ref: d3358b1d42f5092c42f45a68f1e53a16b5a6926f
dir: /silk/silk_pitch_analysis_core.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.
***********************************************************************/
/***********************************************************
* Pitch analyser function
********************************************************** */
#include "silk_SigProc_FIX.h"
#include "silk_pitch_est_defines.h"
#include "silk_debug.h"
#define SCRATCH_SIZE 22
/************************************************************/
/* Internally used functions */
/************************************************************/
void silk_P_Ana_calc_corr_st3(
SKP_int32 cross_corr_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ],/* (O) 3 DIM correlation array */
const SKP_int16 signal[], /* I vector to correlate */
SKP_int start_lag, /* I lag offset to search around */
SKP_int sf_length, /* I length of a 5 ms subframe */
SKP_int nb_subfr, /* I number of subframes */
SKP_int complexity /* I Complexity setting */
);
void silk_P_Ana_calc_energy_st3(
SKP_int32 energies_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ],/* (O) 3 DIM energy array */
const SKP_int16 signal[], /* I vector to calc energy in */
SKP_int start_lag, /* I lag offset to search around */
SKP_int sf_length, /* I length of one 5 ms subframe */
SKP_int nb_subfr, /* I number of subframes */
SKP_int complexity /* I Complexity setting */
);
SKP_int32 silk_P_Ana_find_scaling(
const SKP_int16 *signal,
const SKP_int signal_length,
const SKP_int sum_sqr_len
);
/*************************************************************/
/* FIXED POINT CORE PITCH ANALYSIS FUNCTION */
/*************************************************************/
SKP_int silk_pitch_analysis_core( /* O Voicing estimate: 0 voiced, 1 unvoiced */
const SKP_int16 *signal, /* I Signal of length PE_FRAME_LENGTH_MS*Fs_kHz */
SKP_int *pitch_out, /* O 4 pitch lag values */
SKP_int16 *lagIndex, /* O Lag Index */
SKP_int8 *contourIndex, /* O Pitch contour Index */
SKP_int *LTPCorr_Q15, /* I/O Normalized correlation; input: value from previous frame */
SKP_int prevLag, /* I Last lag of previous frame; set to zero is unvoiced */
const SKP_int32 search_thres1_Q16, /* I First stage threshold for lag candidates 0 - 1 */
const SKP_int search_thres2_Q15, /* I Final threshold for lag candidates 0 - 1 */
const SKP_int Fs_kHz, /* I Sample frequency (kHz) */
const SKP_int complexity, /* I Complexity setting, 0-2, where 2 is highest */
const SKP_int nb_subfr /* I number of 5 ms subframes */
)
{
SKP_int16 signal_8kHz[ PE_MAX_FRAME_LENGTH_ST_2 ];
SKP_int16 signal_4kHz[ PE_MAX_FRAME_LENGTH_ST_1 ];
SKP_int32 filt_state[ 6 ];
SKP_int32 scratch_mem[ 3 * PE_MAX_FRAME_LENGTH ];
SKP_int16 *input_signal_ptr;
SKP_int i, k, d, j;
SKP_int16 C[ PE_MAX_NB_SUBFR ][ ( PE_MAX_LAG >> 1 ) + 5 ];
const SKP_int16 *target_ptr, *basis_ptr;
SKP_int32 cross_corr, normalizer, energy, shift, energy_basis, energy_target;
SKP_int d_srch[ PE_D_SRCH_LENGTH ], Cmax, length_d_srch, length_d_comp;
