def analyse_sound(path): x, sr = librosa.load(path) td = librosa.get_duration(x) print("Time duration of audio: ", td) x = np.reshape(x, [1, -1, 1, 1]) print("Shape of input waveform: ", x.shape) # Setup visible device os.environ["CUDA_VISIBLE_DEVICES"] = args.cuda_device # Load pre-trained model G_name = './models/sound8.npy' param_G = np.load(G_name, encoding='latin1').item() # Init. Session sess_config = tf.ConfigProto() sess_config.allow_soft_placement = True sess_config.gpu_options.allow_growth = True with tf.Session(config=sess_config) as session: # Build model model = Model(session, config=local_config, param_G=param_G) init = tf.global_variables_initializer() session.run(init) model.load() features = ef.extract_feat(model, x, local_config) print("Shape of feature: ", features.shape)
def main(): args = parse_args() local_config['phase'] = args.phase # Setup visible device os.environ["CUDA_VISIBLE_DEVICES"] = args.cuda_device # Make path if not os.path.exists(args.outpath): os.mkdir(args.outpath) # Load pre-trained model param_G = np.load(local_config['param_g_dir'], encoding='latin1').item() \ if args.phase in ['finetune', 'extract'] \ else None # Init. Session sess_config = tf.ConfigProto() sess_config.allow_soft_placement = True sess_config.gpu_options.allow_growth = True with tf.Session(config=sess_config) as session: # Build model model = Model(session, config=local_config, param_G=param_G) if args.phase in ['train', 'finetune']: # Training phase model.train() elif args.phase == 'extract': # import when we need from extract_feat import extract_feat # Feature extractor #sound_sample = np.reshape(np.load('./data/demo.npy', encoding='latin1'), [local_config['batch_size'], -1, 1, 1]) import librosa audio_path = './data/demo.mp3' sound_sample, _ = load_audio(audio_path) sound_sample = preprocess(sound_sample, config=local_config) output = extract_feat(model, sound_sample, args)
def extract_feat_from_link(model, audio_path, config): sound_sample = load_from_link(audio_path) features = extract_feat(model, sound_sample) return features
# coding: utf-8 import pandas as pd from extract_feat import extract_feat dataset = pd.read_csv("extract_dataset/extract_feat_dataset_15.csv") feat = dataset[dataset.time >= "t5"][[ "user_id", "item_id", "item_category", "label" ]] feat = feat.drop_duplicates() feat = extract_feat(feat, dataset, '2014-12-15 00', '2014-12-16 00', '2014-12-14 00', '2014-12-12 00', '2014-12-09 00', '2014-11-18 00') feat.to_csv("feature/feature1.csv", index=False)
# coding: utf-8 import pandas as pd from extract_feat import extract_feat dataset = pd.read_csv("extract_dataset/extract_feat_dataset_18.csv") feat = dataset[dataset.time >= "t5"][["user_id", "item_id", "item_category"]] feat = feat.drop_duplicates() feat = extract_feat(feat, dataset, '2014-12-18 00', '2014-12-19 00', '2014-12-17 00', '2014-12-15 00', '2014-12-12 00', '2014-11-18 00') feat.to_csv("feature/feature4.csv",index=False)
# coding: utf-8 import pandas as pd from extract_feat import extract_feat dataset = pd.read_csv("extract_dataset/extract_feat_dataset_17.csv") feat = dataset[dataset.time >= "t5"][[ "user_id", "item_id", "item_category", "label" ]] feat = feat.drop_duplicates() feat = extract_feat(feat, dataset, '2014-12-17 00', '2014-12-18 00', '2014-12-16 00', '2014-12-14 00', '2014-12-11 00', '2014-11-18 00') feat.to_csv("feature/feature3.csv", index=False)