def create_silent_video(self, youtube=None, preview=False):
log("Creating silent video", prnt=True)
base_path = str(self.podcast.id) + '/' + str(self.episode.id)
video_path = base_path + '.mp4'
with NamedTemporaryFile(suffix='.mp4') as tf:
with NamedTemporaryFile(suffix='.mp3') as _audio:
self.videosyncepisode.update_sync_status({
'event':
'DOWNLOAD_AUDIO',
'timestamp':
arrow.utcnow().isoformat()
})
r = requests.get(self.audio_url, stream=True)
for chunk in r.iter_content(chunk_size=4096):
_audio.write(chunk)
_audio.seek(0)
song = AudioSegment.from_file(File(_audio), format='mp3')
sound_length = math.ceil(len(song) / 1000)
self.videosyncepisode.update_sync_status({
'event':
'CREATE_VIDEO',
'timestamp':
arrow.utcnow().isoformat()
})
clip = ImageClip(self.videosyncepisode.artwork_path,
duration=sound_length)
clip.write_videofile(tf.name, fps=1, audio=_audio.name)
self.videosyncepisode.update_sync_status({
'event':
'UPLOAD_VIDEO',
'timestamp':
arrow.utcnow().isoformat()
})
video_id = self.upload_video(youtube, tf.name)
return video_id
return None
def image_to_video(self, image_path: str, destination: str,
duration: float) -> None:
video = ImageClip(image_path)
video = video.set_duration(duration)
video.write_videofile(destination, fps=24)
def generate_video_for_id(id):
et_dataset_location = f'{reflacx_dataset_location}/main_data/'
mimic_dataset_location = mimiccxr_images_location
table_et_pt1 = pd.read_csv(f'{et_dataset_location}/{id}/fixations.csv')
table_text = pd.read_csv(
f'{et_dataset_location}/{id}/timestamps_transcription.csv')
main_table = pd.read_csv(
f'{et_dataset_location}/metadata_phase_{id[1]}.csv')
image_filepath = main_table[main_table['id'] == id]['image'].values
assert (len(image_filepath) == 1)
max_time_fixation = max(table_et_pt1['timestamp_end_fixation'].values)
max_time_text = max(table_text['timestamp_end_word'].values)
dicom_array, _, _ = open_dicom(
f'{mimic_dataset_location}/{image_filepath[0]}')
dicom_array = dicom_array * 255
from skimage.transform import resize
dicom_array = resize(dicom_array,
(int(dicom_array.shape[0] * scale_video),
int(dicom_array.shape[1] * scale_video)),
anti_aliasing=True)
dicom_array = dicom_array.astype(np.uint8)
#generat ea clip with the original dicom as every frame
my_clip = ImageClip(np.stack(
(dicom_array, ) * 3,
axis=-1)).set_duration(max([max_time_fixation,
max_time_text])).set_fps(fps)
#modify every frame of the video according to the fixations and transcription tables
my_clip = fl(
my_clip,
lambda get_frame, t: scroll(get_frame, t, table_et_pt1, table_text))
#generate the audio from the timestamped transcription
full_audio = AudioSegment.empty()
previous_end = 0
for _, row in table_text.iterrows():
# if start and end of the word are at the same time, it was not captured by the original transcription, so we do not use it in the audio, only in subtitle
if row['timestamp_start_word'] == row['timestamp_end_word']:
continue
print(row['word'])
# text to speech
tts = SaveTTSFile('./create_video_temp.wav')
tts.start(
row['word'].replace('.', 'period').replace(',', 'comma').replace(
'/', 'slash'), row['timestamp_start_word'],
row['timestamp_end_word'])
for i in range(10):
if not os.path.exists('./create_video_temp.wav'):
time.sleep(1)
else:
break
if i > 10:
assert (False)
del (tts)
# add silence between words if they did not end/start at the same time
if row['timestamp_start_word'] > previous_end:
full_audio += AudioSegment.silent(
duration=(row['timestamp_start_word'] - previous_end) * 1000)
print(full_audio.duration_seconds)
print(row['timestamp_start_word'])
assert (abs(full_audio.duration_seconds - row['timestamp_start_word'])
< 0.005)
#change the duration of the word sound to the duration it took for the radiologist to say it
word_audio = AudioSegment.from_file('./create_video_temp.wav',
format="wav")
word_audio = stretch_audio(
word_audio, './create_video_temp.wav',
word_audio.duration_seconds /
(row['timestamp_end_word'] - row['timestamp_start_word']))
os.remove('./create_video_temp.wav')
full_audio += word_audio
assert (abs(full_audio.duration_seconds - row['timestamp_end_word']) <
0.005)
previous_end = row['timestamp_end_word']
full_audio.export("./create_video_temp.wav", format="wav")
audio_background = mpe.AudioFileClip('./create_video_temp.wav')
my_clip = my_clip.set_audio(audio_background)
my_clip.write_videofile(f"movie_{id}.mp4",
audio_codec='aac',
codec="libx264",
temp_audiofile='temp-audio.m4a',
remove_temp=True,
fps=30,
bitrate="5000k")
os.remove('./create_video_temp.wav')
