(f"Downloading: {p}%")
|
78 |
|
79 |
+
@st.cache_data
|
80 |
+
def split_audio_by_vad(audio_data: bytes, ext: str, _vad_model, sensitivity: float, max_duration: int = 30, return_seconds: bool = True):
|
81 |
+
"""Splits audio into chunks based on voice activity detection (VAD).
|
82 |
|
83 |
Args:
|
84 |
+
audio_data (bytes): The audio data as bytes.
|
85 |
+
ext (str): The audio file extension.
|
86 |
+
_vad_model: The VAD model.
|
87 |
+
sensitivity (float): The VAD sensitivity (0.0 to 1.0).
|
88 |
+
max_duration (int): The maximum duration of each chunk in seconds.
|
89 |
+
return_seconds (bool): Whether to return timestamps in seconds.
|
90 |
|
91 |
Returns:
|
92 |
+
list: A list of dictionaries, where each dictionary represents an audio chunk.
|
93 |
+
Returns an empty list if no speech segments are detected or an error occurs.
|
94 |
"""
|
95 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
96 |
if not audio_data:
|
97 |
st.error("No audio data received.")
|
98 |
return []
|
99 |
+
|
100 |
try:
|
101 |
audio = pydub.AudioSegment.from_file(io.BytesIO(audio_data), format=ext)
|
|
|
|
|
102 |
rate = audio.frame_rate
|
103 |
+
|
104 |
+
# Convert to mono if stereo for compatibility with VAD
|
105 |
+
if audio.channels > 1:
|
106 |
+
audio = audio.set_channels(1)
|
107 |
+
|
108 |
+
# Calculate dynamic VAD parameters based on sensitivity
|
109 |
window_size_samples = int(512 + (1536 - 512) * (1 - sensitivity))
|
110 |
speech_threshold = 0.5 + (0.95 - 0.5) * sensitivity
|
111 |
|
|
|
112 |
samples = np.array(audio.get_array_of_samples())
|
113 |
|
|
|
114 |
speech_timestamps = get_speech_timestamps(
|
115 |
+
samples,
|
116 |
_vad_model,
|
117 |
+
sampling_rate=rate,
|
118 |
return_seconds=return_seconds,
|
119 |
window_size_samples=window_size_samples,
|
120 |
threshold=speech_threshold,
|
|
|
124 |
st.warning("No speech segments detected.")
|
125 |
return []
|
126 |
|
|
|
127 |
speech_timestamps[0]["start"] = 0.
|
128 |
speech_timestamps[-1]['end'] = audio.duration_seconds
|
129 |
for i, chunk in enumerate(speech_timestamps[1:], start=1):
|
130 |
+
chunk["start"] = speech_timestamps[i - 1]['end']
|
|
|
|
|
|
|
131 |
|
132 |
+
aggregated_segments = []
|
133 |
+
if speech_timestamps:
|
134 |
+
current_segment_start = speech_timestamps[0]['start']
|
135 |
+
current_segment_end = speech_timestamps[0]['end']
|
136 |
+
for segment in speech_timestamps[1:]:
|
137 |
+
if segment['start'] - current_segment_start >= max_duration:
|
138 |
+
aggregated_segments.append({'start': current_segment_start, 'end': current_segment_end})
|
139 |
+
current_segment_start = segment['start']
|
140 |
+
current_segment_end = segment['end']
|
141 |
+
else:
|
142 |
+
current_segment_end = segment['end']
|
143 |
+
aggregated_segments.append({'start': current_segment_start, 'end': current_segment_end})
|
144 |
+
|
145 |
if not aggregated_segments:
|
146 |
return []
|
147 |
|
|
|
148 |
chunks = []
|
149 |
for segment in aggregated_segments:
|
150 |
start_ms = int(segment['start'] * 1000)
|
151 |
end_ms = int(segment['end'] * 1000)
|
152 |
chunk = audio[start_ms:end_ms]
|
|
|
|
|
153 |
chunk_io = io.BytesIO()
|
154 |
chunk.export(chunk_io, format=ext)
|
|
|
|
|
155 |
chunks.append({
|
156 |
+
'data': chunk_io.getvalue(),
|
157 |
'start': segment['start'],
|
158 |
'end': segment['end']
|
159 |
})
|
160 |
+
chunk_io.close()
|
|
|
161 |
return chunks
|
162 |
except Exception as e:
|
163 |
st.error(f"Error processing audio in split_audio_by_vad: {str(e)}")
|
164 |
return []
|
165 |
finally:
|
|
|
166 |
if 'audio' in locals():
|
167 |
del audio
|
168 |
if 'samples' in locals():
|
|
|
170 |
|
171 |
@st.cache_data
|
172 |
def transcribe_batch(batch, _transcriber, language=None):
|
173 |
+
"""Transcribes a batch of audio chunks.
