-
Notifications
You must be signed in to change notification settings - Fork 0
/
hoac_decoder.py
175 lines (134 loc) · 5.34 KB
/
hoac_decoder.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
173
174
175
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
HOAC - Higher-Order Ambisonics Audio Compression - Decoder
@author: Chris Hold
"""
from pathlib import Path
import time
import numpy as np
import matplotlib.pyplot as plt
import spaudiopy as spa
import safpy
import hoac
SAVE = True
PLOT = False
PLAY = False
# Read
hoac_file = Path("./out_file.hoac")
conf, sig_tc, doa_idx_stream, dif_idx_stream = hoac.read_hoac(file=hoac_file)
# Prepare
N_sph_out = conf['N_sph_in']
num_sh_out = (N_sph_out+1)**2
r_smooth = 0.75
x_tc = np.sqrt(4*np.pi)*sig_tc.get_signals()
fs = sig_tc.fs
num_ch = x_tc.shape[0]
blocksize = conf['blocksize']
hopsize = conf['hopsize']
num_slots = conf['blocksize'] // conf['hopsize']
hSTFT = safpy.afstft.AfSTFT(num_ch, num_sh_out, hopsize, fs)
hSTFT.clear_buffers()
num_bands = hSTFT.num_bands
freqs = hSTFT.center_freqs
f_qt = hoac.get_f_quantizer(num_bands)
f_qt_c = np.asarray([np.mean(freqs[idx[0]: idx[1]]) for idx in f_qt])
qgrid = conf['qgrid']
qdifbins = np.append(conf['qdifbins'], 1)
x_tc = np.hstack((x_tc, np.zeros((num_ch, hSTFT.processing_delay))))
out_sig = np.zeros((num_sh_out, x_tc.shape[1]))
fd_sig_in = np.zeros((num_slots, num_ch, num_bands), dtype='complex64')
A_nm = conf['A_nm']
beta = conf['beta']
B_nm, B_nm_trunc, num_recov = hoac.sph_filterbank_reconstruction(A_nm)
assert num_recov
B_nm_trunc = B_nm_trunc[:, np.newaxis, :, np.newaxis]
B_nm_exp = np.zeros((num_sh_out, num_slots, num_ch, num_bands))
B_nm_exp[:B_nm.shape[0], :, :, :] = B_nm[:, np.newaxis, :, np.newaxis]
N_sph_recov = int(np.sqrt(num_recov) - 1)
C_dif = hoac.get_cov_dif(N_sph_out, num_ch, conf)
orne = num_sh_out / np.trace(A_nm.conj().T @ A_nm)
M_mavg = np.zeros((N_sph_out+1, num_sh_out))
for n in range(N_sph_out+1):
M_mavg[n, n**2:(n+1)**2] = 1/(2*n+1)
num_m = np.asarray([2*n+1 for n in range(N_sph_out+1)])
# Initialize and start timer
start_time = time.time()
doa = np.zeros((num_slots, num_ch, num_bands, 3))
doa_prev = np.zeros_like(doa)
dif = np.zeros((num_slots, num_ch, num_bands))
dif_prev = np.zeros_like(dif)
Y = np.zeros((num_sh_out, num_slots, num_ch, num_bands))
X_nm = np.zeros((num_slots, num_sh_out, num_bands), dtype=np.complex_)
M = np.zeros_like(Y)
M_prev = np.zeros_like(M)
g = np.ones((N_sph_out+1, num_bands))
print(" HOAC decoding.")
start_smpl = 0
idx_blk = 0
while idx_blk < doa_idx_stream.shape[0]:
blk_in = x_tc[:, range(start_smpl, start_smpl+blocksize)]
fd_sig_in[:] = hSTFT.forward(blk_in)
doa[:], dif[:] = hoac.dequantize_dirac_pars(doa_idx_stream[idx_blk, ...],
dif_idx_stream[idx_blk, ...],
freqs, f_qt_c,
qgrid, qdifbins)
M[:], Y[:] = hoac.formulate_M_Y(doa, dif, N_sph_out, B_nm_exp, beta,
num_recov, B_nm_trunc)
if idx_blk == 0:
M_prev[:] = M[:]
M = r_smooth * M + (1. - r_smooth) * M_prev
X_nm[:] = np.einsum('ldsk,dsk->dlk', M, fd_sig_in)
gn = hoac.opt_gain(X_nm, Y, dif,
np.real((fd_sig_in * fd_sig_in.conj())),
C_dif, orne, M_mavg)
np.clip(gn, 0.5, 2., out=gn)
g = r_smooth * gn + (1. - r_smooth) * g
X_nm[:, num_recov:, :] = np.repeat(g, num_m, axis=0)[
np.newaxis, num_recov:, :] * X_nm[:, num_recov:, :]
# back
blk_out = hSTFT.backward(X_nm)
out_sig[:, range(start_smpl, start_smpl+blocksize)] = blk_out
M_prev[:] = M[:]
start_smpl += blocksize
idx_blk += 1
out_sig = out_sig[:, hSTFT.processing_delay:]
print("READY")
Path('./audio').mkdir(parents=True, exist_ok=True)
spa.io.save_audio(spa.sph.n3d_to_sn3d(out_sig).T,
"./audio/out_hoac_ambix.wav", fs)
print('Decoding: ', time.time()-start_time, 'seconds.')
if SAVE:
hrirs_nm = spa.decoder.magls_bin(spa.io.load_sofa_hrirs(
'~/data/HRTFs/THK_KU100/HRIR_L2354.sofa'), N_sph_out)
out_bin = spa.sig.MultiSignal([*spa.decoder.sh2bin(0.3*out_sig, hrirs_nm)],
fs=fs)
out_bin.save("./audio/Hoac_bin.wav", "PCM_16")
uncompressed_sig = spa.sph.sn3d_to_n3d(
spa.io.load_audio("./audio/in_sig_ambix.wav", fs).get_signals()[:num_sh_out, :])
in_bin = spa.sig.MultiSignal([*spa.decoder.sh2bin(0.3*uncompressed_sig, hrirs_nm)],
fs=fs)
in_bin.save("./audio/Input_bin.wav", "PCM_16")
print("RMSE(n) ratio:",
np.round(((M_mavg @ spa.utils.rms(out_sig, axis=-1)) /
(10e-10 + M_mavg @ spa.utils.rms(uncompressed_sig, axis=-1))), 3))
print("RMSE(n) dB:",
np.round(spa.utils.db(
(M_mavg @ spa.utils.rms(out_sig, axis=-1)) /
(10e-10 + M_mavg @ spa.utils.rms(uncompressed_sig, axis=-1))), 3))
print("RMSE dB:",
np.round(np.mean(np.abs(spa.utils.db(
(spa.utils.rms(out_sig, axis=-1)) /
(10e-10 + spa.utils.rms(uncompressed_sig, axis=-1))))), 3))
if PLAY:
print("Playing input")
in_bin.play()
print("Playing decoded")
out_bin.play()
if PLOT:
spa.plot.sh_bar([spa.utils.rms(out_sig) /
(spa.utils.rms(uncompressed_sig+1e-10))], TODB=1,
centered=True, s=800)
spa.plot.sh_rms_map(out_sig+1e-10, TODB=True,
title="Output SHD Signal")
plt.show()