SKP_int16 d_comp[ ( PE_MAX_LAG >> 1 ) + 5 ];
SKP_int32 sum, threshold, temp32, lag_counter;
SKP_int CBimax, CBimax_new, CBimax_old, lag, start_lag, end_lag, lag_new;
SKP_int32 CC[ PE_NB_CBKS_STAGE2_EXT ], CCmax, CCmax_b, CCmax_new_b, CCmax_new;
SKP_int32 energies_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ];
SKP_int32 crosscorr_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ];
SKP_int frame_length, frame_length_8kHz, frame_length_4kHz, max_sum_sq_length;
SKP_int sf_length, sf_length_8kHz, sf_length_4kHz;
SKP_int min_lag, min_lag_8kHz, min_lag_4kHz;
SKP_int max_lag, max_lag_8kHz, max_lag_4kHz;
SKP_int32 contour_bias_Q20, diff, lz, lshift;
SKP_int nb_cbk_search, cbk_size;
SKP_int32 delta_lag_log2_sqr_Q7, lag_log2_Q7, prevLag_log2_Q7, prev_lag_bias_Q15, corr_thres_Q15;
const SKP_int8 *Lag_CB_ptr;
/* Check for valid sampling frequency */
SKP_assert( Fs_kHz == 8 || Fs_kHz == 12 || Fs_kHz == 16 );
/* Check for valid complexity setting */
SKP_assert( complexity >= SILK_PE_MIN_COMPLEX );
SKP_assert( complexity <= SILK_PE_MAX_COMPLEX );
SKP_assert( search_thres1_Q16 >= 0 && search_thres1_Q16 <= (1<<16) );
SKP_assert( search_thres2_Q15 >= 0 && search_thres2_Q15 <= (1<<15) );
/* Setup frame lengths max / min lag for the sampling frequency */
frame_length = ( PE_LTP_MEM_LENGTH_MS + nb_subfr * PE_SUBFR_LENGTH_MS ) * Fs_kHz;
frame_length_4kHz = ( PE_LTP_MEM_LENGTH_MS + nb_subfr * PE_SUBFR_LENGTH_MS ) * 4;
frame_length_8kHz = ( PE_LTP_MEM_LENGTH_MS + nb_subfr * PE_SUBFR_LENGTH_MS ) * 8;
sf_length = PE_SUBFR_LENGTH_MS * Fs_kHz;
sf_length_4kHz = PE_SUBFR_LENGTH_MS * 4;
sf_length_8kHz = PE_SUBFR_LENGTH_MS * 8;
min_lag = PE_MIN_LAG_MS * Fs_kHz;
min_lag_4kHz = PE_MIN_LAG_MS * 4;
min_lag_8kHz = PE_MIN_LAG_MS * 8;
max_lag = PE_MAX_LAG_MS * Fs_kHz - 1;
max_lag_4kHz = PE_MAX_LAG_MS * 4;
max_lag_8kHz = PE_MAX_LAG_MS * 8 - 1;
SKP_memset( C, 0, sizeof( SKP_int16 ) * nb_subfr * ( ( PE_MAX_LAG >> 1 ) + 5) );
/* Resample from input sampled at Fs_kHz to 8 kHz */
if( Fs_kHz == 16 ) {
SKP_memset( filt_state, 0, 2 * sizeof( SKP_int32 ) );
silk_resampler_down2( filt_state, signal_8kHz, signal, frame_length );
} else if ( Fs_kHz == 12 ) {
SKP_memset( filt_state, 0, 6 * sizeof( SKP_int32 ) );
silk_resampler_down2_3( filt_state, signal_8kHz, signal, frame_length );
} else {
SKP_assert( Fs_kHz == 8 );
SKP_memcpy( signal_8kHz, signal, frame_length_8kHz * sizeof(SKP_int16) );
}
/* Decimate again to 4 kHz */
SKP_memset( filt_state, 0, 2 * sizeof( SKP_int32 ) );/* Set state to zero */
silk_resampler_down2( filt_state, signal_4kHz, signal_8kHz, frame_length_8kHz );
/* Low-pass filter */
for( i = frame_length_4kHz - 1; i > 0; i-- ) {
signal_4kHz[ i ] = SKP_ADD_SAT16( signal_4kHz[ i ], signal_4kHz[ i - 1 ] );
}
/*******************************************************************************
** Scale 4 kHz signal down to prevent correlations measures from overflowing
** find scaling as max scaling for each 8kHz(?) subframe
*******************************************************************************/
/* Inner product is calculated with different lengths, so scale for the worst case */
