/* HTS_freqt: frequency transformation */
static void HTS_freqt(HTS_Vocoder * v, const double *c1, const int m1, double *c2, const int m2, const double a)
{
int i, j;
const double b = 1 - a * a;
double *g;
if (m2 > v->freqt_size) {
if (v->freqt_buff != NULL)
HTS_free(v->freqt_buff);
v->freqt_buff = (double *) HTS_calloc(m2 + m2 + 2, sizeof(double));
v->freqt_size = m2;
}
g = v->freqt_buff + v->freqt_size + 1;
for (i = 0; i < m2 + 1; i++)
g[i] = 0.0;
for (i = -m1; i <= 0; i++) {
if (0 <= m2)
g[0] = c1[-i] + a * (v->freqt_buff[0] = g[0]);
if (1 <= m2)
g[1] = b * v->freqt_buff[0] + a * (v->freqt_buff[1] = g[1]);
for (j = 2; j <= m2; j++)
g[j] = v->freqt_buff[j - 1] + a * ((v->freqt_buff[j] = g[j]) - g[j - 1]);
}
HTS_movem(g, c2, m2 + 1);
}
/* HTS_gc2gc: generalized cepstral transformation */
static void HTS_gc2gc(HTS_Vocoder * v, double *c1, const int m1, const double g1, double *c2, const int m2, const double g2)
{
int i, min, k, mk;
double ss1, ss2, cc;
if (m1 > v->gc2gc_size) {
if (v->gc2gc_buff != NULL)
HTS_free(v->gc2gc_buff);
v->gc2gc_buff = (double *) HTS_calloc(m1 + 1, sizeof(double));
v->gc2gc_size = m1;
}
HTS_movem(c1, v->gc2gc_buff, m1 + 1);
c2[0] = v->gc2gc_buff[0];
for (i = 1; i <= m2; i++) {
ss1 = ss2 = 0.0;
min = m1 < i ? m1 : i - 1;
for (k = 1; k <= min; k++) {
mk = i - k;
cc = v->gc2gc_buff[k] * c2[mk];
ss2 += k * cc;
ss1 += mk * cc;
}
if (i <= m1)
c2[i] = v->gc2gc_buff[i] + (g2 * ss2 - g1 * ss1) / i;
else
c2[i] = (g2 * ss2 - g1 * ss1) / i;
}
}
/* HTS_gnorm: gain normalization */
static void HTS_gnorm(double *c1, double *c2, int m, const double g)
{
double k;
if (g != 0.0) {
k = 1.0 + g * c1[0];
for (; m >= 1; m--)
c2[m] = c1[m] / k;
c2[0] = pow(k, 1.0 / g);
} else {
HTS_movem(&c1[1], &c2[1], m);
c2[0] = exp(c1[0]);
}
}
/* HTS_ignorm: inverse gain normalization */
static void HTS_ignorm(double *c1, double *c2, int m, const double g)
{
double k;
if (g != 0.0) {
k = pow(c1[0], g);
for (; m >= 1; m--)
c2[m] = k * c1[m];
c2[0] = (k - 1.0) / g;
} else {
HTS_movem(&c1[1], &c2[1], m);
c2[0] = log(c1[0]);
}
}
/* HTS_mc2b: transform mel-cepstrum to MLSA digital fillter coefficients */
static void HTS_mc2b(double *mc, double *b, int m, const double a)
{
if (mc != b) {
if (a != 0.0) {
b[m] = mc[m];
for (m--; m >= 0; m--)
b[m] = mc[m] - a * b[m + 1];
} else
HTS_movem(mc, b, m + 1);
} else if (a != 0.0)
for (m--; m >= 0; m--)
b[m] -= a * b[m + 1];
}
/* HTS_Vocoder_postfilter_lsp: postfilter for LSP */
static void HTS_Vocoder_postfilter_lsp(HTS_Vocoder * v, double *lsp, size_t m, double alpha, double beta)
{
double e1, e2;
size_t i;
double d1, d2;
if (beta > 0.0 && m > 1) {
if (v->postfilter_size < m) {
if (v->postfilter_buff != NULL)
HTS_free(v->postfilter_buff);
v->postfilter_buff = (double *) HTS_calloc(m + 1, sizeof(double));
v->postfilter_size = m;
}
e1 = HTS_lsp2en(v, lsp, m, alpha);
/* postfiltering */
for (i = 0; i <= m; i++) {
if (i > 1 && i < m) {
d1 = beta * (lsp[i + 1] - lsp[i]);
d2 = beta * (lsp[i] - lsp[i - 1]);
