Audio-Event-Detection/utils.py at main · DibyojyotiS/Audio-Event-Detection · GitHub

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
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
import librosa
import numpy as np
import pdb
import string
from Levenshtein import distance

def wav2feat(wavfile):
    '''
    Input: audio wav file name
    Output: Magnitude spectrogram
    '''
    x, Fs = librosa.load(wavfile, sr=44100, mono=True)
    hop = int(0.01 * Fs) # 10ms
    win = int(0.02 * Fs) # 20ms
    X = librosa.stft(x, n_fft=1024, hop_length=hop, win_length=win, window='hann', center=True, pad_mode='reflect')
    return np.abs(X)

def wavs2feat(wavfiles):
    '''
    Concatenate the audio files listed in wavfiles
    Input: list of audio wav file names
    Output: Magnitude spectrogram of concatenated wav
    '''
    x = []
    for wf in wavfiles:
        x1, Fs = librosa.load(wf, sr=44100, mono=True)
        x.append(x1)
    x = np.hstack(x)
    hop = int(0.01 * Fs) # 10ms
    win = int(0.02 * Fs) # 20ms
    X = librosa.stft(x, n_fft=1024, hop_length=hop, win_length=win, window='hann', center=True, pad_mode='reflect')
    return np.abs(X)

def read_csv(filename):
    id_label = {}
    with open(filename,'r') as fid:
        for line in fid: # '176787-5-0-27.wav,engine_idling\n'
            tokens = line.strip().split(',') # ['176787-5-0-27.wav', 'engine_idling']
            id_label[tokens[0]] = tokens[1]
    return id_label

def editDistance(gt, est):
    '''both are lists of labels
    E.g. gt is "dog_bark-street_music-engine_idling"
    E.g. est is "street_music-engine_idling"
    '''
    gttokens = gt.split('-')
    esttokens = est.split('-')
    # Map token to char
    tokenset = list(set(gttokens+esttokens)) # ['dog_bark', 'siren', 'street_music', 'engine_idling']
    token_char = {}
    for i in range(len(tokenset)):
        token_char[tokenset[i]] = string.ascii_uppercase[i]  # {'dog_bark': 'A', 'siren': 'B', 'street_music': 'C', 'engine_idling': 'D'}
    # convert gt and est to strings
    gtstr = [token_char[t] for t in gttokens]
    gtstr = ''.join(gtstr)  # 'BCA'
    eststr = [token_char[t] for t in esttokens]
    eststr = ''.join(eststr)  #
    # Compare
    editdist = distance(gtstr, eststr) # 1
    score = 1 - editdist/len(gtstr)
    return editdist, score

def evals(gtcsv, estcsv, taskid):
    gt_id_label = read_csv(gtcsv)
    est_id_label = read_csv(estcsv)
    score = 0
    for id in est_id_label:
        if taskid==1:
            if est_id_label[id] == gt_id_label[id]:
                score += 1
        elif taskid==2:
            _, ss = editDistance(gt_id_label[id], est_id_label[id])
            score += ss
        else:
            pdb.set_trace()
            assert False, ["taskid not correct; it is", taskid]
    avgScore = score/len(est_id_label)
    return avgScore

if __name__=="__main__":
    wavs = ['../shared_train/audio_train/180937-7-3-27.wav']
    wavs2feat(wavs)
#     # wavfiles = ['../shared_train/audio_train/180937-7-3-27.wav','../shared_train/audio_train/180937-7-3-27.wav']
#     # X = wavs2feat(wavfiles)
#     # eval('test_task1/labels.csv', 'test_task1/est.csv', 1)
#     editDistance("dog_bark-street_music-engine_idling",
#         "siren-street_music-engine_idling")