max_sum_sq_length = SKP_max_32( sf_length_8kHz, SKP_LSHIFT( sf_length_4kHz, 2 ) );
shift = silk_P_Ana_find_scaling( signal_4kHz, frame_length_4kHz, max_sum_sq_length );
if( shift > 0 ) {
for( i = 0; i < frame_length_4kHz; i++ ) {
signal_4kHz[ i ] = SKP_RSHIFT( signal_4kHz[ i ], shift );
}
}
/******************************************************************************
* FIRST STAGE, operating in 4 khz
******************************************************************************/
target_ptr = &signal_4kHz[ SKP_LSHIFT( sf_length_4kHz, 2 ) ];
for( k = 0; k < nb_subfr >> 1; k++ ) {
/* Check that we are within range of the array */
SKP_assert( target_ptr >= signal_4kHz );
SKP_assert( target_ptr + sf_length_8kHz <= signal_4kHz + frame_length_4kHz );
basis_ptr = target_ptr - min_lag_4kHz;
/* Check that we are within range of the array */
SKP_assert( basis_ptr >= signal_4kHz );
SKP_assert( basis_ptr + sf_length_8kHz <= signal_4kHz + frame_length_4kHz );
normalizer = 0;
cross_corr = 0;
/* Calculate first vector products before loop */
cross_corr = silk_inner_prod_aligned( target_ptr, basis_ptr, sf_length_8kHz );
normalizer = silk_inner_prod_aligned( basis_ptr, basis_ptr, sf_length_8kHz );
normalizer = SKP_ADD_SAT32( normalizer, SKP_SMULBB( sf_length_8kHz, 4000 ) );
temp32 = SKP_DIV32( cross_corr, silk_SQRT_APPROX( normalizer ) + 1 );
C[ k ][ min_lag_4kHz ] = (SKP_int16)SKP_SAT16( temp32 ); /* Q0 */
/* From now on normalizer is computed recursively */
for( d = min_lag_4kHz + 1; d <= max_lag_4kHz; d++ ) {
basis_ptr--;
/* Check that we are within range of the array */
SKP_assert( basis_ptr >= signal_4kHz );
SKP_assert( basis_ptr + sf_length_8kHz <= signal_4kHz + frame_length_4kHz );
cross_corr = silk_inner_prod_aligned( target_ptr, basis_ptr, sf_length_8kHz );
/* Add contribution of new sample and remove contribution from oldest sample */
normalizer +=
SKP_SMULBB( basis_ptr[ 0 ], basis_ptr[ 0 ] ) -
SKP_SMULBB( basis_ptr[ sf_length_8kHz ], basis_ptr[ sf_length_8kHz ] );
temp32 = SKP_DIV32( cross_corr, silk_SQRT_APPROX( normalizer ) + 1 );
C[ k ][ d ] = (SKP_int16)SKP_SAT16( temp32 ); /* Q0 */
}
/* Update target pointer */
target_ptr += sf_length_8kHz;
}
/* Combine two subframes into single correlation measure and apply short-lag bias */
if( nb_subfr == PE_MAX_NB_SUBFR ) {
for( i = max_lag_4kHz; i >= min_lag_4kHz; i-- ) {
sum = (SKP_int32)C[ 0 ][ i ] + (SKP_int32)C[ 1 ][ i ]; /* Q0 */
SKP_assert( SKP_RSHIFT( sum, 1 ) == SKP_SAT16( SKP_RSHIFT( sum, 1 ) ) );
sum = SKP_RSHIFT( sum, 1 ); /* Q-1 */
SKP_assert( SKP_LSHIFT( (SKP_int32)-i, 4 ) == SKP_SAT16( SKP_LSHIFT( (SKP_int32)-i, 4 ) ) );
sum = SKP_SMLAWB( sum, sum, SKP_LSHIFT( -i, 4 ) ); /* Q-1 */
SKP_assert( sum == SKP_SAT16( sum ) );
C[ 0 ][ i ] = (SKP_int16)sum; /* Q-1 */
}
} else {
/* Only short-lag bias */
for( i = max_lag_4kHz; i >= min_lag_4kHz; i-- ) {
sum = (SKP_int32)C[ 0 ][ i ];
