#include "aacenc_utils.h"
#include "aacenc_quantization.h"
-/*
- * Shifts the values as well if compression is possible.
- */
+/* Could be set to 3 to save an additional bit at the cost of little quality */
+#define TNS_Q_BITS 4
+
+/* Coefficient resolution in short windows */
+#define TNS_Q_BITS_IS8 4
+
+/* Define this to save a bit, be warned decoders can't deal with it
+ * so it is not lossless despite what the specifications say */
+// #define TNS_ENABLE_COEF_COMPRESSION
+
+/* TNS will only be used if the LPC gain is within these margins */
+#define TNS_GAIN_THRESHOLD_LOW 1.477f
+#define TNS_GAIN_THRESHOLD_HIGH 7.0f
+#define TNS_GAIN_THRESHOLD_LOW_IS8 0.16f*TNS_GAIN_THRESHOLD_LOW
+#define TNS_GAIN_THRESHOLD_HIGH_IS8 0.26f*TNS_GAIN_THRESHOLD_HIGH
+
static inline int compress_coeffs(int *coef, int order, int c_bits)
{
- int i, res = 0;
+ int i;
const int low_idx = c_bits ? 4 : 2;
const int shift_val = c_bits ? 8 : 4;
const int high_idx = c_bits ? 11 : 5;
+#ifndef TNS_ENABLE_COEF_COMPRESSION
+ return 0;
+#endif /* TNS_ENABLE_COEF_COMPRESSION */
+ for (i = 0; i < order; i++)
+ if (coef[i] >= low_idx && coef[i] <= high_idx)
+ return 0;
for (i = 0; i < order; i++)
- if (coef[i] < low_idx || coef[i] > high_idx)
- res++;
- if (res == order)
- for (i = 0; i < order; i++)
- coef[i] -= (coef[i] > high_idx) ? shift_val : 0;
- return res == order;
+ coef[i] -= (coef[i] > high_idx) ? shift_val : 0;
+ return 1;
}
/**
* Encode TNS data.
- * Coefficient compression saves a single bit per coefficient.
+ * Coefficient compression is simply not lossless as it should be
+ * on any decoder tested and as such is not active.
*/
void ff_aac_encode_tns_info(AACEncContext *s, SingleChannelElement *sce)
{
- int i, w, filt, coef_len, coef_compress = 0;
- const int is8 = sce->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE;
+ int i, w, filt, coef_compress = 0, coef_len;
TemporalNoiseShaping *tns = &sce->tns;
- const int c_bits = is8 ? TNS_Q_BITS_SHORT == 4 : TNS_Q_BITS == 4;
+ const int is8 = sce->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE;
+ const int c_bits = is8 ? TNS_Q_BITS_IS8 == 4 : TNS_Q_BITS == 4;
if (!sce->tns.present)
return;
for (i = 0; i < sce->ics.num_windows; i++) {
put_bits(&s->pb, 2 - is8, sce->tns.n_filt[i]);
- if (tns->n_filt[i]) {
- put_bits(&s->pb, 1, c_bits);
- for (filt = 0; filt < tns->n_filt[i]; filt++) {
- put_bits(&s->pb, 6 - 2 * is8, tns->length[i][filt]);
- put_bits(&s->pb, 5 - 2 * is8, tns->order[i][filt]);
- if (tns->order[i][filt]) {
- coef_compress = compress_coeffs(tns->coef_idx[i][filt],
- tns->order[i][filt], c_bits);
- put_bits(&s->pb, 1, !!tns->direction[i][filt]);
- put_bits(&s->pb, 1, !!coef_compress);
- coef_len = c_bits + 3 - coef_compress;
- for (w = 0; w < tns->order[i][filt]; w++)
- put_bits(&s->pb, coef_len, tns->coef_idx[i][filt][w]);
- }
- }
+ if (!tns->n_filt[i])
+ continue;
+ put_bits(&s->pb, 1, c_bits);
+ for (filt = 0; filt < tns->n_filt[i]; filt++) {
+ put_bits(&s->pb, 6 - 2 * is8, tns->length[i][filt]);
+ put_bits(&s->pb, 5 - 2 * is8, tns->order[i][filt]);
+ if (!tns->order[i][filt])
+ continue;
+ put_bits(&s->pb, 1, tns->direction[i][filt]);
+ coef_compress = compress_coeffs(tns->coef_idx[i][filt],
+ tns->order[i][filt], c_bits);
+ put_bits(&s->pb, 1, coef_compress);
+ coef_len = c_bits + 3 - coef_compress;
+ for (w = 0; w < tns->order[i][filt]; w++)
+ put_bits(&s->pb, coef_len, tns->coef_idx[i][filt][w]);
}
}
}
IndividualChannelStream *ics = &sce->ics;
int w, filt, m, i, top, order, bottom, start, end, size, inc;
const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
- float lpc[TNS_MAX_ORDER];
+ float lpc[TNS_MAX_ORDER], tmp[TNS_MAX_ORDER+1];
for (w = 0; w < ics->num_windows; w++) {
bottom = ics->num_swb;
}
start += w * 128;
- // ar filter
- for (m = 0; m < size; m++, start += inc)
- for (i = 1; i <= FFMIN(m, order); i++)
- sce->coeffs[start] += lpc[i-1]*sce->pcoeffs[start - i*inc];
