-
Notifications
You must be signed in to change notification settings - Fork 1
/
peakshift.py
172 lines (129 loc) · 5.2 KB
/
peakshift.py
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
import sys
import numpy as np
import scipy.signal
import matplotlib.pyplot as plt
import soundfile
import pvc
# TODO: This doesn't work correctly, and grabs small peaks rather than the larger-scale formants
if __name__ == "__main__":
block_size = 4096
n_blocks = 4
vol_thresh = 0.01
peak_border = 4
# if len(sys.argv) < 5:
# print("Usage: {} <in_filename> <out_filename> <length_mult> <pitch_mult> [block_size={}] [n_blocks={}]".format(
# sys.argv[0], block_size, n_blocks
# ))
# sys.exit()
in_filename = "audio/test.flac" # sys.argv[1]
out_filename = "audio/test_formants.flac" # sys.argv[2]
pitch_formant = [1, 1, 1, 1, 1] # float(sys.argv[4])
pitch_shift = 1.5
if len(sys.argv) >= 6:
block_size = int(sys.argv[5])
if len(sys.argv) >= 7:
n_blocks = int(sys.argv[6])
in_shift = block_size // n_blocks
out_shift = in_shift
in_file = soundfile.SoundFile(in_filename)
rate = in_file.samplerate
n_blocks = int(np.ceil(in_file.frames / in_shift))
out_length = int(n_blocks * out_shift + block_size)
# print("from", in_file.frames, "to", out_length)
out_data = np.zeros(out_length)
t_pvc = pvc.PhaseVocoder(rate, block_size)
indices = np.arange(t_pvc.fft_size)
print(t_pvc.fft_size, n_blocks)
spectrogram = np.zeros((t_pvc.fft_size, n_blocks))
out_spectrogram = np.zeros((t_pvc.fft_size, n_blocks))
p_x = []
p_y = []
index = 0
t = 0
for block in in_file.blocks(blocksize=block_size, overlap=(block_size - in_shift)):
if block.shape[0] != block_size:
block = np.pad(block, ((0, block_size - block.shape[0]), (0, 0)))
magnitude, phase, frequency = t_pvc.analyze(block, in_shift)
spectrogram[:, index] = np.log(magnitude)
if np.max(np.abs(block)) >= vol_thresh:
peaks, properties = scipy.signal.find_peaks(
magnitude, prominence=8
) # , height=np.mean(magnitude))
print(peaks)
# plt.plot(t_pvc.freq, magnitude)
# plt.scatter(t_pvc.freq[peaks], magnitude[peaks])
# plt.show()
target_peaks = len(pitch_formant)
n_peaks = len(peaks)
p_x.extend((index,) * len(peaks))
p_y.extend(peaks)
# if n_peaks > target_peaks:
# peaks2 = peaks[:target_peaks]
last_peak = 0
end = 0
first_start = 0
layers = []
for i, peak in enumerate(peaks):
if i >= target_peaks:
break
start = 0
if i == 0:
start = max(peak - peak_border, 0)
first_start = start
else:
start = np.argmin(magnitude[last_peak:peak]) + last_peak
if i < n_peaks - 1:
next_peak = peaks[i + 1]
end = np.argmin(magnitude[peak:next_peak]) + peak - 1
else:
end = min(peak + peak_border, magnitude.size - 1)
layers.append((start, end, pitch_formant[i]))
last_peak = peak
layers.append((0, first_start, pitch_shift))
layers.append((end, magnitude.size - 1, pitch_shift))
mask = np.zeros(magnitude.size)
mag_sum = np.zeros(magnitude.size)
freq_sum = np.zeros(magnitude.size)
for layer in layers:
start = layer[0]
end = layer[1]
pitch = layer[2]
if end > start:
peak_mag = np.zeros(magnitude.size)
peak_freq = np.zeros(magnitude.size)
peak_mask = np.zeros(magnitude.size)
peak_mask[start:end] = 1
peak_mag[start:end] = magnitude[start:end]
peak_freq[start:end] = frequency[start:end]
peak_mag = np.interp(indices / pitch, indices, peak_mag, 0, 0)
peak_freq = (
np.interp(indices / pitch, indices, peak_freq, 0, 0) * pitch
)
peak_mask = np.interp(indices / pitch, indices, peak_mask, 0, 0)
mag_sum += peak_mag
freq_sum += peak_freq
mask += peak_mask
to_modify = mask > 0
freq_sum[to_modify] /= mask[to_modify]
magnitude = mag_sum
frequency = freq_sum
out_spectrogram[:, index] = np.log(magnitude)
out_block = t_pvc.synthesize(magnitude, frequency, out_shift)
out_data[t : t + block_size] += out_block
t += out_shift
index += 1
in_file.close()
out_data = out_data / np.max(np.abs(out_data))
soundfile.write(out_filename, out_data, rate)
plt.subplot(211)
ax1 = plt.imshow(
spectrogram, origin="lower"
) # , extent=[0,n_blocks*in_shift,0,t_pvc.freq[-1]]) #, aspect=0.01)
plt.scatter(p_x, p_y)
plt.ylim((0, 100))
plt.subplot(212)
ax2 = plt.imshow(out_spectrogram, origin="lower")
plt.ylim((0, 100))
# plt.yticks(t_pvc.freq)
# plt.xticks(np.arange(n_blocks)*in_shift/rate)
plt.show()