File size: 5,205 Bytes
c7e47b2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
"""Generate activation confidence annotations.
"""
from __future__ import division
import scipy.signal
import numpy as np
import librosa
import medleydb
import os
import argparse


def compute_activation_confidence(mtrack, win_len=4096, lpf_cutoff=0.075,
                                 theta=0.15, var_lambda=20.0,
                                 amplitude_threshold=0.01):
    """Create the activation confidence annotation for a multitrack. The final
    activation matrix is computed as:
        `C[i, t] = 1 - (1 / (1 + e**(var_lambda * (H[i, t] - theta))))`
    where H[i, t] is the energy of stem `i` at time `t`

    Parameters
    ----------
    mtrack : MultiTrack
        Multitrack object
    win_len : int, default=4096
        Number of samples in each window
    lpf_cutoff : float, default=0.075
        Lowpass frequency cutoff fraction
    theta : float
        Controls the threshold of activation.
    var_labmda : float
        Controls the slope of the threshold function.
    amplitude_threshold : float
        Energies below this value are set to 0.0

    Returns
    -------
    C : np.array
        Array of activation confidence values shape (n_conf, n_stems)
    stem_index_list : list
        List of stem indices in the order they appear in C

    """
    H = []
    stem_index_list = []

    # MATLAB equivalent to @hanning(win_len)
    win = scipy.signal.windows.hann(win_len + 2)[1:-1]

    for stem_idx, track in mtrack.stems.items():
        audio, rate = librosa.load(track.audio_path, sr=44100, mono=True)
        H.append(track_energy(audio.T, win_len, win))
        stem_index_list.append(stem_idx)

    # list to numpy array
    H = np.array(H)

    # normalization (to overall energy and # of sources)
    E0 = np.sum(H, axis=0)
    
    H = len(mtrack.stems) * H / np.max(E0)
    # binary thresholding for low overall energy events
    H[:, E0 < amplitude_threshold] = 0.0

    # LP filter
    b, a = scipy.signal.butter(2, lpf_cutoff, 'low')
    H = scipy.signal.filtfilt(b, a, H, axis=1)

    # logistic function to semi-binarize the output; confidence value
    C = 1.0 - (1.0 / (1.0 + np.exp(np.dot(var_lambda, (H - theta)))))

    # add time column
    time = librosa.core.frames_to_time(
        np.arange(C.shape[1]), sr=rate, hop_length=win_len // 2
    )

    # stack time column to matrix
    C_out = np.vstack((time, C))
    return C_out.T, stem_index_list


def track_energy(wave, win_len, win):
    """Compute the energy of an audio signal

    Parameters
    ----------
    wave : np.array
        The signal from which to compute energy
    win_len: int
        The number of samples to use in energy computation
    win : np.array
        The windowing function to use in energy computation

    Returns
    -------
    energy : np.array
        Array of track energy

    """
    hop_len = win_len // 2

    wave = np.lib.pad(
        wave, pad_width=(win_len-hop_len, 0), mode='constant', constant_values=0
    )

    # post padding
    wave = librosa.util.fix_length(
        wave, int(win_len * np.ceil(len(wave) / win_len))
    )

    # cut into frames
    wavmat = librosa.util.frame(wave, frame_length=win_len, hop_length=hop_len)

    # Envelope follower
    wavmat = hwr(wavmat) ** 0.5  # half-wave rectification + compression

    return np.mean((wavmat.T * win), axis=1)


def hwr(x):
    """ Half-wave rectification.

    Parameters
    ----------
    x : array-like
        Array to half-wave rectify

    Returns
    -------
    x_hwr : array-like
        Half-wave rectified array

    """
    return (x + np.abs(x)) / 2


def write_activations_to_csv(mtrack, activations, stem_index_list,
                             overwrite_existing=False):
    """Write computed activations to the multitrack's activation confidence
    file.

    Parameters
    ----------
    mtrack : MultiTrack
        Multitrack object
    activations : np.array
        Matrix of stem activations
    stem_index_list : list
        List of stem indices
    overwrite_existing : bool, default=False
        If True, overwrites an existing activation confidence file

    """
    stem_str = ",".join(
        ["S%02d" % stem_idx for stem_idx in stem_index_list]
    )
    if not os.path.exists(mtrack.activation_conf_fpath) or overwrite_existing:
        np.savetxt(
            mtrack.activation_conf_fpath,
            activations,
            header='time,{}'.format(stem_str),
            delimiter=',',
            fmt='%.4f',
            comments=''
        )


def main(args):
    """Main function. Computes the activation confidence annotation for a given
    multitrack id.
    """
    mtrack = medleydb.MultiTrack(args.track_id)
    if os.path.exists(mtrack.activation_conf_fpath):
        return True

    activations, index_list = compute_activation_confidence(mtrack)
    write_activations_to_csv(mtrack, activations, index_list)


if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="")
    parser.add_argument("track_id",
                        type=str,
                        default="LizNelson_Rainfall",
                        help="MedleyDB track id. Ex. LizNelson_Rainfall")
    main(parser.parse_args())