int phase, phase_model;
float ex_phase[1];
+ int voiced, voiced_1, voiced_2, voiced_synth;
int postfilt;
float bg_est;
int hand_snr;
FILE *fsnr;
-
+
+ MODEL model_1, model_2, model_3, model_synth, model_a, model_b;
+ int transition;
+
+ int vf=0;
+
+ voiced_1 = voiced_2 = 0;
+ model_1.Wo = TWO_PI/P_MIN;
+ model_1.L = floor(PI/model_1.Wo);
+ for(i=1; i<=model_1.L; i++) {
+ model_1.A[i] = 0.0;
+ model_1.phi[i] = 0.0;
+ }
+ model_synth = model_3 = model_2 = model_1;
+
if (argc < 3) {
printf("usage: sinedec InputFile ModelFile [-o OutputFile] [-o lpc Order]\n");
printf(" [--dump DumpFilePrefix]\n");
bg_est = 0.0;
postfilt = switch_present("--postfilter",argc,argv);
+ transition = 0;
+
/* Initialise ------------------------------------------------------------*/
init_decoder();
dump_Sn(Sn);
dft_speech(); dump_Sw(Sw);
- dump_model(&model);
+ //dump_model(&model);
/* optional phase modelling - make sure this happens before LPC
modelling of {Am} as first order model fit doesn't work well
assert(order == PHASE_LPC_ORD);
dump_ak(ak_phase, PHASE_LPC_ORD);
- snr = phase_model_first_order(ak_phase, H, &n_min, &min_Am);
+ snr = phase_model_first_order(ak_phase, H, &n_min, &min_Am, &voiced);
dump_snr(snr);
if (phase_model == 0) {
model.phi[i] = 0;
if (hand_snr)
fscanf(fsnr,"%f\n",&snr);
- phase_synth_zero_order(snr, H, ex_phase);
+ phase_synth_zero_order(voiced, H, ex_phase, voiced);
}
if (phase_model == 1) {
- phase_synth_first_order(snr, H, n_min, min_Am);
+ phase_synth_first_order(voiced, H, n_min, min_Am, voiced);
}
if (postfilt)
- postfilter(&model, snr>2.0, &bg_est);
+ postfilter(&model, voiced, &bg_est);
dump_phase_(&model.phi[0]);
}
dump_quantised_model(&model);
}
+ #define DEC
+ #ifdef DEC
+ /* Decimate to 20ms frame rate. In the code we only send
+ off frames to the receiver. To simulate this on odd
+ frames the model parameters pass straight thru. On even
+ frames we interpolate from adjacent odd frames. A one
+ frame delay is required for the odd frames.
+ */
+
+ /*
+ frames Transmitted to Rx Decimator output
+
+ 0 n 0.5*model(-3) + 0.5*model(-1)
+ 1 y model(-1)
+ 2 n 0.5*model(-1) + 0.5*model(1)
+ 3 y model(1)
+ 4 n 0.5*model(1) + 0.5*model(3)
+ 5 y model(3)
+ */
+
+ /*
+ TODO:
+ [ ] Voicing decision
+ [ ] unvoiced
+ [ ] amplitudes
+ [ ] phases
+ [ ] Wo
+ [ ] unvoiced
+ [ ] amplitudes
+ [ ] phases
+ + OK to run zero phase model on 10ms rate, using info
+ from adjacent 20 ms frames
+ [ ] Wo
+ */
+
+ dump_model(&model_2);
+
+ if (frames%2) {
+
+ /* odd frames use the original model parameters */
+
+ model_synth = model_2;
+ transition = 0;
+
+ }
+ else {
+ /* even frame so we need to synthesise the model parameters by
+ interpolating between adjacent frames */
+
+ model_synth = model_2;
+ voiced_synth = voiced && voiced_2;
+ if (fabs(model_1.Wo - model_3.Wo) < 0.1*model_1.Wo) {
+ /* If the Wo of adjacent frames is within 10% we synthesise a
+ continuous track through this frame by linear interpolation
+ of the amplitudes and Wo. This is typical of a strongly
+ voiced frame.