+ if (!s->options.ltp) { // ar filter
+ for (m = 0; m < size; m++, start += inc) {
+ for (i = 1; i <= FFMIN(m, order); i++) {
+ sce->coeffs[start] += lpc[i-1]*sce->pcoeffs[start - i*inc];
+ }
+ }
+ } else { // ma filter
+ for (m = 0; m < size; m++, start += inc) {
+ tmp[0] = sce->pcoeffs[start];
+ for (i = 1; i <= FFMIN(m, order); i++)
+ sce->coeffs[start] += lpc[i-1]*tmp[i];
+ for (i = order; i > 0; i--)
+ tmp[i] = tmp[i - 1];
+ }
+ }
}
}
}
void ff_aac_search_for_tns(AACEncContext *s, SingleChannelElement *sce)
{
TemporalNoiseShaping *tns = &sce->tns;
+ double gain, coefs[MAX_LPC_ORDER];
int w, w2, g, count = 0;
const int mmm = FFMIN(sce->ics.tns_max_bands, sce->ics.max_sfb);
const int is8 = sce->ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE;
- const int c_bits = is8 ? TNS_Q_BITS_SHORT == 4 : TNS_Q_BITS == 4;
+ const int c_bits = is8 ? TNS_Q_BITS_IS8 == 4 : TNS_Q_BITS == 4;
+ const int slant = sce->ics.window_sequence[0] == LONG_STOP_SEQUENCE ? 1 :
+ sce->ics.window_sequence[0] == LONG_START_SEQUENCE ? 0 : 2;
int sfb_start = av_clip(tns_min_sfb[is8][s->samplerate_index], 0, mmm);
int sfb_end = av_clip(sce->ics.num_swb, 0, mmm);
+ int order = is8 ? 5 : s->profile == FF_PROFILE_AAC_LOW ? 12 : TNS_MAX_ORDER;
for (w = 0; w < sce->ics.num_windows; w++) {
- int use_tns;
- int order = is8 ? 5 : s->profile == FF_PROFILE_AAC_LOW ? 12 : TNS_MAX_ORDER;
+ float en[2] = {0.0f, 0.0f};
int coef_start = w*sce->ics.num_swb + sce->ics.swb_offset[sfb_start];
int coef_len = sce->ics.swb_offset[sfb_end] - sce->ics.swb_offset[sfb_start];
- float e_ratio = 0.0f, threshold = 0.0f, spread = 0.0f, en[2] = {0.0, 0.0f};
- double gain = 0.0f, coefs[MAX_LPC_ORDER] = {0};
for (g = 0; g < sce->ics.num_swb; g++) {
if (w*16+g < sfb_start || w*16+g > sfb_end)
en[1] += band->energy;
else
en[0] += band->energy;
- threshold += band->threshold;
- spread += band->spread;
}
}
coef_len, order, coefs);
if (!order || gain < TNS_GAIN_THRESHOLD_LOW || gain > TNS_GAIN_THRESHOLD_HIGH)
- use_tns = 0;
- else if ((en[0]+en[1]) < TNS_GAIN_THRESHOLD_LOW*threshold || spread < TNS_SPREAD_THRESHOLD)
- use_tns = 0;
- else
- use_tns = 1;
-
- if (use_tns) {
- e_ratio = en[0]/en[1];
- if (is8 || order < 2 || (e_ratio > TNS_E_RATIO_LOW && e_ratio < TNS_E_RATIO_HIGH)) {
- tns->n_filt[w] = 1;
- for (g = 0; g < tns->n_filt[w]; g++) {
- tns->length[w][g] = sfb_end - sfb_start;
- tns->direction[w][g] = en[0] < en[1];
- tns->order[w][g] = order;
- quantize_coefs(coefs, tns->coef_idx[w][g], tns->coef[w][g],
- order, c_bits);
- }
- } else { /* 2 filters due to energy disbalance */
- tns->n_filt[w] = 2;
- for (g = 0; g < tns->n_filt[w]; g++) {
- tns->direction[w][g] = en[g] < en[!g];
- tns->order[w][g] = !g ? order/2 : order - tns->order[w][g-1];
- tns->length[w][g] = !g ? (sfb_end - sfb_start)/2 : \
+ continue;
+ if (is8 && (gain < TNS_GAIN_THRESHOLD_LOW_IS8 || gain > TNS_GAIN_THRESHOLD_HIGH_IS8))
+ continue;
+ if (is8 || order < 2) {
+ tns->n_filt[w] = 1;
+ for (g = 0; g < tns->n_filt[w]; g++) {
+ tns->length[w][g] = sfb_end - sfb_start;
+ tns->direction[w][g] = slant != 2 ? slant : en[0] < en[1];
+ tns->order[w][g] = order;
+ quantize_coefs(coefs, tns->coef_idx[w][g], tns->coef[w][g],
+ order, c_bits);
+ }
+ } else { /* 2 filters due to energy disbalance */
+ tns->n_filt[w] = 2;
+ for (g = 0; g < tns->n_filt[w]; g++) {
+ tns->direction[w][g] = slant != 2 ? slant : en[g] < en[!g];
+ tns->order[w][g] = !g ? order/2 : order - tns->order[w][g-1];
+ tns->length[w][g] = !g ? (sfb_end - sfb_start)/2 : \
(sfb_end - sfb_start) - tns->length[w][g-1];
- quantize_coefs(&coefs[!g ? 0 : order - tns->order[w][g-1]],
- tns->coef_idx[w][g], tns->coef[w][g],
- tns->order[w][g], c_bits);
- }
+ quantize_coefs(&coefs[!g ? 0 : order - tns->order[w][g-1]],
+ tns->coef_idx[w][g], tns->coef[w][g],
+ tns->order[w][g], c_bits);
}
- count++;
}
+ count += tns->n_filt[w];
}
sce->tns.present = !!count;
}
projects / ffmpeg / blobdiff