% [ ] switches to turn on/off quantisation
% [ ] rename mask_sample_freqs, do we need "mask" any more
+#{
+ [ ] refactor
+#}
+
function [fvec_log amps_log] = newamp1_batch(samname, optional_Am_out_name, optional_Aw_out_name)
newamp;
more off;
model = load(model_name);
[frames nc] = size(model);
+ K = 40;
+ [rate_K_surface rate_K_sample_freqs_kHz] = resample_const_rate_f(model, K);
+
+ #{
+ So given surface, lets look at spectral content and see if we can reduce it while maintaining
+ speech quality. First step is to dft across time and plot.
+ #}
+
+ %Nf=20; Nf2 = floor(Nf/2)+1;
+ %b = fir1(Nf,0.25)
+ %b = zeros(1, Nf2);
+ Nf = 4; Nf2 = 5;
+ [b a]= cheby1(4, 1, 0.20);
+ dft_surface = zeros(frames,K);
+ rate_K_surface_filt = zeros(frames,K);
+ dft_surface_filt = zeros(frames,K);
+ for k=1:K
+ dft_surface(:,k) = fft(rate_K_surface(:,k).*hanning(frames));
+ rate_K_surface_filt(:,k) = filter(b, a, rate_K_surface(:,k));
+ dft_surface_filt(:,k) = fft(rate_K_surface_filt(:,k).*hanning(frames));
+ end
+ figure(1); clf;
+ mesh(abs(dft_surface))
+ figure(2); clf;
+ mesh(abs(dft_surface_filt))
+
+ Fs = 100; Ts = 1/Fs;
+ figure(3);
+ subplot(211);
+ h = freqz(b,a,Fs/2);
+ plot(1:Fs/2, 20*log10(abs(h)))
+ axis([1 Fs/2 -80 0])
+ ylabel('Gain (dB)')
+ grid;
+ subplot(212)
+ [g w] = grpdelay(b,a);
+ plot(w*Fs/pi, 1000*g*Ts)
+ %axis([1 Fs/2 0 0.5])
+ xlabel('Frequency (Hz)');
+ ylabel('Delay (ms)')
+ grid
+
+ figure(4); clf; stem(filter(b,a,[1 zeros(1,20)]))
+
+ % adjust for time offset due to filtering
+
+ rate_K_surface_filt = [rate_K_surface_filt(Nf2:frames,:); zeros(Nf2, K)];
+
+ % back down to rate L
+
+ model_ = resample_rate_L(model, rate_K_surface_filt, rate_K_sample_freqs_kHz);
+
if nargin == 2
Am_out_name = optional_Am_out_name;
else
fam = fopen(Am_out_name,"wb");
+ Wo_out_name = sprintf("%s_Wo.out", samname);
+ fWo = fopen(Wo_out_name,"wb");
+
+ for f=1:frames
+ printf("%d ", f);
+ Wo = model_(f,1);
+ L = min([model_(f,2) max_amp-1]);
+ Am = model_(f,3:(L+2));
+
+ Am_ = zeros(1,max_amp);
+ Am_(2:L) = Am(1:L-1);
+ fwrite(fam, Am_, "float32");
+ fwrite(fWo, Wo, "float32");
+ end
+
+#{
% encoder loop ------------------------------------------------------
fvec_log = []; amps_log = [];
- % prime initial values so first pass iterpolator works. TODO: improve this
-
for f=1:frames
+ printf("%d ", f);
Wo = model(f,1);
L = min([model(f,2) max_amp-1]);
Am = model(f,3:(L+2));
fvec_log = [fvec_log; fvec];
amps_log = [amps_log; amps];
- model_(f,1) = Wo; model_(f,2) = L; model_(f,3:(L+2)) = 10 .^ (AmdB_(1:L)/20);
+ model_(f,1) = Wo; model_(f,2) = L; model_(f,3:(L+2)) = 10 .^ (AmdB(1:L)/20);
end
for f=1:frames
+ if 0
if (f > 1) && (e(f) > (e(f-1)+3))
decimate = 2;
else
decimate = 4;
end
+ end
printf("%d ", decimate);
-
+
AmdB_ = decimate_frame_rate(model_, decimate, f, frames);
L = length(AmdB_);
Am_ = zeros(1,max_amp);
