ref: 4a897680e34dc59aebc0dfa8076e38338fbe48e9
parent: f347dd3b3d9c1ffef4f707d39208ffa4750370ab
author: Jean-Marc Valin <[email protected]>
date: Tue Dec 11 19:45:15 EST 2007
Working on some stability issues (appears to be solved by making the pitch projection less aggressive). Also, fixed a 64-bit overflow in the stereo mode and added a "band rotation" function.
--- a/libcelt/bands.c
+++ b/libcelt/bands.c
@@ -35,7 +35,75 @@
#include "vq.h"
#include "cwrs.h"
+/* Applies a series of rotations so that pulses are spread like a two-sided expo
+nential */
+static void exp_rotation(float *X, int len, float theta, int dir)
+{+ int i;
+ float c, s;
+ c = cos(theta);
+ s = sin(theta);
+ if (dir > 0)
+ {+ for (i=0;i<(len/2)-1;i++)
+ {+ float x1, x2;
+ x1 = X[2*i];
+ x2 = X[2*i+2];
+ X[2*i] = c*x1 - s*x2;
+ X[2*i+2] = c*x2 + s*x1;
+
+ x1 = X[2*i+1];
+ x2 = X[2*i+3];
+ X[2*i+1] = c*x1 - s*x2;
+ X[2*i+3] = c*x2 + s*x1;
+ }
+ for (i=(len/2)-3;i>=0;i--)
+ {+ float x1, x2;
+ x1 = X[2*i];
+ x2 = X[2*i+2];
+ X[2*i] = c*x1 - s*x2;
+ X[2*i+2] = c*x2 + s*x1;
+
+ x1 = X[2*i+1];
+ x2 = X[2*i+3];
+ X[2*i+1] = c*x1 - s*x2;
+ X[2*i+3] = c*x2 + s*x1;
+ }
+ } else {+ for (i=0;i<(len/2)-2;i++)
+ {+ float x1, x2;
+ x1 = X[2*i];
+ x2 = X[2*i+2];
+ X[2*i] = c*x1 + s*x2;
+ X[2*i+2] = c*x2 - s*x1;
+
+ x1 = X[2*i+1];
+ x2 = X[2*i+3];
+ X[2*i+1] = c*x1 + s*x2;
+ X[2*i+3] = c*x2 - s*x1;
+ }
+
+ for (i=(len/2)-2;i>=0;i--)
+ {+ float x1, x2;
+ x1 = X[2*i];
+ x2 = X[2*i+2];
+ X[2*i] = c*x1 + s*x2;
+ X[2*i+2] = c*x2 - s*x1;
+
+ x1 = X[2*i+1];
+ x2 = X[2*i+3];
+ X[2*i+1] = c*x1 + s*x2;
+ X[2*i+3] = c*x2 - s*x1;
+ }
+ }
+}
+
+
/* Compute the energy in each of the bands */
void compute_band_energies(const CELTMode *m, float *X, float *bank)
{@@ -159,16 +227,18 @@
q = m->nbPulses[i];
if (q>0) {float n = sqrt(B*(eBands[i+1]-eBands[i]));
- alg_quant(X+B*eBands[i], B*(eBands[i+1]-eBands[i]), q, P+B*eBands[i], enc);
+ alg_quant(X+B*eBands[i], B*(eBands[i+1]-eBands[i]), q, P+B*eBands[i], 0.7, enc);
for (j=B*eBands[i];j<B*eBands[i+1];j++)
norm[j] = X[j] * n;
- //printf ("%f ", log2(ncwrs(B*(eBands[i+1]-eBands[i]), q))/(B*(eBands[i+1]-eBands[i])));+ //printf ("%f ", log2(ncwrs64(B*(eBands[i+1]-eBands[i]), q))/(B*(eBands[i+1]-eBands[i])));+ //printf ("%f ", log2(ncwrs64(B*(eBands[i+1]-eBands[i]), q))); } else {float n = sqrt(B*(eBands[i+1]-eBands[i]));
copy_quant(X+B*eBands[i], B*(eBands[i+1]-eBands[i]), -q, norm, B, eBands[i], enc);
for (j=B*eBands[i];j<B*eBands[i+1];j++)
norm[j] = X[j] * n;
- //printf ("%f ", (1+log2(eBands[i]-(eBands[i+1]-eBands[i]))+log2(ncwrs(B*(eBands[i+1]-eBands[i]), -q)))/(B*(eBands[i+1]-eBands[i])));+ //printf ("%f ", (1+log2(eBands[i]-(eBands[i+1]-eBands[i]))+log2(ncwrs64(B*(eBands[i+1]-eBands[i]), -q)))/(B*(eBands[i+1]-eBands[i])));+ //printf ("%f ", (1+log2(eBands[i]-(eBands[i+1]-eBands[i]))+log2(ncwrs64(B*(eBands[i+1]-eBands[i]), -q))));}