sum = SKP_SMLAWB( sum, sum, SKP_LSHIFT( -i, 4 ) ); /* Q-1 */
C[ 0 ][ i ] = (SKP_int16)sum; /* Q-1 */
}
}
/* Sort */
length_d_srch = SKP_ADD_LSHIFT32( 4, complexity, 1 );
SKP_assert( 3 * length_d_srch <= PE_D_SRCH_LENGTH );
silk_insertion_sort_decreasing_int16( &C[ 0 ][ min_lag_4kHz ], d_srch, max_lag_4kHz - min_lag_4kHz + 1, length_d_srch );
/* Escape if correlation is very low already here */
target_ptr = &signal_4kHz[ SKP_SMULBB( sf_length_4kHz, nb_subfr ) ];
energy = silk_inner_prod_aligned( target_ptr, target_ptr, SKP_LSHIFT( sf_length_4kHz, 2 ) );
energy = SKP_ADD_SAT32( energy, 1000 ); /* Q0 */
Cmax = (SKP_int)C[ 0 ][ min_lag_4kHz ]; /* Q-1 */
threshold = SKP_SMULBB( Cmax, Cmax ); /* Q-2 */
/* Compare in Q-2 domain */
if( SKP_RSHIFT( energy, 4 + 2 ) > threshold ) {
SKP_memset( pitch_out, 0, nb_subfr * sizeof( SKP_int ) );
*LTPCorr_Q15 = 0;
*lagIndex = 0;
*contourIndex = 0;
return 1;
}
threshold = SKP_SMULWB( search_thres1_Q16, Cmax );
for( i = 0; i < length_d_srch; i++ ) {
/* Convert to 8 kHz indices for the sorted correlation that exceeds the threshold */
if( C[ 0 ][ min_lag_4kHz + i ] > threshold ) {
d_srch[ i ] = SKP_LSHIFT( d_srch[ i ] + min_lag_4kHz, 1 );
} else {
length_d_srch = i;
break;
}
}
SKP_assert( length_d_srch > 0 );
for( i = min_lag_8kHz - 5; i < max_lag_8kHz + 5; i++ ) {
d_comp[ i ] = 0;
}
for( i = 0; i < length_d_srch; i++ ) {
d_comp[ d_srch[ i ] ] = 1;
}
/* Convolution */
for( i = max_lag_8kHz + 3; i >= min_lag_8kHz; i-- ) {
d_comp[ i ] += d_comp[ i - 1 ] + d_comp[ i - 2 ];
}
length_d_srch = 0;
for( i = min_lag_8kHz; i < max_lag_8kHz + 1; i++ ) {
if( d_comp[ i + 1 ] > 0 ) {
d_srch[ length_d_srch ] = i;
length_d_srch++;
}
}
/* Convolution */
for( i = max_lag_8kHz + 3; i >= min_lag_8kHz; i-- ) {
d_comp[ i ] += d_comp[ i - 1 ] + d_comp[ i - 2 ] + d_comp[ i - 3 ];
}
length_d_comp = 0;
for( i = min_lag_8kHz; i < max_lag_8kHz + 4; i++ ) {
if( d_comp[ i ] > 0 ) {
d_comp[ length_d_comp ] = i - 2;
length_d_comp++;
}
}
/**********************************************************************************
** SECOND STAGE, operating at 8 kHz, on lag sections with high correlation
*************************************************************************************/
/******************************************************************************
** Scale signal down to avoid correlations measures from overflowing
*******************************************************************************/
/* find scaling as max scaling for each subframe */
shift = silk_P_Ana_find_scaling( signal_8kHz, frame_length_8kHz, sf_length_8kHz );
if( shift > 0 ) {
for( i = 0; i < frame_length_8kHz; i++ ) {
signal_8kHz[ i ] = SKP_RSHIFT( signal_8kHz[ i ], shift );
}
}
/*********************************************************************************
* Find energy of each subframe projected onto its history, for a range of delays
*********************************************************************************/
SKP_memset( C, 0, PE_MAX_NB_SUBFR * ( ( PE_MAX_LAG >> 1 ) + 5 ) * sizeof( SKP_int16 ) );
