f6e781ab8b
- Merged the LPC stabilization from NLSF2A_stable.c into NLSF2A.c - The bandwidth expansion in NLSF2A() now operates on int32 LPC coefficients in Q17 domain (instead of int16 Q12 coefficients) - The function bwexpander_32() has a more precise way of updating the chirp variable (round to nearest, instead of round down) - Changed a few variables in NLSF_stabilize() from int16 to int32 to avoid signed wrap-around (no difference in results as the wrap-around would always be reversed later) - The LSF codebook for WB speech has a quantization stepsize of 0.15 (was 0.16). This doesn't break the bitstream, although it slightly limits quality of signals encoded with the old version and decoded with the new one (I can't really hear it and PESQ gives high scores as well). I does improve handling of tonal signals. - As discussed: the Q-domain of the poly function is now in Q16 (was Q20) - As discussed: limiting the LSFs in NLSF_decode() to 0...32767 - The silk_NLSF_DELTA_MIN values were lowered to deal with a possible future situation with less or no input HP filtering.
183 lines
9.5 KiB
C
183 lines
9.5 KiB
C
/***********************************************************************
|
|
Copyright (c) 2006-2011, Skype Limited. All rights reserved.
|
|
Redistribution and use in source and binary forms, with or without
|
|
modification, (subject to the limitations in the disclaimer below)
|
|
are permitted provided that the following conditions are met:
|
|
- Redistributions of source code must retain the above copyright notice,
|
|
this list of conditions and the following disclaimer.
|
|
- Redistributions in binary form must reproduce the above copyright
|
|
notice, this list of conditions and the following disclaimer in the
|
|
documentation and/or other materials provided with the distribution.
|
|
- Neither the name of Skype Limited, nor the names of specific
|
|
contributors, may be used to endorse or promote products derived from
|
|
this software without specific prior written permission.
|
|
NO EXPRESS OR IMPLIED LICENSES TO ANY PARTY'S PATENT RIGHTS ARE GRANTED
|
|
BY THIS LICENSE. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
|
|
CONTRIBUTORS ''AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING,
|
|
BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
|
|
FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
|
|
COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
|
|
INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
|
|
NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
|
|
USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
|
|
ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
|
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
|
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
***********************************************************************/
|
|
|
|
#include "silk_main.h"
|
|
|
|
#define STORE_LSF_DATA_FOR_TRAINING 0
|
|
|
|
/***********************/
|
|
/* NLSF vector encoder */
|
|
/***********************/
|
|
SKP_int32 silk_NLSF_encode( /* O Returns RD value in Q25 */
|
|
SKP_int8 *NLSFIndices, /* I Codebook path vector [ LPC_ORDER + 1 ] */
|
|
SKP_int16 *pNLSF_Q15, /* I/O Quantized NLSF vector [ LPC_ORDER ] */
|
|
