-
Notifications
You must be signed in to change notification settings - Fork 0
/
Readme.txt
47 lines (31 loc) · 1.36 KB
/
Readme.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
#dataset_name = ['AMG_1608', 'CH_818']
#feature_name = ['melSpec_lw', 'pitch+lw', 'auto']
audio_path = '/mnt/data/Wayne/Dataset'
audio_npy_path = '/mnt/data/Wayne/#dataset_name/Train_X@220505Hz.npy'
feature_path = 'mnt/data/Wayne/#dataset_name/Train_X@220505Hz_#feature_name.npy'
0. Reformat the audio files
# Tansfer audio files into 22.05KHz and clip into 29 seconds
functions/Transfer_funcs.audio_to_wav()
# Generate combined audio files and labels for training
functions/Transfer_funcs.wav_to_npy()
1. Extract features
# Extract autocorrelation-based tempogram
feature_extraction/MATLAB-Tempogram-Toolbox_1.0/test_TempogramToolbox.m
# Extract pitch salience representation
feature_extraction/predict_on_audio.py
ex: python predict_on_audio.py '/mnt/data/Wayne/Dataset/AMG_1608_wav@22050Hz' 'pitch' '/mnt/data/Wayne/AMG_1608_pitch+lw@22050Hz'
# Extract log-mel spectrogram
functions/model_structure.extract_melspec()
# Transfer different features into the same npy format
feature_extraction/Extract_features.py
2. Pre-training
# Within-dataset experiment
Training/AMG1608_CV.py
# Cross-dataset experiment
Training/AMG1608_train.py
3. Adversarial discriminative domain adaptation
# Cross-dataset experiment
Training/WADDA_S_AMG1608_T_CH818.py
4. Testing
# Unconfirmed
Testing/Multi_fusion_pred_S_AMG1608_T_CH818_find_by_loss.py