target_ptr = &signal_8kHz[ PE_LTP_MEM_LENGTH_MS * 8 ];
for( k = 0; k < nb_subfr; k++ ) {
/* Check that we are within range of the array */
SKP_assert( target_ptr >= signal_8kHz );
SKP_assert( target_ptr + sf_length_8kHz <= signal_8kHz + frame_length_8kHz );
energy_target = silk_inner_prod_aligned( target_ptr, target_ptr, sf_length_8kHz );
// ToDo: Calculate 1 / energy_target here and save one division inside next for loop
for( j = 0; j < length_d_comp; j++ ) {
d = d_comp[ j ];
basis_ptr = target_ptr - d;
/* Check that we are within range of the array */
SKP_assert( basis_ptr >= signal_8kHz );
SKP_assert( basis_ptr + sf_length_8kHz <= signal_8kHz + frame_length_8kHz );
cross_corr = silk_inner_prod_aligned( target_ptr, basis_ptr, sf_length_8kHz );
energy_basis = silk_inner_prod_aligned( basis_ptr, basis_ptr, sf_length_8kHz );
if( cross_corr > 0 ) {
energy = SKP_max( energy_target, energy_basis ); /* Find max to make sure first division < 1.0 */
lz = silk_CLZ32( cross_corr );
lshift = SKP_LIMIT_32( lz - 1, 0, 15 );
temp32 = SKP_DIV32( SKP_LSHIFT( cross_corr, lshift ), SKP_RSHIFT( energy, 15 - lshift ) + 1 ); /* Q15 */
SKP_assert( temp32 == SKP_SAT16( temp32 ) );
temp32 = SKP_SMULWB( cross_corr, temp32 ); /* Q(-1), cc * ( cc / max(b, t) ) */
temp32 = SKP_ADD_SAT32( temp32, temp32 ); /* Q(0) */
lz = silk_CLZ32( temp32 );
lshift = SKP_LIMIT_32( lz - 1, 0, 15 );
energy = SKP_min( energy_target, energy_basis );
C[ k ][ d ] = SKP_DIV32( SKP_LSHIFT( temp32, lshift ), SKP_RSHIFT( energy, 15 - lshift ) + 1 ); // Q15
} else {
C[ k ][ d ] = 0;
}
}
target_ptr += sf_length_8kHz;
}
/* search over lag range and lags codebook */
/* scale factor for lag codebook, as a function of center lag */
CCmax = SKP_int32_MIN;
CCmax_b = SKP_int32_MIN;
CBimax = 0; /* To avoid returning undefined lag values */
lag = -1; /* To check if lag with strong enough correlation has been found */
if( prevLag > 0 ) {
if( Fs_kHz == 12 ) {
prevLag = SKP_DIV32_16( SKP_LSHIFT( prevLag, 1 ), 3 );
} else if( Fs_kHz == 16 ) {
prevLag = SKP_RSHIFT( prevLag, 1 );
}
prevLag_log2_Q7 = silk_lin2log( (SKP_int32)prevLag );
} else {
prevLag_log2_Q7 = 0;
}
SKP_assert( search_thres2_Q15 == SKP_SAT16( search_thres2_Q15 ) );
/* Setup stage 2 codebook based on number of subframes */
if( nb_subfr == PE_MAX_NB_SUBFR ) {
cbk_size = PE_NB_CBKS_STAGE2_EXT;
Lag_CB_ptr = &silk_CB_lags_stage2[ 0 ][ 0 ];
if( Fs_kHz == 8 && complexity > SILK_PE_MIN_COMPLEX ) {
/* If input is 8 khz use a larger codebook here because it is last stage */
nb_cbk_search = PE_NB_CBKS_STAGE2_EXT;
} else {
nb_cbk_search = PE_NB_CBKS_STAGE2;
}
corr_thres_Q15 = SKP_RSHIFT( SKP_SMULBB( search_thres2_Q15, search_thres2_Q15 ), 13 );
} else {
cbk_size = PE_NB_CBKS_STAGE2_10MS;
Lag_CB_ptr = &silk_CB_lags_stage2_10_ms[ 0 ][ 0 ];
nb_cbk_search = PE_NB_CBKS_STAGE2_10MS;
corr_thres_Q15 = SKP_RSHIFT( SKP_SMULBB( search_thres2_Q15, search_thres2_Q15 ), 14 );
}
for( k = 0; k < length_d_srch; k++ ) {
d = d_srch[ k ];
for( j = 0; j < nb_cbk_search; j++ ) {
CC[ j ] = 0;