|
174 |
+
|
175 |
+
Args:
|
176 |
+
batch (list): A list of audio chunk dictionaries.
|
177 |
+
_transcriber: The transcription model.
|
178 |
+
language (str, optional): The language of the audio (e.g., "en", "es"). Defaults to None (auto-detection).
|
179 |
+
|
180 |
+
Returns:
|
181 |
+
list: A list of dictionaries, each containing the transcription, start, and end time of a chunk.
|
182 |
+
Returns an empty list if an error occurs.
|
183 |
+
"""
|
184 |
transcriptions = []
|
185 |
for i, chunk_data in enumerate(batch):
|
186 |
try:
|
187 |
generate_kwargs = {
|
188 |
"task": "transcribe",
|
189 |
+
"return_timestamps": True,
|
190 |
+
"language": language
|
191 |
}
|
192 |
+
|
|
|
|
|
193 |
transcription = _transcriber(
|
194 |
+
chunk_data['data'],
|
195 |
generate_kwargs=generate_kwargs
|
196 |
)
|
197 |
transcriptions.append({
|
|
|
206 |
|
207 |
# --- Streamlit App ---
|
208 |
def setup_ui():
|
209 |
+
"""Sets up the Streamlit user interface."""
|
210 |
st.title("YouTube Video Transcriber")
|
211 |
+
|
212 |
+
col1, col2, col3, col4 = st.columns(4)
|
213 |
+
with col1:
|
214 |
+
transcribe_option = st.checkbox("Transcribe", value=True)
|
215 |
+
with col2:
|
216 |
+
download_audio_option = st.checkbox("Download Audio", value=False)
|
217 |
+
with col3:
|
218 |
+
download_video_option = st.checkbox("Download Video", value=False)
|
219 |
+
with col4:
|
220 |
+
pass
|
221 |
+
|
222 |
+
video_url = st.text_input("YouTube Video Link:", key="video_url")
|
223 |
+
language = st.text_input("Language (two-letter code, e.g., 'en', 'es', leave empty for auto-detection):", max_chars=2, key="language")
|
224 |
+
batch_size = st.number_input("Batch Size", min_value=1, value=2, key="batch_size")
|
225 |
+
vad_sensitivity = st.slider("VAD Sensitivity", min_value=0.0, max_value=1.0, value=0.1, step=0.05, key="vad_sensitivity")
|
226 |
+
|
227 |
+
# Use session state to manage audio format selection and reset
|
228 |
+
if 'reset_audio_format' not in st.session_state:
|
229 |
+
st.session_state.reset_audio_format = False
|
230 |
+
|
231 |
+
if 'audio_format' not in st.session_state or st.session_state.reset_audio_format:
|
232 |
+
st.session_state.audio_format = "wav" # Default value
|
233 |
+
st.session_state.reset_audio_format = False
|
234 |
+
|
235 |
+
audio_format = st.selectbox("Audio Format", ["wav", "mp3", "ogg", "flac"], key="audio_format_widget", index=["wav", "mp3", "ogg", "flac"].index(st.session_state.audio_format))
|
236 |
+
st.session_state.audio_format = audio_format
|
237 |
+
|
238 |
+
if download_video_option:
|
239 |
+
video_format = st.selectbox("Video Format", ["mp4", "webm"], index=0, key="video_format")
|
240 |
+
else:
|
241 |
+
video_format = "mp4"
|
242 |
+
|
243 |
+
process_button = st.button("Process")
|
244 |
+
|
245 |
+
return video_url, language, batch_size, transcribe_option, download_audio_option, download_video_option, process_button, vad_sensitivity, audio_format, video_format
|
246 |
|
247 |
@st.cache_resource
|
248 |
def initialize_models():
|
249 |
+
"""Initializes the transcription and VAD models."""
|
250 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
251 |
transcriber = load_transcriber(device)
|
252 |
vad_model = load_vad_model()
|
253 |
return transcriber, vad_model
|
254 |
|
255 |
+
def process_transcription(video_url, vad_sensitivity, batch_size, transcriber, vad_model, audio_format, language=None):
|
256 |
+
"""Downloads, processes, and transcribes the audio from a YouTube video.
