def merge_subtitle(sub_a, sub_b, delta):
out = WebVTTFile()
intervals = [item.start.ordinal for item in sub_a]
intervals.extend([item.end.ordinal for item in sub_a])
intervals.extend([item.start.ordinal for item in sub_b])
intervals.extend([item.end.ordinal for item in sub_b])
intervals.sort()
j = k = 0
for i in xrange(1, len(intervals)):
start = WebVTTTime.from_ordinal(intervals[i - 1])
end = WebVTTTime.from_ordinal(intervals[i])
if (end - start) > delta:
text_a, j = find_subtitle(sub_a, start, end, j)
text_b, k = find_subtitle(sub_b, start, end, k)
text = join_lines(text_a, text_b)
if len(text) > 0:
item = WebVTTItem(0, start, end, text)
out.append(item)
out.clean_indexes()
return out
def merge_subtitle(sub_a, sub_b, delta):
out = WebVTTFile()
intervals = [item.start.ordinal for item in sub_a]
intervals.extend([item.end.ordinal for item in sub_a])
intervals.extend([item.start.ordinal for item in sub_b])
intervals.extend([item.end.ordinal for item in sub_b])
intervals.sort()
j = k = 0
for i in xrange(1, len(intervals)):
start = WebVTTTime.from_ordinal(intervals[i - 1])
end = WebVTTTime.from_ordinal(intervals[i])
if (end-start) > delta:
text_a, j = find_subtitle(sub_a, start, end, j)
text_b, k = find_subtitle(sub_b, start, end, k)
text = join_lines(text_a, text_b)
if len(text) > 0:
item = WebVTTItem(0, start, end, text)
out.append(item)
out.clean_indexes()
return out
def main(options):
# Ensure ffmpeg is around
if not run_ffmpeg(['-version']):
log.error(
"ffmpeg needs to be available to strip audio from the video file.")
exit(1)
with NamedTemporaryFile(delete=True) as vid_file:
log.info("Downloading %s - this might take a while." % options.vid_url)
response = get(options.vid_url, stream=True)
total_length = response.headers.get("content-length")
if total_length is None: # no content length header
log.info("Unknown length - can't predict how long this will take.")
f.write(response.content)
else:
bar = ProgressBar(max_value=int(total_length))
dl = 0
for data in response.iter_content(chunk_size=DOWNLOAD_CHUNK_SIZE):
dl += len(data)
vid_file.write(data)
vid_file.flush()
bar.update(dl)
log.info("Download done. Stripping audio.")
(wav_file, wav_file_name) = mkstemp('.wav')
result = run_ffmpeg([
"-y", "-i", vid_file.name, "-vn", "-acodec", "pcm_s16le", "-ar",
"16000", "-ac", "1", wav_file_name
])
if not result:
close(wav_file)
log.error("ffmpeg failed. Bailing.")
exit(1)
fs, audio = wav.read(wav_file_name)
close(wav_file)
log.info("Will write VTT to %s" % options.output)
# Make sure the WAV is to code...
log.info("Loading up WAV file...")
if fs != 16000:
log.error("Only 16000hz WAV files are usable.")
exit(1)
total_samples = len(audio)
duration_hours, duration_minutes, duration_seconds = sample_index_to_time(
len(audio))
log.info("Approximate duration: %d:%02d:%02d" %
(duration_hours, duration_minutes, duration_seconds))
# Let's load up DeepSpeech and get it ready.
log.info("Loading pre-trained DeepSpeech model...")
root_model_dir = path.join(options.deepspeech_model_dir, MODEL_DIR)
model = path.join(root_model_dir, MODEL_FILE)
alphabet = path.join(root_model_dir, MODEL_ALPHABET)
lang_model = path.join(root_model_dir, MODEL_LANG_MODEL)
trie = path.join(root_model_dir, MODEL_TRIE)
deepspeech = Model(model, N_FEATURES, N_CONTEXT, alphabet, BEAM_WIDTH)
log.info("Done loading model.")
log.info("Loading language model...")
deepspeech.enableDecoderWithLM(alphabet, lang_model, trie, LM_WEIGHT,
WORD_COUNT_WEIGHT, VALID_WORD_COUNT_WEIGHT)
log.info("Done loading model.")
playhead = 0
out = WebVTTFile()
bar = ProgressBar(max_value=total_samples)
while playhead < (total_samples - 1):
end_point = min(playhead + AUDIO_SEGMENT_SAMPLES, (total_samples - 1))
segment = audio[playhead:end_point]
inference = deepspeech.stt(segment, fs)
log.debug("Inferred: %s" % inference)
start_hours, start_minutes, start_seconds = sample_index_to_time(
playhead)
playhead = end_point
end_hours, end_minutes, end_seconds = sample_index_to_time(playhead)
if not inference or inference == "ah":
continue
for search, replace in INFERENCE_REPLACEMENTS.iteritems():
inference = sub(r"\b" + search + r"\b", replace, inference)
inference = fill(inference, width=MAX_CAPTION_WIDTH)
start = WebVTTTime(start_hours, start_minutes, start_seconds)
end = WebVTTTime(end_hours, end_minutes, end_seconds)
item = WebVTTItem(0, start, end, inference)
out.append(item)
bar.update(playhead)
out.save(options.output, encoding="utf-8")
out.clean_indexes()
out.save(options.output, encoding="utf-8")