for( i = 0; i < nb_subfr; i++ ) {
/* Try all codebooks */
CC[ j ] = CC[ j ] + (SKP_int32)C[ i ][ d + matrix_ptr( Lag_CB_ptr, i, j, cbk_size )];
}
}
/* Find best codebook */
CCmax_new = SKP_int32_MIN;
CBimax_new = 0;
for( i = 0; i < nb_cbk_search; i++ ) {
if( CC[ i ] > CCmax_new ) {
CCmax_new = CC[ i ];
CBimax_new = i;
}
}
/* Bias towards shorter lags */
lag_log2_Q7 = silk_lin2log( (SKP_int32)d ); /* Q7 */
SKP_assert( lag_log2_Q7 == SKP_SAT16( lag_log2_Q7 ) );
SKP_assert( nb_subfr * SILK_FIX_CONST( PE_SHORTLAG_BIAS, 15 ) == SKP_SAT16( nb_subfr * SILK_FIX_CONST( PE_SHORTLAG_BIAS, 15 ) ) );
CCmax_new_b = CCmax_new - SKP_RSHIFT( SKP_SMULBB( nb_subfr * SILK_FIX_CONST( PE_SHORTLAG_BIAS, 15 ), lag_log2_Q7 ), 7 ); /* Q15 */
/* Bias towards previous lag */
SKP_assert( nb_subfr * SILK_FIX_CONST( PE_PREVLAG_BIAS, 15 ) == SKP_SAT16( nb_subfr * SILK_FIX_CONST( PE_PREVLAG_BIAS, 15 ) ) );
if( prevLag > 0 ) {
delta_lag_log2_sqr_Q7 = lag_log2_Q7 - prevLag_log2_Q7;
SKP_assert( delta_lag_log2_sqr_Q7 == SKP_SAT16( delta_lag_log2_sqr_Q7 ) );
delta_lag_log2_sqr_Q7 = SKP_RSHIFT( SKP_SMULBB( delta_lag_log2_sqr_Q7, delta_lag_log2_sqr_Q7 ), 7 );
prev_lag_bias_Q15 = SKP_RSHIFT( SKP_SMULBB( nb_subfr * SILK_FIX_CONST( PE_PREVLAG_BIAS, 15 ), *LTPCorr_Q15 ), 15 ); /* Q15 */
prev_lag_bias_Q15 = SKP_DIV32( SKP_MUL( prev_lag_bias_Q15, delta_lag_log2_sqr_Q7 ), delta_lag_log2_sqr_Q7 + ( 1 << 6 ) );
CCmax_new_b -= prev_lag_bias_Q15; /* Q15 */
}
if ( CCmax_new_b > CCmax_b && /* Find maximum biased correlation */
CCmax_new > corr_thres_Q15 && /* Correlation needs to be high enough to be voiced */
silk_CB_lags_stage2[ 0 ][ CBimax_new ] <= min_lag_8kHz /* Lag must be in range */
) {
CCmax_b = CCmax_new_b;
CCmax = CCmax_new;
lag = d;
CBimax = CBimax_new;
}
}
if( lag == -1 ) {
/* No suitable candidate found */
SKP_memset( pitch_out, 0, nb_subfr * sizeof( SKP_int ) );
*LTPCorr_Q15 = 0;
*lagIndex = 0;
*contourIndex = 0;
return 1;
}
if( Fs_kHz > 8 ) {
/******************************************************************************
** Scale input signal down to avoid correlations measures from overflowing
*******************************************************************************/
/* find scaling as max scaling for each subframe */
shift = silk_P_Ana_find_scaling( signal, frame_length, sf_length );
if( shift > 0 ) {
/* Move signal to scratch mem because the input signal should be unchanged */
/* Reuse the 32 bit scratch mem vector, use a 16 bit pointer from now */
input_signal_ptr = (SKP_int16*)scratch_mem;
for( i = 0; i < frame_length; i++ ) {
input_signal_ptr[ i ] = SKP_RSHIFT( signal[ i ], shift );
}
} else {
input_signal_ptr = (SKP_int16*)signal;
}
/*********************************************************************************/
/* Search in original signal */
CBimax_old = CBimax;
/* Compensate for decimation */
SKP_assert( lag == SKP_SAT16( lag ) );
if( Fs_kHz == 12 ) {
lag = SKP_RSHIFT( SKP_SMULBB( lag, 3 ), 1 );
} else if( Fs_kHz == 16 ) {
lag = SKP_LSHIFT( lag, 1 );
} else {
lag = SKP_SMULBB( lag, 3 );