const silk_NLSF_CB_struct *psNLSF_CB, /* I Codebook object */
|
|
const SKP_int16 *pW_QW, /* I NLSF weight vector [ LPC_ORDER ] */
|
|
const SKP_int NLSF_mu_Q20, /* I Rate weight for the RD optimization */
|
|
const SKP_int nSurvivors, /* I Max survivors after first stage */
|
|
const SKP_int signalType /* I Signal type: 0/1/2 */
|
|
)
|
|
{
|
|
SKP_int i, s, ind1, bestIndex, prob_Q8, bits_q7;
|
|
SKP_int32 W_tmp_Q9;
|
|
SKP_int32 err_Q26[ NLSF_VQ_MAX_VECTORS ];
|
|
SKP_int32 RD_Q25[ NLSF_VQ_MAX_SURVIVORS ];
|
|
SKP_int tempIndices1[ NLSF_VQ_MAX_SURVIVORS ];
|
|
SKP_int8 tempIndices2[ NLSF_VQ_MAX_SURVIVORS * MAX_LPC_ORDER ];
|
|
SKP_int16 res_Q15[ MAX_LPC_ORDER ];
|
|
SKP_int16 res_Q10[ MAX_LPC_ORDER ];
|
|
SKP_int16 NLSF_tmp_Q15[ MAX_LPC_ORDER ];
|
|
SKP_int16 W_tmp_QW[ MAX_LPC_ORDER ];
|
|
SKP_int16 W_adj_Q5[ MAX_LPC_ORDER ];
|
|
SKP_uint8 pred_Q8[ MAX_LPC_ORDER ];
|
|
SKP_int16 ec_ix[ MAX_LPC_ORDER ];
|
|
const SKP_uint8 *pCB_element, *iCDF_ptr;
|
|
|
|
#if STORE_LSF_DATA_FOR_TRAINING
|
|
SKP_int16 pNLSF_Q15_orig[MAX_LPC_ORDER ];
|
|
DEBUG_STORE_DATA( NLSF.dat, pNLSF_Q15, psNLSF_CB->order * sizeof( SKP_int16 ) );
|
|
DEBUG_STORE_DATA( WNLSF.dat, pW_Q5, psNLSF_CB->order * sizeof( SKP_int16 ) );
|
|
DEBUG_STORE_DATA( NLSF_mu.dat, &NLSF_mu_Q20, sizeof( SKP_int ) );
|
|
DEBUG_STORE_DATA( sigType.dat, &signalType, sizeof( SKP_int ) );
|
|
SKP_memcpy(pNLSF_Q15_orig, pNLSF_Q15, sizeof( pNLSF_Q15_orig ));
|
|
#endif
|
|
|
|
SKP_assert( nSurvivors <= NLSF_VQ_MAX_SURVIVORS );
|
|
SKP_assert( signalType >= 0 && signalType <= 2 );
|
|
SKP_assert( NLSF_mu_Q20 <= 32767 && NLSF_mu_Q20 >= 0 );
|
|
|
|
/* NLSF stabilization */
|
|
silk_NLSF_stabilize( pNLSF_Q15, psNLSF_CB->deltaMin_Q15, psNLSF_CB->order );
|
|
|
|
/* First stage: VQ */
|
|
silk_NLSF_VQ( err_Q26, pNLSF_Q15, psNLSF_CB->CB1_NLSF_Q8, psNLSF_CB->nVectors, psNLSF_CB->order );
|
|
|
|
/* Sort the quantization errors */
|
|
silk_insertion_sort_increasing( err_Q26, tempIndices1, psNLSF_CB->nVectors, nSurvivors );
|
|
|
|
/* Loop over survivors */
|
|
for( s = 0; s < nSurvivors; s++ ) {
|
|
ind1 = tempIndices1[ s ];
|
|
|
|
/* Residual after first stage */
|
|
pCB_element = &psNLSF_CB->CB1_NLSF_Q8[ ind1 * psNLSF_CB->order ];
|
|
for( i = 0; i < psNLSF_CB->order; i++ ) {
|
|
NLSF_tmp_Q15[ i ] = SKP_LSHIFT16( ( SKP_int16 )pCB_element[ i ], 7 );
|
|
res_Q15[ i ] = pNLSF_Q15[ i ] - NLSF_tmp_Q15[ i ];
|
|
}
|
|
|
|
/* Weights from codebook vector */
|
|
silk_NLSF_VQ_weights_laroia( W_tmp_QW, NLSF_tmp_Q15, psNLSF_CB->order );
|
|
|
|
/* Apply square-rooted weights */
|
|
for( i = 0; i < psNLSF_CB->order; i++ ) {
|
|
W_tmp_Q9 = silk_SQRT_APPROX( SKP_LSHIFT( ( SKP_int32 )W_tmp_QW[ i ], 18 - NLSF_W_Q ) );
|
|
res_Q10[ i ] = ( SKP_int16 )SKP_RSHIFT( SKP_SMULBB( res_Q15[ i ], W_tmp_Q9 ), 14 );
|
|
}
|
|
|
|
/* Modify input weights accordingly */
|
|
for( i = 0; i < psNLSF_CB->order; i++ ) {
|
|
W_adj_Q5[ i ] = SKP_DIV32_16( SKP_LSHIFT( ( SKP_int32 )pW_QW[ i ], 5 ), W_tmp_QW[ i ] );
|
|
}
|
|
|
|
/* Unpack entropy table indices and predictor for current CB1 index */
|
|
silk_NLSF_unpack( ec_ix, pred_Q8, psNLSF_CB, ind1 );
|
|
|
|
/* Trellis quantizer */
|
|
RD_Q25[ s ] = silk_NLSF_del_dec_quant( &tempIndices2[ s * MAX_LPC_ORDER ], res_Q10, W_adj_Q5, pred_Q8, ec_ix,
|
|
