SILK fixes following last codec WG meeting

decoder:
- fixed incorrect scaling of filter states for the smallest quantization
  step sizes
- NLSF2A now limits the prediction gain of LPC filters

encoder:
- increased damping of LTP coefficients in LTP analysis
- increased white noise fraction in noise shaping LPC analysis
- introduced maximum total prediction gain.  Used by Burg's method to
  exit early if prediction gain is exceeded.  This improves packet
  loss robustness and numerical robustness in Burg's method
- Prefiltered signal is now in int32 Q10 domain, from int16 Q0
- Increased max number of iterations in CBR gain control loop from 5 to 6
- Removed useless code from LTP scaling control
- Optimization: smarter LPC loop unrolling
- Switched default win32 compile mode to be floating-point

resampler:
- made resampler have constant delay of 0.75 ms; removed delay
  compensation from silk code.
- removed obsolete table entries (~850 Bytes)
- increased downsampling filter order from 16 to 18/24/36 (depending on
  frequency ratio)
- reoptimized filter coefficients
This commit is contained in:
Koen Vos 2011-12-13 14:47:31 -05:00 committed by Jean-Marc Valin
parent 6619a73637
commit bf75c8ec4d
71 changed files with 961 additions and 1005 deletions

View file

@ -41,11 +41,11 @@ void silk_decode_core(
const opus_int pulses[ MAX_FRAME_LENGTH ] /* I Pulse signal */
)
{
opus_int i, j, k, lag = 0, start_idx, sLTP_buf_idx, NLSF_interpolation_flag, signalType;
opus_int i, k, lag = 0, start_idx, sLTP_buf_idx, NLSF_interpolation_flag, signalType;
opus_int16 *A_Q12, *B_Q14, *pxq, A_Q12_tmp[ MAX_LPC_ORDER ];
opus_int16 sLTP[ MAX_FRAME_LENGTH ];
opus_int32 sLTP_Q16[ 2 * MAX_FRAME_LENGTH ];
opus_int32 LTP_pred_Q14, LPC_pred_Q10, Gain_Q10, inv_gain_Q16, inv_gain_Q32, gain_adj_Q16, rand_seed, offset_Q10;
opus_int32 sLTP_Q15[ 2 * MAX_FRAME_LENGTH ];
opus_int32 LTP_pred_Q13, LPC_pred_Q10, Gain_Q10, inv_gain_Q16, inv_gain_Q31, gain_adj_Q16, rand_seed, offset_Q10;
opus_int32 *pred_lag_ptr, *pexc_Q10, *pres_Q10;
opus_int32 res_Q10[ MAX_SUB_FRAME_LENGTH ];
opus_int32 sLPC_Q14[ MAX_SUB_FRAME_LENGTH + MAX_LPC_ORDER ];
@ -64,7 +64,7 @@ void silk_decode_core(
rand_seed = psDec->indices.Seed;
for( i = 0; i < psDec->frame_length; i++ ) {
rand_seed = silk_RAND( rand_seed );
psDec->exc_Q10[ i ] = silk_LSHIFT( ( opus_int32 )pulses[ i ], 10 );
psDec->exc_Q10[ i ] = silk_LSHIFT( (opus_int32)pulses[ i ], 10 );
if( psDec->exc_Q10[ i ] > 0 ) {
psDec->exc_Q10[ i ] -= QUANT_LEVEL_ADJUST_Q10;
} else
@ -95,7 +95,6 @@ void silk_decode_core(
Gain_Q10 = silk_RSHIFT( psDecCtrl->Gains_Q16[ k ], 6 );
inv_gain_Q16 = silk_INVERSE32_varQ( psDecCtrl->Gains_Q16[ k ], 32 );
inv_gain_Q16 = silk_min( inv_gain_Q16, silk_int16_MAX );
/* Calculate Gain adjustment factor */
gain_adj_Q16 = 1 << 16;
@ -141,19 +140,19 @@ void silk_decode_core(
A_Q12, psDec->ltp_mem_length - start_idx, psDec->LPC_order );
/* After rewhitening the LTP state is unscaled */
inv_gain_Q32 = silk_LSHIFT( inv_gain_Q16, 16 );
inv_gain_Q31 = silk_LSHIFT( inv_gain_Q16, 15 );
if( k == 0 ) {
/* Do LTP downscaling to reduce inter-packet dependency */
inv_gain_Q32 = silk_LSHIFT( silk_SMULWB( inv_gain_Q32, psDecCtrl->LTP_scale_Q14 ), 2 );
inv_gain_Q31 = silk_LSHIFT( silk_SMULWB( inv_gain_Q31, psDecCtrl->LTP_scale_Q14 ), 2 );
}
for( i = 0; i < lag + LTP_ORDER/2; i++ ) {
sLTP_Q16[ sLTP_buf_idx - i - 1 ] = silk_SMULWB( inv_gain_Q32, sLTP[ psDec->ltp_mem_length - i - 1 ] );
sLTP_Q15[ sLTP_buf_idx - i - 1 ] = silk_SMULWB( inv_gain_Q31, sLTP[ psDec->ltp_mem_length - i - 1 ] );
}
} else {
/* Update LTP state when Gain changes */
if( gain_adj_Q16 != 1 << 16 ) {
for( i = 0; i < lag + LTP_ORDER/2; i++ ) {
