def rate_sentence(videos, s):
# TODO seed ?
print(f'generating combos for "{s}"')
combos = sm.process_sm(s, videos)
print(f'rating "{s}"')
rates = []
for c in combos[:NB_COMBOS]:
rate, wave = sentence_mixing.video_creator.audio.concat_segments(
c.get_audio_phonems())
wave = (sentence_mixing.logic.audio_analysis.resample(
np.sum(wave, axis=1).reshape((-1, )), rate,
16000).astype("int16").copy(order="C"))
decoder = get_decoder()
decoder.start_utt() # begin utterance processing
decoder.process_raw(
wave, False, True
) # process audio data with recognition enabled (no_search = False), as a full utterance (full_utt = True)
decoder.end_utt() # stop utterance processing
recognized_sentence = decoder.hyp().hypstr
rates.append(sentence_distance(s, recognized_sentence))
print(f"returning from {s}")
return rates
def analyze(self, interrupt_callback):
self.set_combos([
Combo(c, self, i) for i, c in enumerate(
sm.process_sm(
self._sentence,
self.project.videos,
interrupt_callback=interrupt_callback,
))
])
if __name__ == "__main__":
parser = argparse.ArgumentParser(description=DESCRIPTION)
parser.add_argument(
"-s", "--seed", default=params.DEFAULT_SEED, help=SEED_HELP,
)
parser.add_argument(
"sentence",
metavar="TARGET_SENTENCE",
action="store",
help=TARGET_SENTENCE_HELP,
)
parser.add_argument(
"config_path",
metavar="CONFIG_PATH",
action="store",
help=CONFIG_PATH_HELP,
)
parser.add_argument(
"video_urls",
metavar="VIDEO_URL",
nargs="+",
action="store",
help=VIDEO_URL_HELP,
)
args = parser.parse_args()
sm.prepare_sm_config_file(args.config_path)
videos = sm.get_videos(args.video_urls)
print(sm.process_sm(args.sentence, videos, args.seed)[0])
def loop_interface(audio_command, video_futures):
total_timestamps = []
total_text = ""
timestamps_buffer = []
timestamps_buffer_sentence = []
sentence = get_sentence(None)
videos = None
while sentence != "":
timestamps = []
combo = None
available_combos = []
edit = False
store = False
valid = False
load_audio_index = None
i = 0
while not valid and not edit:
# Stores previous audio in buffer
if store:
timestamps_buffer.append(timestamps)
timestamps_buffer_sentence.append(sentence)
store = False
if load_audio_index is not None:
timestamps = timestamps_buffer[load_audio_index]
load_audio_index = None
else:
if len(available_combos) == 0:
bad_sentence = True
while bad_sentence:
try:
if videos is None:
print("downloading...")
videos = list(video_futures)[0]
available_combos = sm.process_sm(sentence, videos)
bad_sentence = False
except KeyError as e:
print(e, "not recognized")
sentence = get_sentence(total_text)
except PhonemError as e:
print(
e,
"Try to change your sentence or add more videos.",
)
sentence = get_sentence(total_text)
except TokenAmbiguityError as e:
print(
e,
"Please change this word",
)
sentence = get_sentence(total_text)
combo = available_combos.pop(0)
timestamps = combo.get_audio_phonems()
print(combo_displayer(combo))
concat_wav(AUDIO_FILE_PATH, timestamps)
os.system(audio_command.format(AUDIO_FILE_PATH))
if timestamps_buffer_sentence:
print("Stashed audios:")
for i, stashed_sentence in enumerate(
timestamps_buffer_sentence):
print(i, ".", stashed_sentence)
print("")
line = input(
"Enter 'y' to validate, 'e' to edit the sentence, 's' to store this audio in the buffer, 'l' + index for loading previously stored audio, otherwise just press enter: "
)
valid = line == "y"
edit = line == "e"
store = line == "s"
if line.startswith("l "):
index = line.split(" ")[1]
if index.isdigit():
index = int(index)
if -1 < index < len(timestamps_buffer):
load_audio_index = index
print(load_audio_index)
i += 1
clear_screen()
if not edit:
total_timestamps.extend(timestamps)
total_text += "\n" + sentence
save(total_timestamps, total_text, name="video.json")
sentence = get_sentence(total_text)
clear_screen()
return total_timestamps, total_text, videos
parser.add_argument(
"sentence",
metavar="TARGET_SENTENCE",
action="store",
help=TARGET_SENTENCE_HELP,
)
parser.add_argument(
"config_path",
metavar="CONFIG_PATH",
action="store",
help=CONFIG_PATH_HELP,
)
parser.add_argument(
"video_urls",
metavar="VIDEO_URL",
nargs="+",
action="store",
help=VIDEO_URL_HELP,
)
args = parser.parse_args()
out_dir = args.out_dir
sm.prepare_sm_config_file(args.config_path)
videos = sm.get_videos(args.video_urls)
combos = sm.process_sm(args.sentence, videos, args.seed)
print(combo_displayer(combos[0]))
sentence_mixing.video_creator.audio.concat_wav(
cli_interface.AUDIO_FILE_PATH, combos[0].get_audio_phonems())