}
lag = SKP_LIMIT_int( lag, min_lag, max_lag );
start_lag = SKP_max_int( lag - 2, min_lag );
end_lag = SKP_min_int( lag + 2, max_lag );
lag_new = lag; /* to avoid undefined lag */
CBimax = 0; /* to avoid undefined lag */
SKP_assert( SKP_LSHIFT( CCmax, 13 ) >= 0 );
*LTPCorr_Q15 = (SKP_int)silk_SQRT_APPROX( SKP_LSHIFT( CCmax, 13 ) ); /* Output normalized correlation */
CCmax = SKP_int32_MIN;
/* pitch lags according to second stage */
for( k = 0; k < nb_subfr; k++ ) {
pitch_out[ k ] = lag + 2 * silk_CB_lags_stage2[ k ][ CBimax_old ];
}
/* Calculate the correlations and energies needed in stage 3 */
silk_P_Ana_calc_corr_st3( crosscorr_st3, input_signal_ptr, start_lag, sf_length, nb_subfr, complexity );
silk_P_Ana_calc_energy_st3( energies_st3, input_signal_ptr, start_lag, sf_length, nb_subfr, complexity );
lag_counter = 0;
SKP_assert( lag == SKP_SAT16( lag ) );
contour_bias_Q20 = SKP_DIV32_16( SILK_FIX_CONST( PE_FLATCONTOUR_BIAS, 20 ), lag );
/* Setup cbk parameters acording to complexity setting and frame length */
if( nb_subfr == PE_MAX_NB_SUBFR ) {
nb_cbk_search = (SKP_int)silk_nb_cbk_searchs_stage3[ complexity ];
cbk_size = PE_NB_CBKS_STAGE3_MAX;
Lag_CB_ptr = &silk_CB_lags_stage3[ 0 ][ 0 ];
} else {
nb_cbk_search = PE_NB_CBKS_STAGE3_10MS;
cbk_size = PE_NB_CBKS_STAGE3_10MS;
Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ];
}
for( d = start_lag; d <= end_lag; d++ ) {
for( j = 0; j < nb_cbk_search; j++ ) {
cross_corr = 0;
energy = 0;
for( k = 0; k < nb_subfr; k++ ) {
SKP_assert( PE_MAX_NB_SUBFR == 4 );
energy += SKP_RSHIFT( energies_st3[ k ][ j ][ lag_counter ], 2 ); /* use mean, to avoid overflow */
SKP_assert( energy >= 0 );
cross_corr += SKP_RSHIFT( crosscorr_st3[ k ][ j ][ lag_counter ], 2 ); /* use mean, to avoid overflow */
}
if( cross_corr > 0 ) {
/* Divide cross_corr / energy and get result in Q15 */
lz = silk_CLZ32( cross_corr );
/* Divide with result in Q13, cross_corr could be larger than energy */
lshift = SKP_LIMIT_32( lz - 1, 0, 13 );
CCmax_new = SKP_DIV32( SKP_LSHIFT( cross_corr, lshift ), SKP_RSHIFT( energy, 13 - lshift ) + 1 );
CCmax_new = SKP_SAT16( CCmax_new );
CCmax_new = SKP_SMULWB( cross_corr, CCmax_new );
/* Saturate */
if( CCmax_new > SKP_RSHIFT( SKP_int32_MAX, 3 ) ) {
CCmax_new = SKP_int32_MAX;
} else {
CCmax_new = SKP_LSHIFT( CCmax_new, 3 );
}
/* Reduce depending on flatness of contour */
diff = SKP_int16_MAX - SKP_RSHIFT( SKP_MUL( contour_bias_Q20, j ), 5 ); /* Q20 -> Q15 */
SKP_assert( diff == SKP_SAT16( diff ) );
CCmax_new = SKP_LSHIFT( SKP_SMULWB( CCmax_new, diff ), 1 );
} else {
CCmax_new = 0;
}
if( CCmax_new > CCmax &&
( d + silk_CB_lags_stage3[ 0 ][ j ] ) <= max_lag
) {
CCmax = CCmax_new;
lag_new = d;
CBimax = j;
}
}
lag_counter++;
}
for( k = 0; k < nb_subfr; k++ ) {
pitch_out[ k ] = lag_new + matrix_ptr( Lag_CB_ptr, k, CBimax, cbk_size );
}
*lagIndex = (SKP_int16)( lag_new - min_lag);
*contourIndex = (SKP_int8)CBimax;
} else {
/* Save Lags and correlation */
CCmax = SKP_max( CCmax, 0 );