v->postfilter_buff[i] = lsp[i - 1] + d2 + (d2 * d2 * ((lsp[i + 1] - lsp[i - 1]) - (d1 + d2))) / ((d2 * d2) + (d1 * d1));
} else {
v->postfilter_buff[i] = lsp[i];
}
}
HTS_movem(v->postfilter_buff, lsp, m + 1);
e2 = HTS_lsp2en(v, lsp, m, alpha);
if (e1 != e2) {
if (v->use_log_gain)
lsp[0] += 0.5 * log(e1 / e2);
else
lsp[0] *= sqrt(e1 / e2);
}
}
}
/* HTS_Vocoder_synthesize: pulse/noise excitation and MLSA/MGLSA filster based waveform synthesis */
void HTS_Vocoder_synthesize(HTS_Vocoder * v, size_t m, double lf0, double *spectrum, size_t nlpf, double *lpf, double alpha, double beta, double volume, double *rawdata, HTS_Audio * audio)
{
double x;
int i, j;
short xs;
int rawidx = 0;
double p;
/* lf0 -> pitch */
if (lf0 == LZERO)
p = 0.0;
else if (lf0 <= MIN_LF0)
p = v->rate / MIN_F0;
else if (lf0 >= MAX_LF0)
p = v->rate / MAX_F0;
else
p = v->rate / exp(lf0);
/* first time */
if (v->is_first == TRUE) {
HTS_Vocoder_initialize_excitation(v, p, nlpf);
if (v->stage == 0) { /* for MCP */
HTS_mc2b(spectrum, v->c, m, alpha);
} else { /* for LSP */
HTS_movem(spectrum, v->c, m + 1);
HTS_lsp2mgc(v, v->c, v->c, m, alpha);
HTS_mc2b(v->c, v->c, m, alpha);
HTS_gnorm(v->c, v->c, m, v->gamma);
for (i = 1; i <= m; i++)
v->c[i] *= v->gamma;
}
v->is_first = FALSE;
}
HTS_Vocoder_start_excitation(v, p);
if (v->stage == 0) { /* for MCP */
HTS_Vocoder_postfilter_mcp(v, spectrum, m, alpha, beta);
HTS_mc2b(spectrum, v->cc, m, alpha);
for (i = 0; i <= m; i++)
v->cinc[i] = (v->cc[i] - v->c[i]) / v->fprd;
} else { /* for LSP */
HTS_Vocoder_postfilter_lsp(v, spectrum, m, alpha, beta);
HTS_check_lsp_stability(spectrum, m);
HTS_lsp2mgc(v, spectrum, v->cc, m, alpha);
HTS_mc2b(v->cc, v->cc, m, alpha);
HTS_gnorm(v->cc, v->cc, m, v->gamma);
for (i = 1; i <= m; i++)
v->cc[i] *= v->gamma;
for (i = 0; i <= m; i++)
v->cinc[i] = (v->cc[i] - v->c[i]) / v->fprd;
}
for (j = 0; j < v->fprd; j++) {
x = HTS_Vocoder_get_excitation(v, lpf);
if (v->stage == 0) { /* for MCP */
if (x != 0.0)
x *= exp(v->c[0]);
x = HTS_mlsadf(x, v->c, m, alpha, PADEORDER, v->d1);
} else { /* for LSP */
if (!NGAIN)
x *= v->c[0];
x = HTS_mglsadf(x, v->c, m, alpha, v->stage, v->d1);
}
x *= volume;
/* output */
if (rawdata)
rawdata[rawidx++] = x;
if (audio) {
if (x > 32767.0)
xs = 32767;
else if (x < -32768.0)
xs = -32768;
else
xs = (short) x;
HTS_Audio_write(audio, xs);
}
for (i = 0; i <= m; i++)
v->c[i] += v->cinc[i];
}
HTS_Vocoder_end_excitation(v, p);
HTS_movem(v->cc, v->c, m + 1);
}
/* HTS_lsp2lpc: transform LSP to LPC */
static void HTS_lsp2lpc(HTS_Vocoder * v, double *lsp, double *a, const int m)
{
int i, k, mh1, mh2, flag_odd;
double xx, xf, xff;
double *p, *q;
double *a0, *a1, *a2, *b0, *b1, *b2;
flag_odd = 0;
if (m % 2 == 0)
mh1 = mh2 = m / 2;
else {
mh1 = (m + 1) / 2;
mh2 = (m - 1) / 2;
flag_odd = 1;
}
if (m > v->lsp2lpc_size) {
if (v->lsp2lpc_buff != NULL)
HTS_free(v->lsp2lpc_buff);
v->lsp2lpc_buff = (double *) HTS_calloc(5 * m + 6, sizeof(double));
v->lsp2lpc_size = m;
}
p = v->lsp2lpc_buff + m;
q = p + mh1;
a0 = q + mh2;
a1 = a0 + (mh1 + 1);