sLTP_Q16[ sLTP_buf_idx - i - 1 ] = silk_SMULWW( gain_adj_Q16, sLTP_Q16[ sLTP_buf_idx - i - 1 ] );
sLTP_Q15[ sLTP_buf_idx - i - 1 ] = silk_SMULWW( gain_adj_Q16, sLTP_Q15[ sLTP_buf_idx - i - 1 ] );
}
}
}
@ -161,22 +160,22 @@ void silk_decode_core(
/* Long-term prediction */
if( signalType == TYPE_VOICED ) {
/* Setup pointer */
pred_lag_ptr = &sLTP_Q16[ sLTP_buf_idx - lag + LTP_ORDER / 2 ];
/* Set up pointer */
pred_lag_ptr = &sLTP_Q15[ sLTP_buf_idx - lag + LTP_ORDER / 2 ];
for( i = 0; i < psDec->subfr_length; i++ ) {
/* Unrolled loop */
LTP_pred_Q14 = silk_SMULWB( pred_lag_ptr[ 0 ], B_Q14[ 0 ] );
LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ -1 ], B_Q14[ 1 ] );
LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ -2 ], B_Q14[ 2 ] );
LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ -3 ], B_Q14[ 3 ] );
LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ -4 ], B_Q14[ 4 ] );
LTP_pred_Q13 = silk_SMULWB( pred_lag_ptr[ 0 ], B_Q14[ 0 ] );
LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ -1 ], B_Q14[ 1 ] );
LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ -2 ], B_Q14[ 2 ] );
LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ -3 ], B_Q14[ 3 ] );
LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ -4 ], B_Q14[ 4 ] );
pred_lag_ptr++;
/* Generate LPC excitation */
pres_Q10[ i ] = silk_ADD32( pexc_Q10[ i ], silk_RSHIFT_ROUND( LTP_pred_Q14, 4 ) );
pres_Q10[ i ] = silk_ADD32( pexc_Q10[ i ], silk_RSHIFT_ROUND( LTP_pred_Q13, 3 ) );
/* Update states */
sLTP_Q16[ sLTP_buf_idx ] = silk_LSHIFT( pres_Q10[ i ], 6 );
sLTP_Q15[ sLTP_buf_idx ] = silk_LSHIFT( pres_Q10[ i ], 5 );
sLTP_buf_idx++;
}
} else {
@ -184,7 +183,8 @@ void silk_decode_core(
}
for( i = 0; i < psDec->subfr_length; i++ ) {
/* Partially unrolled */
/* Short-term prediction */
silk_assert( psDec->LPC_order == 10 || psDec->LPC_order == 16 );
LPC_pred_Q10 = silk_SMULWB( sLPC_Q14[ MAX_LPC_ORDER + i - 1 ], A_Q12_tmp[ 0 ] );
LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 2 ], A_Q12_tmp[ 1 ] );
LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 3 ], A_Q12_tmp[ 2 ] );
@ -195,17 +195,23 @@ void silk_decode_core(
LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 8 ], A_Q12_tmp[ 7 ] );
LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 9 ], A_Q12_tmp[ 8 ] );
LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 10 ], A_Q12_tmp[ 9 ] );
for( j = 10; j < psDec->LPC_order; j++ ) {
LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - j - 1 ], A_Q12_tmp[ j ] );
if( psDec->LPC_order == 16 ) {
LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 11 ], A_Q12_tmp[ 10 ] );
LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 12 ], A_Q12_tmp[ 11 ] );
LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 13 ], A_Q12_tmp[ 12 ] );
LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 14 ], A_Q12_tmp[ 13 ] );
LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 15 ], A_Q12_tmp[ 14 ] );
LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 16 ], A_Q12_tmp[ 15 ] );
}
/* Add prediction to LPC excitation */
sLPC_Q14[ MAX_LPC_ORDER + i ] = silk_LSHIFT( silk_ADD32( pres_Q10[ i ], LPC_pred_Q10 ), 4 );
/* Scale with Gain */
pxq[ i ] = ( opus_int16 )silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( sLPC_Q14[ MAX_LPC_ORDER + i ], Gain_Q10 ), 8 ) );
pxq[ i ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( sLPC_Q14[ MAX_LPC_ORDER + i ], Gain_Q10 ), 8 ) );
}
/* DEBUG_STORE_DATA( dec.pcm, pxq, psDec->subfr_length * sizeof( opus_int16 ) ) */
/* Update LPC filter state */
silk_memcpy( sLPC_Q14, &sLPC_Q14[ psDec->subfr_length ], MAX_LPC_ORDER * sizeof( opus_int32 ) );
pexc_Q10 += psDec->subfr_length;