psNLSF_CB->ec_Rates_Q5, psNLSF_CB->quantStepSize_Q16, psNLSF_CB->invQuantStepSize_Q6, NLSF_mu_Q20, psNLSF_CB->order );
|
|
|
|
/* Add rate for first stage */
|
|
iCDF_ptr = &psNLSF_CB->CB1_iCDF[ ( signalType >> 1 ) * psNLSF_CB->nVectors ];
|
|
if( ind1 == 0 ) {
|
|
prob_Q8 = 256 - iCDF_ptr[ ind1 ];
|
|
} else {
|
|
prob_Q8 = iCDF_ptr[ ind1 - 1 ] - iCDF_ptr[ ind1 ];
|
|
}
|
|
bits_q7 = ( 8 << 7 ) - silk_lin2log( prob_Q8 );
|
|
RD_Q25[ s ] = SKP_SMLABB( RD_Q25[ s ], bits_q7, SKP_RSHIFT( NLSF_mu_Q20, 2 ) );
|
|
}
|
|
|
|
/* Find the lowest rate-distortion error */
|
|
silk_insertion_sort_increasing( RD_Q25, &bestIndex, nSurvivors, 1 );
|
|
|
|
NLSFIndices[ 0 ] = ( SKP_int8 )tempIndices1[ bestIndex ];
|
|
SKP_memcpy( &NLSFIndices[ 1 ], &tempIndices2[ bestIndex * MAX_LPC_ORDER ], psNLSF_CB->order * sizeof( SKP_int8 ) );
|
|
|
|
/* Decode */
|
|
silk_NLSF_decode( pNLSF_Q15, NLSFIndices, psNLSF_CB );
|
|
|
|
#if STORE_LSF_DATA_FOR_TRAINING
|
|
{
|
|
/* code for training the codebooks */
|
|
SKP_int32 RD_dec_Q22, Dist_Q22_dec, Rate_Q7, diff_Q15;
|
|
ind1 = NLSFIndices[ 0 ];
|
|
silk_NLSF_unpack( ec_ix, pred_Q8, psNLSF_CB, ind1 );
|
|
|
|
pCB_element = &psNLSF_CB->CB1_NLSF_Q8[ ind1 * psNLSF_CB->order ];
|
|
for( i = 0; i < psNLSF_CB->order; i++ ) {
|
|
NLSF_tmp_Q15[ i ] = SKP_LSHIFT16( ( SKP_int16 )pCB_element[ i ], 7 );
|
|
}
|
|
silk_NLSF_VQ_weights_laroia( W_tmp_QW, NLSF_tmp_Q15, psNLSF_CB->order );
|
|
for( i = 0; i < psNLSF_CB->order; i++ ) {
|
|
W_tmp_Q9 = silk_SQRT_APPROX( SKP_LSHIFT( ( SKP_int32 )W_tmp_QW[ i ], 18 - NLSF_W_Q ) );
|
|
res_Q15[ i ] = pNLSF_Q15_orig[ i ] - NLSF_tmp_Q15[ i ];
|
|
res_Q10[ i ] = (SKP_int16)SKP_RSHIFT( SKP_SMULBB( res_Q15[ i ], W_tmp_Q9 ), 14 );
|
|
DEBUG_STORE_DATA( NLSF_res_q10.dat, &res_Q10[ i ], sizeof( SKP_int16 ) );
|
|
res_Q15[ i ] = pNLSF_Q15[ i ] - NLSF_tmp_Q15[ i ];
|
|
res_Q10[ i ] = (SKP_int16)SKP_RSHIFT( SKP_SMULBB( res_Q15[ i ], W_tmp_Q9 ), 14 );
|
|
DEBUG_STORE_DATA( NLSF_resq_q10.dat, &res_Q10[ i ], sizeof( SKP_int16 ) );
|
|
}
|
|
|
|
Dist_Q22_dec = 0;
|
|
for( i = 0; i < psNLSF_CB->order; i++ ) {
|
|
diff_Q15 = pNLSF_Q15_orig[ i ] - pNLSF_Q15[ i ];
|
|
Dist_Q22_dec += ( ( (diff_Q15 >> 5) * (diff_Q15 >> 5) ) * pW_Q5[ i ] ) >> 3;
|
|
}
|
|
iCDF_ptr = &psNLSF_CB->CB1_iCDF[ ( signalType >> 1 ) * psNLSF_CB->nVectors ];
|
|
if( ind1 == 0 ) {
|
|
prob_Q8 = 256 - iCDF_ptr[ ind1 ];
|
|
} else {
|
|
prob_Q8 = iCDF_ptr[ ind1 - 1 ] - iCDF_ptr[ ind1 ];
|
|
}
|
|
Rate_Q7 = ( 8 << 7 ) - silk_lin2log( prob_Q8 );
|
|
for( i = 0; i < psNLSF_CB->order; i++ ) {
|
|
Rate_Q7 += ((int)psNLSF_CB->ec_Rates_Q5[ ec_ix[ i ] + SKP_LIMIT( NLSFIndices[ i + 1 ] + NLSF_QUANT_MAX_AMPLITUDE, 0, 2 * NLSF_QUANT_MAX_AMPLITUDE ) ] ) << 2;
|
|
if( SKP_abs( NLSFIndices[ i + 1 ] ) >= NLSF_QUANT_MAX_AMPLITUDE ) {
|
|
Rate_Q7 += 128 << ( SKP_abs( NLSFIndices[ i + 1 ] ) - NLSF_QUANT_MAX_AMPLITUDE );
|
|
}
|
|
}
|
|
RD_dec_Q22 = Dist_Q22_dec + Rate_Q7 * NLSF_mu_Q20 >> 5;
|
|
DEBUG_STORE_DATA( dec_dist_q22.dat, &Dist_Q22_dec, sizeof( SKP_int32 ) );
|
|
DEBUG_STORE_DATA( dec_rate_q7.dat, &Rate_Q7, sizeof( SKP_int32 ) );
|
|
DEBUG_STORE_DATA( dec_rd_q22.dat, &RD_dec_Q22, sizeof( SKP_int32 ) );
|
|
}
|
|
DEBUG_STORE_DATA( NLSF_ind.dat, NLSFIndices, (psNLSF_CB->order+1) * sizeof( SKP_int8 ) );
|
|
#endif
|
|
|
|
return RD_Q25[ 0 ];
|
|
}
|