a2 = a1 + (mh1 + 1);
b0 = a2 + (mh1 + 1);
b1 = b0 + (mh2 + 1);
b2 = b1 + (mh2 + 1);
HTS_movem(lsp, v->lsp2lpc_buff, m);
for (i = 0; i < mh1 + 1; i++)
a0[i] = 0.0;
for (i = 0; i < mh1 + 1; i++)
a1[i] = 0.0;
for (i = 0; i < mh1 + 1; i++)
a2[i] = 0.0;
for (i = 0; i < mh2 + 1; i++)
b0[i] = 0.0;
for (i = 0; i < mh2 + 1; i++)
b1[i] = 0.0;
for (i = 0; i < mh2 + 1; i++)
b2[i] = 0.0;
/* lsp filter parameters */
for (i = k = 0; i < mh1; i++, k += 2)
p[i] = -2.0 * cos(v->lsp2lpc_buff[k]);
for (i = k = 0; i < mh2; i++, k += 2)
q[i] = -2.0 * cos(v->lsp2lpc_buff[k + 1]);
/* impulse response of analysis filter */
xx = 1.0;
xf = xff = 0.0;
for (k = 0; k <= m; k++) {
if (flag_odd) {
a0[0] = xx;
b0[0] = xx - xff;
xff = xf;
xf = xx;
} else {
a0[0] = xx + xf;
b0[0] = xx - xf;
xf = xx;
}
for (i = 0; i < mh1; i++) {
a0[i + 1] = a0[i] + p[i] * a1[i] + a2[i];
a2[i] = a1[i];
a1[i] = a0[i];
}
for (i = 0; i < mh2; i++) {
b0[i + 1] = b0[i] + q[i] * b1[i] + b2[i];
b2[i] = b1[i];
b1[i] = b0[i];
}
if (k != 0)
a[k - 1] = -0.5 * (a0[mh1] + b0[mh2]);
xx = 0.0;
}
for (i = m - 1; i >= 0; i--)
a[i + 1] = -a[i];
a[0] = 1.0;
}
/* HTS_Vocoder_synthesize: pulse/noise excitation and MLSA/MGLSA filster based waveform synthesis */
void HTS_Vocoder_synthesize(HTS_Vocoder *v, const int m, double lf0,
double *spectrum, double alpha, double beta,
short *rawdata)
{
double x;
int i, j;
short xs;
int rawidx = 0;
double p;
/* lf0 -> pitch */
if (lf0 == LZERO)
p = 0.0;
else
p = v->rate / exp(lf0);
/* first time */
if (v->p1 < 0.0) {
if (v->gauss & (v->seed != 1))
v->next = HTS_srnd((unsigned) v->seed);
HTS_Vocoder_initialize_excitation(v);
if (v->stage != 0) { /* for LSP */
if (v->use_log_gain)
v->c[0] = LZERO;
else
v->c[0] = ZERO;
for (i = 0; i <= m; i++)
v->c[i] = i * PI / (m + 1);
HTS_lsp2mgc(v, v->c, v->c, m, alpha);
HTS_mc2b(v->c, v->c, m, alpha);
HTS_gnorm(v->c, v->c, m, v->gamma);
for (i = 1; i <= m; i++)
v->c[i] *= v->gamma;
}
}
HTS_Vocoder_start_excitation(v, p);
if (v->stage == 0) { /* for MCP */
HTS_Vocoder_postfilter_mcp(v, spectrum, m, alpha, beta);
HTS_mc2b(spectrum, v->cc, m, alpha);
for (i = 0; i <= m; i++)
v->cinc[i] = (v->cc[i] - v->c[i]) * v->iprd / v->fprd;
} else { /* for LSP */
HTS_lsp2mgc(v, spectrum, v->cc, m, alpha);
HTS_mc2b(v->cc, v->cc, m, alpha);
HTS_gnorm(v->cc, v->cc, m, v->gamma);
for (i = 1; i <= m; i++)
v->cc[i] *= v->gamma;
for (i = 0; i <= m; i++)
v->cinc[i] = (v->cc[i] - v->c[i]) * v->iprd / v->fprd;
}
for (j = 0, i = (v->iprd + 1) / 2; j < v->fprd; j++) {
x = HTS_Vocoder_get_excitation(v, j, i);
if (v->stage == 0) { /* for MCP */
if (x != 0.0)
x *= exp(v->c[0]);
x = HTS_mlsadf(x, v->c, m, alpha, PADEORDER, v->d1, v->pade);
} else { /* for LSP */
if (!NGAIN)
x *= v->c[0];
x = HTS_mglsadf(x, v->c, m, alpha, v->stage, v->d1);
}
xs = (short) (1.00*x);
if (rawdata)
rawdata[rawidx++] = xs;
if (v->audio)
HTS_Audio_write(v->audio, xs);
if (!--i) {
for (i = 0; i <= m; i++)
v->c[i] += v->cinc[i];
i = v->iprd;
}
}
HTS_Vocoder_end_excitation(v);
HTS_movem(v->cc, v->c, m + 1);
}