def compose_from_transcript(files, search, searchtype):
"""Takes transcripts created by audiogrep/pocketsphinx, a search and search type
and returns a list of timestamps for creating a supercut"""
final_segments = []
if searchtype in ['re', 'word', 'franken', 'fragment']:
if searchtype == 're':
searchtype = 'sentence'
segments = audiogrep.search(search, files, mode=searchtype, regex=True)
for seg in segments:
seg['file'] = seg['file'].replace('.transcription.txt', '')
seg['line'] = seg['words']
final_segments.append(seg)
elif searchtype in ['hyper', 'pos']:
for s in audiogrep.convert_timestamps(files):
for w in s['words']:
if search_line(w[0], search, searchtype):
seg = {
'file': s['file'].replace('.transcription.txt',''),
'line': w[0],
'start': float(w[1]),
'end': float(w[2])
}
final_segments.append(seg)
return final_segments
def compose_from_transcript(files, search, searchtype):
"""Takes transcripts created by audiogrep/pocketsphinx, a search and search type
and returns a list of timestamps for creating a supercut"""
final_segments = []
if searchtype in ['re', 'word', 'franken', 'fragment']:
if searchtype == 're':
searchtype = 'sentence'
segments = audiogrep.search(search, files, mode=searchtype, regex=True)
for seg in segments:
seg['file'] = seg['file'].replace('.transcription.txt', '')
seg['line'] = seg['words']
final_segments.append(seg)
elif searchtype in ['hyper', 'pos']:
for s in audiogrep.convert_timestamps(files):
for w in s['words']:
if search_line(w[0], search, searchtype):
seg = {
'file': s['file'].replace('.transcription.txt', ''),
'line': w[0],
'start': float(w[1]),
'end': float(w[2])
}
final_segments.append(seg)
return final_segments
def get_ngrams(inputfile, n=1, use_transcript=False, use_vtt=False):
'''
Get ngrams from a text
Sourced from:
https://gist.github.com/dannguyen/93c2c43f4e65328b85af
'''
words = []
if use_transcript:
for s in audiogrep.convert_timestamps(inputfile):
for w in s['words']:
words.append(w[0])
elif use_vtt:
vtts = get_vtt_files(inputfile)
for vtt in vtts:
with open(vtt['vtt'], 'r') as infile:
sentences = parse_auto_sub(infile.read())
for s in sentences:
for w in s['words']:
words.append(w['word'])
else:
text = ''
srts = get_subtitle_files(inputfile)
for srt in srts:
lines = clean_srt(srt)
if lines:
for timespan in lines.keys():
line = lines[timespan].strip()
text += line + ' '
words = re.split(r'[.?!,:\"]+\s*|\s+', text)
ngrams = zip(*[words[i:] for i in range(n)])
return ngrams
def extract_words(files,
padding,
use_uuid=False,
confidence=0.0,
output_directory='extracted_words'):
''' Extracts individual words form files and exports them to individual files. '''
segments = []
for s in audiogrep.convert_timestamps(files):
for w in s['words']:
if w[3] < confidence:
continue
try:
float(w[1])
except:
continue
seg = {
'word': w[0],
'file': s['file'].replace('.transcription.txt', ''),
'line': w[0],
'start': float(w[1]),
'end': float(w[2])
}
segments.append(seg)
composition = segments
# apply padding and sync
for c in composition:
c['start'] = c['start'] - padding
c['end'] = c['end'] + padding
all_filenames = set([c['file'] for c in composition])
videofileclips = dict([(f, VideoFileClip(f)) for f in all_filenames])
cut_clips = []
for c in composition:
try:
clip = videofileclips[c['file']]
start = max(0, c['start'] - padding)
end = min(c['end'] + padding, clip.duration)
subclip = clip.subclip(start, end)
cut_clips.append((c['word'], subclip))
except Exception as e:
continue
from collections import defaultdict
wc = defaultdict(int)
for word, clip in cut_clips:
print word, clip
if use_uuid:
word_id = str(uuid.uuid1())
else:
wc[word] += 1
word_id = str(wc[word])
path = output_directory + "/" + word
if not os.path.exists(path):
os.makedirs(path)
clip.to_videofile(path + "/" + word_id + ".mp4",
codec="libx264",
temp_audiofile='temp-audio.m4a',
audio_codec='aac',
remove_temp=True,
fps=23)
def test_convert_timestamps():
filename = os.path.join(os.path.dirname(__file__), 'data/test.mp3')
sentences = audiogrep.convert_timestamps([filename])
words = {}
for sentence in sentences:
for word in sentence['words']:
words[word[0]] = True
assert 'fashion' in words
assert len(sentences) == 9
import sys
import moviepy.editor as mp
import audiogrep
videofile = sys.argv[1]
# get the sentence timestamps
sentences = audiogrep.convert_timestamps([videofile])
timestamps = []
for sentence in sentences:
timestamps += sentence['words']
# alphabetize the list
timestamps.sort(key=lambda x: x[0])
# we could limit how many clips here
# words = words[100:200]
original_video = mp.VideoFileClip(videofile)
clips = []
for timestamp in timestamps:
word = timestamp[0]
start = float(timestamp[1])
end = float(timestamp[2])
# skip this clip if the word is shorter than 5 characters