Modifying the L1 metric for tf analysis by merging together all frequencies.

Still need to make it use the right channel in stereo.
This commit is contained in:
Jean-Marc Valin 2010-10-15 02:18:47 -04:00
parent 4a8c1f11b1
commit 8200b2d16c

View file

@ -423,6 +423,26 @@ static const signed char tf_select_table[4][8] = {
{0, -2, 0, -3, 2, 0, 1 -1}, {0, -2, 0, -3, 2, 0, 1 -1},
}; };
celt_word32 l1_metric(const celt_norm *tmp, int N, int LM, int width)
{
int i, j;
static const celt_word16 sqrtM_1[4] = {Q15ONE, QCONST16(0.70711f,15), QCONST16(0.5f,15), QCONST16(0.35355f,15)};
celt_word32 L1;
celt_word16 bias;
L1=0;
for (i=0;i<1<<LM;i++)
{
celt_word32 L2 = 0;
for (j=0;j<N>>LM;j++)
L2 = MAC16_16(L2, tmp[(j<<LM)+i], tmp[(j<<LM)+i]);
L1 += celt_sqrt(L2);
}
L1 = MULT16_32_Q15(sqrtM_1[LM], L1);
bias = QCONST16(.25f,15)*LM/width;
L1 = MAC16_32_Q15(L1, bias, L1);
return L1;
}
static int tf_analysis(const CELTMode *m, celt_word16 *bandLogE, celt_word16 *oldBandE, static int tf_analysis(const CELTMode *m, celt_word16 *bandLogE, celt_word16 *oldBandE,
int len, int C, int isTransient, int *tf_res, int nbCompressedBytes, celt_norm *X, int len, int C, int isTransient, int *tf_res, int nbCompressedBytes, celt_norm *X,
int N0, int LM, int *tf_sum) int N0, int LM, int *tf_sum)
@ -452,7 +472,7 @@ static int tf_analysis(const CELTMode *m, celt_word16 *bandLogE, celt_word16 *ol
else if (nbCompressedBytes<100) else if (nbCompressedBytes<100)
lambda = 2; lambda = 2;
else else
lambda = 1; lambda = 2;
ALLOC(metric, len, int); ALLOC(metric, len, int);
ALLOC(tmp, (m->eBands[len]-m->eBands[len-1])<<LM, celt_norm); ALLOC(tmp, (m->eBands[len]-m->eBands[len-1])<<LM, celt_norm);
@ -468,31 +488,29 @@ static int tf_analysis(const CELTMode *m, celt_word16 *bandLogE, celt_word16 *ol
N = (m->eBands[i+1]-m->eBands[i])<<LM; N = (m->eBands[i+1]-m->eBands[i])<<LM;
for (j=0;j<N;j++) for (j=0;j<N;j++)
tmp[j] = X[j+(m->eBands[i]<<LM)]; tmp[j] = X[j+(m->eBands[i]<<LM)];
if (C==2) /* FIXME: Do something with the right channel */
/*if (C==2)
for (j=0;j<N;j++) for (j=0;j<N;j++)
tmp[j] = ADD16(tmp[j],X[N0+j+(m->eBands[i]<<LM)]); tmp[j] = ADD16(tmp[j],X[N0+j+(m->eBands[i]<<LM)]);*/
L1=0; L1 = l1_metric(tmp, N, isTransient ? LM : 0, N>>LM);
for (j=0;j<N;j++)
L1 += ABS16(tmp[j]);
/* Biasing towards better freq resolution (because of spreading) */
if (isTransient)
L1 += MULT16_32_Q15(QCONST16(.08,15), L1);
else
L1 -= MULT16_32_Q15(QCONST16(.08,15), L1);
best_L1 = L1; best_L1 = L1;
/*printf ("%f ", L1);*/ /*printf ("%f ", L1);*/
for (k=0;k<LM;k++) for (k=0;k<LM;k++)
{ {
int B;
if (isTransient)
B = (LM-k-1);
else
B = k+1;
if (isTransient) if (isTransient)
haar1(tmp, N>>(LM-k), 1<<(LM-k)); haar1(tmp, N>>(LM-k), 1<<(LM-k));
else else
haar1(tmp, N>>k, 1<<k); haar1(tmp, N>>k, 1<<k);
L1=0; L1 = l1_metric(tmp, N, B, N>>LM);
for (j=0;j<N;j++)
L1 += ABS16(tmp[j]);
/*printf ("%f ", L1);*/
if (L1 < best_L1) if (L1 < best_L1)
{ {
best_L1 = L1; best_L1 = L1;
@ -508,7 +526,7 @@ static int tf_analysis(const CELTMode *m, celt_word16 *bandLogE, celt_word16 *ol
} }
/*printf("\n");*/ /*printf("\n");*/
/* FIXME: Figure out how to set this */ /* FIXME: Figure out how to set this */
tf_select = 1; tf_select = 0;
cost0 = 0; cost0 = 0;
cost1 = lambda; cost1 = lambda;
@ -743,6 +761,12 @@ int celt_encode_with_ec_float(CELTEncoder * restrict st, const celt_sig * pcm, c
/* Band normalisation */ /* Band normalisation */
normalise_bands(st->mode, freq, X, bandE, effEnd, C, M); normalise_bands(st->mode, freq, X, bandE, effEnd, C, M);
ALLOC(tf_res, st->mode->nbEBands, int);
/* Needs to be before coarse energy quantization because otherwise the energy gets modified */
tf_select = tf_analysis(st->mode, bandLogE, oldBandE, effEnd, C, isTransient, tf_res, nbAvailableBytes, X, N, LM, &tf_sum);
for (i=effEnd;i<st->end;i++)
tf_res[i] = tf_res[effEnd-1];
NN = M*st->mode->eBands[effEnd]; NN = M*st->mode->eBands[effEnd];
if (shortBlocks && !transient_shift) if (shortBlocks && !transient_shift)
{ {
@ -790,12 +814,6 @@ int celt_encode_with_ec_float(CELTEncoder * restrict st, const celt_sig * pcm, c
mdct_shape(st->mode, X, mdct_weight_pos+1, M, N, mdct_weight_shift, effEnd, C, 0, M); mdct_shape(st->mode, X, mdct_weight_pos+1, M, N, mdct_weight_shift, effEnd, C, 0, M);
} }
ALLOC(tf_res, st->mode->nbEBands, int);
/* Needs to be before coarse energy quantization because otherwise the energy gets modified */
tf_select = tf_analysis(st->mode, bandLogE, oldBandE, effEnd, C, isTransient, tf_res, nbAvailableBytes, X, N, LM, &tf_sum);
for (i=effEnd;i<st->end;i++)
tf_res[i] = tf_res[effEnd-1];
ALLOC(error, C*st->mode->nbEBands, celt_word16); ALLOC(error, C*st->mode->nbEBands, celt_word16);
quant_coarse_energy(st->mode, st->start, st->end, effEnd, bandLogE, quant_coarse_energy(st->mode, st->start, st->end, effEnd, bandLogE,
oldBandE, nbCompressedBytes*8, st->mode->prob, oldBandE, nbCompressedBytes*8, st->mode->prob,