-
Notifications
You must be signed in to change notification settings - Fork 7
/
replace_stft_with_fb.m
78 lines (69 loc) · 2.18 KB
/
replace_stft_with_fb.m
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
clear all; close all; clc
% oversampled DFT filter bank
%% assume 16 KHz input
fb = FilterBankStruct( );
fb.T = 256/32; % 16 ms FFT size, first scale down the problem
fb.B = 128/32; % 8 ms hop size
Lh = 512/32;
Lg = 512/32;
fb.tau0 = 256/32 - 1; % 16 ms latency
fb.symmetry = [-1;0;0];
% fb.momentum = 2; % to match sidelobe tails with Hann as well
eta = 1e4;
lambda = 0.0; % set lambda > 0 if there is overshoot
best_cost = inf;
best_fb = fb;
for num_trial = 1 : 100
fb.h = rand(Lh,1); fb.g = rand(Lg,1);
[fb, cost, recon_err, iter] = FilterBankDesign(fb, eta, lambda, 100);
fprintf('Trial: %g; cost: %g; reconstruction error: %g; iterations %g\n', num_trial, cost, recon_err, iter)
if cost < best_cost
best_cost = cost;
best_fb = fb;
end
end
fb = FilterBankStruct( );
fb.T = 256; % 16 ms FFT size
fb.B = 128; % 8 ms hop size
fb.tau0 = 256 - 1; % 16 ms latency
fb.symmetry = [-1;0;0];
% fb.momentum = 2; % to match sidelobe tails with Hann as well
fb.h = kron(best_fb.h, ones(32, 1));
fb.g = kron(best_fb.g, ones(32, 1));
eta = 1e4;
lambda = 0.0; % set lambda > 0 if there is overshoot
[fb, cost, recon_err, iter] = FilterBankDesign(fb, eta, lambda, 1000);
fprintf('Refinement. Cost: %g; reconstruction error: %g; iterations %g\n', cost, recon_err, iter)
figure;
subplot(2,1,1)
plot(fb.h, 'b-');
hold on; plot(fb.g, 'k-')
legend('Analysis filter', 'Synthesis filter')
xlabel('Time')
ylabel('Impulse response')
subplot(2,1,2)
fft_size = 32768;
H = 20*log10(abs(fft(fb.h, fft_size)));
plot(pi*[0:fft_size/2-1]/(fft_size/2-1), H(1:end/2), 'b-')
G = 20*log10(abs(fft(fb.g, fft_size)));
hold on; plot(pi*[0:fft_size/2-1]/(fft_size/2-1), G(1:end/2), 'k-')
xlabel('\omega')
ylabel('Magnitude in dB')
figure
[H, F] = freqz(conv(fb.h, fb.g), 1, 32768, 16000);
H = H/max(abs(H));
plot(F, 20*log10(abs(H)))
xlabel('Hz')
ylabel('Analysis-synthesis magnitude in dB')
% STFT with Hann window
h = sqrt(hann(256, 'periodic'));
g = h(end:-1:1);
[H, F] = freqz(conv(h, g), 1, 32768, 16000);
H = H/max(abs(H));
hold on; plot(F, 20*log10(abs(H)))
xlim([0, 200])
ylim([-80, 0])
xlabel('Hz')
ylabel('Analysis-synthesis magnitude in dB')
legend('filterbank', 'sqrt(Hann window)')
title('filterbank vs STFT')