+ */
+ transition = 0;
+
+ /* continuous track through this frame */
+ #define T
+ #ifdef T
+ model_synth.Wo = (model_1.Wo + model_3.Wo)/2.0;
+ if (model_1.L > model_3.L)
+ model_synth.L = model_3.L;
+ else
+ model_synth.L = model_1.L;
+ #endif
+ for(i=1; i<=model_synth.L; i++) {
+ model_synth.A[i] = (model_3.A[i] + model_1.A[i])/2.0;
+ /* cheat on phases for now, these were constructed using
+ LPC model from actual speech for this frame - fix later */
+ model_synth.phi[i] = model_2.phi[i];
+ }
+ vf++;
+ }
+ else {
+ /*
+ transition frame, adjacent frames have different Wo and
+ L so set up two sets of model parameters based on
+ previous and next frame. We then synthesise both of
+ them and add them together in the time domain. Note
+ the adjustments to phase to shift the timing of the
+ model parameters forward or back N samples.
+
+ This case is typical of unvoiced speech or background noise
+ of a voiced to unvoiced transition.
+ */
+
+ transition = 1;
+
+ memcpy(&model_a, &model_3, sizeof(model));
+ memcpy(&model_b, &model_1, sizeof(model));
+ for(i=1; i<=model_a.L; i++) {
+ model_a.A[i] /= 2.0;
+ model_a.phi[i] += model_a.Wo*i*N;
+ }
+ for(i=1; i<=model_b.L; i++) {
+ model_b.A[i] /= 2.0;
+ model_b.phi[i] -= model_b.Wo*i*N;
+ }
+ }
+ }
+
+ voiced_2 = voiced_1;
+ voiced_1 = voiced;
+
+ model_3 = model_2;
+ model_2 = model_1;
+ model_1 = model;
+ model = model_synth;
+#endif
+
/* Synthesise speech */
if (fout != NULL) {
- synthesise_mixed(Pn,&model,Sn_);
+ if (transition) {
+ synthesise_mixed(Pn,&model_a,Sn_,1);
+ synthesise_mixed(Pn,&model_b,Sn_,0);
+ }
+ else {
+ synthesise_mixed(Pn,&model,Sn_,1);
+ }
/* Save output speech to disk */
}
//printf("gmin = %f\n", get_gmin());
+ printf("vf = %d\n", vf);
if (fout != NULL)
fclose(fout);
#include "sine.h"
void synthesise_mixed(
- float Pn[], /* time domain Parzen window */
+ float Pn[], /* time domain Parzen window */
MODEL *model, /* ptr to model parameters for this frame */
- float Sn_[] /* time domain synthesised signal */
+ float Sn_[], /* time domain synthesised signal */
+ int shift /* if non-zero update memories */
)
{
int i,l,j,b; /* loop variables */
COMP Sw_[FFT_DEC]; /* DFT of synthesised signal */
- /* Update memories */
+ if (shift) {
+ /* Update memories */
- for(i=0; i<N-1; i++) {
- Sn_[i] = Sn_[i+N];
+ for(i=0; i<N-1; i++) {
+ Sn_[i] = Sn_[i+N];
+ }
+ Sn_[N-1] = 0.0;
}
- Sn_[N-1] = 0.0;
for(i=0; i<FFT_DEC; i++) {
Sw_[i].real = 0.0;
/* Overlap add to previous samples */
for(i=0; i<N-1; i++) {
- Sn_[i] += Sw_[FFT_DEC-N+1+i].real*Pn[i];
+ Sn_[i] += Sw_[FFT_DEC-N+1+i].real*Pn[i];
}
- for(i=N-1,j=0; i<2*N; i++,j++)
- Sn_[i] = Sw_[j].real*Pn[i];
+
+ if (shift)
+ for(i=N-1,j=0; i<2*N; i++,j++)
+ Sn_[i] = Sw_[j].real*Pn[i];
+ else
+ for(i=N-1,j=0; i<2*N; i++,j++)
+ Sn_[i] += Sw_[j].real*Pn[i];
+
}
/*---------------------------------------------------------------------------*\
Result: when tested was no difference in output speech quality. The
partial unvoiced sound when using zero phase model was found to be
- due mis-laignment of teh LPC analysis window and accidental addition
+ due mis-alignment of the LPC analysis window and accidental addition
of a random phase component. So we are sticking with synthesise_mixed()
above for now.
projects / freetel-svn-tracking.git / commitdiff