}
//printf ("\n");@@ -189,7 +259,7 @@
q = m->nbPulses[i];
if (q>0) {float n = sqrt(B*(eBands[i+1]-eBands[i]));
- alg_unquant(X+B*eBands[i], B*(eBands[i+1]-eBands[i]), q, P+B*eBands[i], dec);
+ alg_unquant(X+B*eBands[i], B*(eBands[i+1]-eBands[i]), q, P+B*eBands[i], 0.7, dec);
for (j=B*eBands[i];j<B*eBands[i+1];j++)
norm[j] = X[j] * n;
} else {@@ -203,4 +273,18 @@
}
for (i=B*eBands[m->nbEBands];i<B*eBands[m->nbEBands+1];i++)
X[i] = 0;
+}
+
+void band_rotation(const CELTMode *m, float *X, int dir)
+{+ int i, B;
+ const int *eBands = m->eBands;
+ B = m->nbMdctBlocks*m->nbChannels;
+ for (i=0;i<m->nbEBands;i++)
+ {+ float theta;
+ theta = pow(.1f,1.f*abs(m->nbPulses[i])/(B*(eBands[i+1]-eBands[i])));
+ exp_rotation(X+B*eBands[i], B*(eBands[i+1]-eBands[i]), theta, dir);
+ }
+ //printf ("\n");}
--- a/libcelt/bands.h
+++ b/libcelt/bands.h
@@ -51,4 +51,6 @@
void unquant_bands(const CELTMode *m, float *X, float *P, ec_dec *dec);
+void band_rotation(const CELTMode *m, float *X, int dir);
+
#endif /* BANDS_H */
--- a/libcelt/celt.c
+++ b/libcelt/celt.c
@@ -266,6 +266,9 @@
normalise_bands(st->mode, P, bandEp);
}
+ band_rotation(st->mode, X, -1);
+ band_rotation(st->mode, P, -1);
+
quant_energy(st->mode, bandE, st->oldBandE, &st->enc);
/* Pitch prediction */
@@ -295,6 +298,8 @@
//printf ("\n");}
+ band_rotation(st->mode, X, 1);
+
/* Synthesis */
denormalise_bands(st->mode, X, bandE);
@@ -495,6 +500,7 @@
compute_band_energies(st->mode, P, bandEp);
normalise_bands(st->mode, P, bandEp);
}
+ band_rotation(st->mode, P, -1);
/* Get the pitch gains */
unquant_pitch(gains, st->mode->nbPBands, &dec);
@@ -505,6 +511,8 @@
/* Decode fixed codebook and merge with pitch */
unquant_bands(st->mode, X, P, &dec);
+ band_rotation(st->mode, X, 1);
+
/* Synthesis */
denormalise_bands(st->mode, X, bandE);
--- a/libcelt/modes.c
+++ b/libcelt/modes.c
@@ -39,7 +39,7 @@
const int qpulses1[NBANDS128] = {7, 5, 5, 5, 4, 5, 4, 5, 5, 4, -2, 0, 0, 0, 0}; const int qpulses2[NBANDS128] = {28,24,20,16,24,20, 18, 12, 10, 10,-7, -4, 0, 0, 0};-const int qpulses2b[NBANDS128] = {32,28,24,20,28,24, 22, 18, 16, 15,-12, -12, 12, 12, 0};+const int qpulses2s[NBANDS128] = {38,30,24,20,24,20, 18, 16, 14, 20,-20,-14, -8, -8, -5}; const int pbank1[PBANDS128+2] = {0, 4, 8, 12, 20, PITCH_END128, 128};@@ -94,7 +94,7 @@
qpulses2 /**< nbPulses */
};
-/* Stereo mode (doesn't work yet) */
+/* Stereo mode around 120 kbps */
const CELTMode mode4 = {256, /**< frameSize */
128, /**< mdctSize */
@@ -107,7 +107,7 @@
qbank1, /**< eBands */
pbank1, /**< pBands*/
- qpulses1 /**< nbPulses */
+ qpulses2s /**< nbPulses */
};
const CELTMode const *celt_mode1 = &mode1;
--- a/libcelt/vq.c
+++ b/libcelt/vq.c
@@ -38,9 +38,9 @@
