def main(model_folder, example_folder):
config_file, caffemodel_file, net_proto = get_filenames(model_folder)
sbd.SbdConfig(config_file)
ctm_file, pitch_file, energy_file = get_audio_files(example_folder)
# parse ctm_file, pitch_file and energy_file
parser = AudioParser(ctm_file, pitch_file, energy_file)
parser.parse()
classifier = load_audio_classifier(model_folder)
data = classifier.predict_audio(parser)
print(data)
def classifyAudioLexical():
assert request.method == 'POST'
# get example folder
example_folder = os.path.join(route_folder, AUDIO_EXAMPLE_FOLDER, request.form['example'])
ctm_file, pitch_file, energy_file = get_audio_files(example_folder)
# parse ctm_file, pitch_file and energy_file
parser = AudioParser()
talks = parser.parse(ctm_file)
# predict audio
load_config(AUDIO_MODEL_FOLDER, request.form['audio_folder'])
audio_probs = audio_classifier.predict(InputAudio(talks))
# predict lexical
load_config(LEXICAL_MODEL_FOLDER, request.form['lexical_folder'])
input_text = InputText(talks)
lexical_probs = lexical_classifier.predict(input_text)
# get config parameter
(lexical_window_size, lexical_punctuation_pos, pos_tagging) = lexical_classifier.get_lexical_parameter()
(audio_window_size, audio_punctuation_pos) = audio_classifier.get_audio_parameter()
# write audio results
audio_classes = ["NONE", "PERIOD"]
all_audio_probs = convert_probabilities(len(input_text.tokens), audio_punctuation_pos, audio_probs, audio_classes)
file_name = os.path.join(route_folder, AUDIO_EXAMPLE_FOLDER, request.form['example'] + ".result")
resultWriter = ResultWriter(audio_classes)
resultWriter.writeToFile(file_name, input_text.tokens, all_audio_probs)
# fusion
fusion = ThresholdFusion()
fusion.init_parameters(lexical_punctuation_pos, lexical_window_size, audio_punctuation_pos, audio_window_size)
fusion_probs = fusion.fuse(len(input_text.tokens), lexical_probs, audio_probs)
# convert it into json
jsonConverter = JsonConverter(lexical_punctuation_pos, lexical_window_size, audio_punctuation_pos, audio_window_size, pos_tagging)
data = jsonConverter.convert_fusion(input_text.tokens, fusion_probs, lexical_probs, audio_probs)
return json.dumps(data)
parser = argparse.ArgumentParser(description='evaluates the fusion.')
parser.add_argument('ctm_file', help="path to ctm_file", default="evaluation_data/data/tst2011_0.ctm", nargs='?')
parser.add_argument('vectorfile', help='the google news word vector', default='evaluation_data/GoogleNews-vectors-negative300.bin', nargs='?')
parser.add_argument('lexical_model_folder', help="path to lexical models", default="evaluation_data/lexical_models", nargs='?')
parser.add_argument('audio_model_folder', help="path to audio models", default="evaluation_data/audio_models", nargs='?')
parser.add_argument('--release', help="whether to test in release mode", action='store_true')
args = parser.parse_args()
if args.release:
vector = Word2VecFile(args.vectorfile)
else:
vector = None
# get all talks
print("Reading all talks ...")
audio_parser = AudioParser()
talks = audio_parser.parse(args.ctm_file)
# get all lexical models
lexical_models = []
for dirname, dirnames, filenames in os.walk(args.lexical_model_folder):
for subdirname in dirnames:
lexical_models.append(os.path.join(dirname, subdirname))
# get all audio models
audio_models = []
for dirname, dirnames, filenames in os.walk(args.audio_model_folder):
for subdirname in dirnames:
audio_models.append(os.path.join(dirname, subdirname))