|
257 |
+
|
258 |
+
Args:
|
259 |
+
video_url (str): The URL of the YouTube video.
|
260 |
+
vad_sensitivity (float): The VAD sensitivity.
|
261 |
+
batch_size (int): The batch size for transcription.
|
262 |
+
transcriber: The transcription model.
|
263 |
+
vad_model: The VAD model.
|
264 |
+
language (str, optional): The language of the audio. Defaults to None.
|
265 |
+
|
266 |
+
Returns:
|
267 |
+
tuple: (full_transcription, audio_data, audio_format, info) or (None, None, None, None) on error.
|
268 |
+
"""
|
269 |
+
audio_data, audio_format, info = download_and_convert_audio(video_url, audio_format)
|
270 |
if not audio_data:
|
271 |
+
return None, None, None, None
|
272 |
+
|
273 |
+
chunks = split_audio_by_vad(audio_data, audio_format, vad_model, vad_sensitivity)
|
274 |
if not chunks:
|
275 |
+
return None, None, None, None
|
276 |
|
277 |
total_chunks = len(chunks)
|
278 |
transcriptions = []
|
279 |
+
progress_bar = st.progress(0)
|
280 |
for i in range(0, total_chunks, batch_size):
|
281 |
batch = chunks[i:i + batch_size]
|
282 |
batch_transcriptions = transcribe_batch(batch, transcriber, language)
|
283 |
transcriptions.extend(batch_transcriptions)
|
284 |
+
progress_bar.progress((i + len(batch)) / total_chunks)
|
285 |
|
286 |
+
progress_bar.empty()
|
287 |
st.success("Transcription complete!")
|
288 |
|
|
|
289 |
full_transcription = ""
|
290 |
for chunk in transcriptions:
|
291 |
start_time = format_seconds(chunk['start'])
|
292 |
end_time = format_seconds(chunk['end'])
|
293 |
full_transcription += f"[{start_time} - {end_time}]: {chunk['text'].strip()}\n\n"
|
294 |
+
|
295 |
+
return full_transcription, audio_data, audio_format, info
|
296 |
|
297 |
def format_seconds(seconds):
|
298 |
"""Formats seconds into HH:MM:SS string."""
|
|
|
300 |
hours, minutes = divmod(minutes, 60)
|
301 |
return f"{int(hours):02}:{int(minutes):02}:{int(seconds):02}"
|
302 |
|
303 |
+
def download_video(video_url, video_format):
|
304 |
+
"""Downloads video from YouTube using yt-dlp."""
|
305 |
+
status_message = st.empty()
|
306 |
+
status_message.text("Downloading video...")
|
307 |
+
try:
|
308 |
+
ydl_opts = {
|
309 |
+
'format': f'bestvideo[ext={video_format}]+bestaudio[ext=m4a]/best[ext={video_format}]/best',
|
310 |
+
'outtmpl': '%(title)s.%(ext)s',
|
311 |
+
'noplaylist': True,
|
312 |
+
'progress_hooks': [lambda d: update_download_progress(d, status_message)],
|
313 |
+
}
|
314 |
+
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
|
315 |
+
info_dict = ydl.extract_info(video_url, download=True)
|
316 |
+
video_filename = ydl.prepare_filename(info_dict)
|
317 |
+
video_title = info_dict.get("title", "video")
|
318 |
+
status_message.text(f"Video downloaded: {video_title}")
|
319 |
+
|
320 |
+
with open(video_filename, 'rb') as video_file:
|
321 |
+
video_bytes = video_file.read()
|
322 |
+
|
323 |
+
os.remove(video_filename)
|
324 |
+
|
325 |
+
return video_bytes, video_filename, info_dict
|
326 |
+
except Exception as e:
|
327 |
+
st.error(f"Error during video download: {e}")
|
328 |
+
return None, None, None
|
329 |
+
|
330 |
+
import random
|
331 |
+
import streamlit as st
|
332 |
+
import io
|
333 |
+
import os
|
334 |
+
from transformers import pipeline
|
335 |
+
import torch
|
336 |
+
import yt_dlp
|
337 |
+
from silero_vad import load_silero_vad, get_speech_timestamps
|
338 |
+
import numpy as np
|
339 |
+
import pydub
|
340 |
+
|
341 |
+
# ... (rest of your code, including model loading, audio functions, etc.)
|
342 |
+
|
343 |
def main():
|
344 |
+
"""Main function to run the Streamlit application."""