/* Improved algebraic pulse-base quantiser. The signal x is replaced by the sum of the pitch
a combination of pulses such that its norm is still equal to 1. The only difference with
the quantiser above is that the search is more complete. */
-void alg_quant(float *x, int N, int K, float *p, ec_enc *enc)
+void alg_quant(float *x, int N, int K, float *p, float alpha, ec_enc *enc)
{- int L = 5;
+ int L = 3;
//float tata[200];
float y[L][N];
int iy[L][N];
@@ -55,7 +55,6 @@
float Rpp=0, Rxp=0;
float gain[L];
int maxL = 1;
- float alpha = .9;
for (j=0;j<N;j++)
Rpp += p[j]*p[j];
@@ -188,9 +187,40 @@
// printf ("%d ", iy[0][i]);pulse2comb(N, K, comb, signs, iy[0]);
ec_enc_uint64(enc,icwrs64(N, K, comb, signs),ncwrs64(N, K));
+
+ /* Recompute the gain in one pass (to reduce errors) */
+ if (0) {+ float Ryp=0;
+ float Rpp=0;
+ float Ryy=0;
+ float g=0;
+ for (i=0;i<N;i++)
+ Rpp += p[i]*p[i];
+
+ for (i=0;i<N;i++)
+ Ryp += iy[0][i]*p[i];
+
+ for (i=0;i<N;i++)
+ y[0][i] = iy[0][i] - alpha*Ryp*p[i];
+
+ /* Recompute after the projection (I think it's right) */
+ Ryp = 0;
+ for (i=0;i<N;i++)
+ Ryp += y[0][i]*p[i];
+
+ for (i=0;i<N;i++)
+ Ryy += y[0][i]*y[0][i];
+
+ g = (sqrt(Ryp*Ryp + Ryy - Ryy*Rpp) - Ryp)/Ryy;
+
+ for (i=0;i<N;i++)
+ x[i] = p[i] + g*y[0][i];
+
+ }
+
}
-static const float pg[5] = {1.f, .82f, .75f, 0.7f, 0.6f};+static const float pg[5] = {1.f, .6f, .45f, 0.35f, 0.25f};/* Finds the right offset into Y and copy it */
void copy_quant(float *x, int N, int K, float *Y, int B, int N0, ec_enc *enc)
@@ -257,11 +287,11 @@
E = .8/sqrt(E);
for (j=0;j<N;j++)
P[j] *= E;
- alg_quant(x, N, K, P, enc);
+ alg_quant(x, N, K, P, 0, enc);
}
}
-void alg_unquant(float *x, int N, int K, float *p, ec_dec *dec)
+void alg_unquant(float *x, int N, int K, float *p, float alpha, ec_dec *dec)
{int i;
celt_uint64_t id;
@@ -269,7 +299,6 @@
int signs[K];
int iy[N];
float y[N];
- float alpha = .9;
float Rpp=0, Ryp=0, Ryy=0;
float g;
@@ -344,6 +373,6 @@
E = .8/sqrt(E);
for (j=0;j<N;j++)
P[j] *= E;
- alg_unquant(x, N, K, P, dec);
+ alg_unquant(x, N, K, P, 0, dec);
}
}
--- a/libcelt/vq.h
+++ b/libcelt/vq.h
@@ -39,9 +39,9 @@
/* Algebraic pulse-base quantiser. The signal x is replaced by the sum of the pitch
a combination of pulses such that its norm is still equal to 1. The only difference with
the quantiser above is that the search is more complete. */
-void alg_quant(float *x, int N, int K, float *p, ec_enc *enc);
+void alg_quant(float *x, int N, int K, float *p, float alpha, ec_enc *enc);
-void alg_unquant(float *x, int N, int K, float *p, ec_dec *dec);
+void alg_unquant(float *x, int N, int K, float *p, float alpha, ec_dec *dec);
/* Finds the right offset into Y and copy it */
void copy_quant(float *x, int N, int K, float *Y, int B, int N0, ec_enc *enc);