Am_(2:L) = 10 .^ (AmdB_(1:L-1)/20); % C array doesnt use A[0]
fwrite(fam, Am_, "float32");
- Am_ = zeros(1,max_amp);
+ fwrite(fWo, Wo, "float32");
end
+#}
fclose(fam);
+ fclose(fWo);
printf("\n")
- figure(1); clf;
- plot(e,'+-');
endfunction
+
+#{
+ [X] resample A(mWo,t) to fixed rate, creating a surface
+ [X] resample surface at half frame increments
+ + this will test resampling in freq is OK
+ [X] listen to a few files and determine if any distortion
+ [X] experiment with filtering in time
+ [ ] decimate to 25Hz
+ [ ] resample to 100Hz
+ [ ] listen to a few files and compare to just filtering at 12Hz
+#}
+
+function [rate_K_surface rate_K_sample_freqs_kHz] = resample_const_rate_f(model, K=50)
+
+ % convert rate L=pi/Wo amplitude samples to fixed rate K
+
+ max_amp = 80;
+ [frames col] = size(model);
+ rate_K_sample_freqs_kHz = (1:K)*4/K;
+ rate_K_surface = zeros(frames, K);
+
+ for f=1:frames
+ Wo = model(f,1);
+ L = min([model(f,2) max_amp-1]);
+ Am = model(f,3:(L+2));
+ AmdB = 20*log10(Am);
+ rate_L_sample_freqs_kHz = (1:L)*Wo*4/pi;
+
+ rate_K_surface(f,:) = interp1(rate_L_sample_freqs_kHz, AmdB, rate_K_sample_freqs_kHz, "spline", "extrap");
+ end
+endfunction
+
+
+function model_ = resample_rate_L(model, rate_K_surface, rate_K_sample_freqs_kHz, K=50)
+ max_amp = 80;
+ [frames col] = size(model);
+
+ model_ = zeros(frames, max_amp+3);
+ for f=1:frames-1
+ Wo = (model(f,1) + model(f+1,1))/2;
+ L = min(pi/Wo, max_amp-1);
+ rate_L_sample_freqs_kHz = (1:L)*Wo*4/pi;
+
+ % back down to rate L
+
+ AmdB_ = interp1(rate_K_sample_freqs_kHz, rate_K_surface(f,:), rate_L_sample_freqs_kHz, "spline", "extrap");
+
+ % todo a way to save all model params ... Am, Wo, L, (v?) Start with current facility then
+ % make it better
+
+ model_(f,1) = Wo; model_(f,2) = L; model_(f,3:(L+2)) = 10 .^ (AmdB_(1:L)/20);
+ end
+endfunction
+
+
+function model_ = resample_half_frame_offset(model, rate_K_surface, rate_K_sample_freqs_kHz, K=50)
+ max_amp = 80;
+ [frames col] = size(model);
+
+ % Check distortion by shifting half a frame and returning model for
+ % synthesis Hmm how to handle Wo? Well lets assume it's OK and
+ % average it. If it sounds OK then it must be OK as well and Am
+ % interpolation. If not then will do some more thinking.
+
+ model_ = zeros(frames, max_amp+3);
+ for f=1:frames-1
+ Wo = (model(f,1) + model(f+1,1))/2;
+ L = min(pi/Wo, max_amp-1);
+ rate_L_sample_freqs_kHz = (1:L)*Wo*4/pi;
+
+ % interpolate at half frame offset
+
+ half = 0.5*rate_K_surface(f,:) + 0.5*rate_K_surface(f+1,:);
+
+ % back down to rate L
+
+ AmdB_ = interp1(rate_K_sample_freqs_kHz, half, rate_L_sample_freqs_kHz, "spline", "extrap");
+
+ % todo a way to save all model params ... Am, Wo, L, (v?) Start with current facility then
+ % make it better
+
+ model_(f,1) = Wo; model_(f,2) = L; model_(f,3:(L+2)) = 10 .^ (AmdB_(1:L)/20);
+ end
+endfunction
projects / freetel-svn-tracking.git / commitdiff