|
345 |
+
|
346 |
+
# Initialize session state variables
|
347 |
+
if 'full_transcription' not in st.session_state:
|
348 |
+
st.session_state.full_transcription = None
|
349 |
+
if 'audio_data' not in st.session_state:
|
350 |
+
st.session_state.audio_data = None
|
351 |
+
if 'info' not in st.session_state:
|
352 |
+
st.session_state.info = None
|
353 |
+
if 'video_data' not in st.session_state:
|
354 |
+
st.session_state.video_data = None
|
355 |
+
if 'video_filename' not in st.session_state:
|
356 |
+
st.session_state.video_filename = None
|
357 |
+
|
358 |
transcriber, vad_model = initialize_models()
|
359 |
+
|
360 |
+
# Call setup_ui() to get UI element values
|
361 |
+
video_url, language, batch_size, transcribe_option, download_audio_option, download_video_option, process_button, vad_sensitivity, audio_format, video_format = setup_ui()
|
362 |
+
|
363 |
+
transcription_output = st.empty()
|
364 |
+
if st.session_state.full_transcription:
|
365 |
+
transcription_output.text_area("Transcription:", value=st.session_state.full_transcription, height=300, key=random.random())
|
366 |
+
|
367 |
+
if process_button:
|
368 |
+
st.session_state.full_transcription = None
|
369 |
+
st.session_state.audio_data = None
|
370 |
+
st.session_state.info = None
|
371 |
+
st.session_state.video_data = None
|
372 |
+
st.session_state.video_filename = None
|
373 |
+
st.session_state.reset_audio_format = True
|
374 |
+
|
375 |
if not video_url:
|
376 |
st.error("Please enter a YouTube video link.")
|
377 |
return
|
378 |
+
|
379 |
+
if transcribe_option:
|
380 |
+
st.session_state.full_transcription, st.session_state.audio_data, st.session_state.audio_format, st.session_state.info = process_transcription(video_url, vad_sensitivity, batch_size, transcriber, vad_model, audio_format, language)
|
381 |
+
if st.session_state.full_transcription:
|
382 |
+
transcription_output.text_area("Transcription:", value=st.session_state.full_transcription, height=300, key=random.random())
|
383 |
+
|
384 |
+
if download_audio_option:
|
385 |
+
if st.session_state.audio_data is None or st.session_state.audio_format is None or st.session_state.info is None:
|
386 |
+
st.session_state.audio_data, st.session_state.audio_format, st.session_state.info = download_and_convert_audio(video_url, audio_format)
|
387 |
+
|
388 |
+
if download_video_option:
|
389 |
+
st.session_state.video_data, st.session_state.video_filename, st.session_state.info = download_video(video_url, video_format)
|
390 |
+
|
391 |
+
# Download button logic (moved after setup_ui() call)
|
392 |
+
col1, col2, col3 = st.columns(3)
|
393 |
+
with col1:
|
394 |
+
if st.session_state.full_transcription and transcribe_option:
|
395 |
+
st.download_button(
|
396 |
+
label="Download Transcription (TXT)",
|
397 |
+
data=st.session_state.full_transcription,
|
398 |
+
file_name=f"{st.session_state.info['id']}_transcription.txt",
|
399 |
+
mime="text/plain"
|
400 |
+
)
|
401 |
+
|
402 |
+
with col2:
|
403 |
+
# Now download_audio_option is defined
|
404 |
+
if st.session_state.audio_data is not None and download_audio_option:
|
405 |
+
st.download_button(
|
406 |
+
label=f"Download Audio ({st.session_state.audio_format})",
|
407 |
+
data=st.session_state.audio_data,
|
408 |
+
file_name=f"{st.session_state.info['id']}.{st.session_state.audio_format}",
|
409 |
+
mime=f"audio/{st.session_state.audio_format}"
|
410 |
+
)
|
411 |
+
|
412 |
+
with col3:
|
413 |
+
if st.session_state.video_data is not None and download_video_option:
|
414 |
+
st.download_button(
|
415 |
+
label="Download Video",
|
416 |
+
data=st.session_state.video_data,
|
417 |
+
file_name=st.session_state.video_filename,
|
418 |
+
mime=f"video/{video_format}"
|
419 |
+
)
|
420 |
|
421 |
if __name__ == "__main__":
|
422 |
main()
|