*LTPCorr_Q15 = (SKP_int)silk_SQRT_APPROX( SKP_LSHIFT( CCmax, 13 ) ); /* Output normalized correlation */
for( k = 0; k < nb_subfr; k++ ) {
pitch_out[ k ] = lag + matrix_ptr( Lag_CB_ptr, k, CBimax, cbk_size );
}
*lagIndex = (SKP_int16)( lag - min_lag_8kHz );
*contourIndex = (SKP_int8)CBimax;
}
SKP_assert( *lagIndex >= 0 );
/* return as voiced */
return 0;
}
/*************************************************************************/
/* Calculates the correlations used in stage 3 search. In order to cover */
/* the whole lag codebook for all the searched offset lags (lag +- 2), */
/*************************************************************************/
void silk_P_Ana_calc_corr_st3(
SKP_int32 cross_corr_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ],/* (O) 3 DIM correlation array */
const SKP_int16 signal[], /* I vector to correlate */
SKP_int start_lag, /* I lag offset to search around */
SKP_int sf_length, /* I length of a 5 ms subframe */
SKP_int nb_subfr, /* I number of subframes */
SKP_int complexity /* I Complexity setting */
)
{
const SKP_int16 *target_ptr, *basis_ptr;
SKP_int32 cross_corr;
SKP_int i, j, k, lag_counter, lag_low, lag_high;
SKP_int nb_cbk_search, delta, idx, cbk_size;
SKP_int32 scratch_mem[ SCRATCH_SIZE ];
const SKP_int8 *Lag_range_ptr, *Lag_CB_ptr;
SKP_assert( complexity >= SILK_PE_MIN_COMPLEX );
SKP_assert( complexity <= SILK_PE_MAX_COMPLEX );
if( nb_subfr == PE_MAX_NB_SUBFR ){
Lag_range_ptr = &silk_Lag_range_stage3[ complexity ][ 0 ][ 0 ];
Lag_CB_ptr = &silk_CB_lags_stage3[ 0 ][ 0 ];
nb_cbk_search = silk_nb_cbk_searchs_stage3[ complexity ];
cbk_size = PE_NB_CBKS_STAGE3_MAX;
} else {
SKP_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1);
Lag_range_ptr = &silk_Lag_range_stage3_10_ms[ 0 ][ 0 ];
Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ];
nb_cbk_search = PE_NB_CBKS_STAGE3_10MS;
cbk_size = PE_NB_CBKS_STAGE3_10MS;
}
target_ptr = &signal[ SKP_LSHIFT( sf_length, 2 ) ]; /* Pointer to middle of frame */
for( k = 0; k < nb_subfr; k++ ) {
lag_counter = 0;
/* Calculate the correlations for each subframe */
lag_low = matrix_ptr( Lag_range_ptr, k, 0, 2 );
lag_high = matrix_ptr( Lag_range_ptr, k, 1, 2 );
for( j = lag_low; j <= lag_high; j++ ) {
basis_ptr = target_ptr - ( start_lag + j );
cross_corr = silk_inner_prod_aligned( (SKP_int16*)target_ptr, (SKP_int16*)basis_ptr, sf_length );
SKP_assert( lag_counter < SCRATCH_SIZE );
scratch_mem[ lag_counter ] = cross_corr;
lag_counter++;
}
delta = matrix_ptr( Lag_range_ptr, k, 0, 2 );
for( i = 0; i < nb_cbk_search; i++ ) {
/* Fill out the 3 dim array that stores the correlations for */
/* each code_book vector for each start lag */
idx = matrix_ptr( Lag_CB_ptr, k, i, cbk_size ) - delta;
for( j = 0; j < PE_NB_STAGE3_LAGS; j++ ) {
SKP_assert( idx + j < SCRATCH_SIZE );
SKP_assert( idx + j < lag_counter );
cross_corr_st3[ k ][ i ][ j ] = scratch_mem[ idx + j ];
}
}
target_ptr += sf_length;
}
}
/********************************************************************/
/* Calculate the energies for first two subframes. The energies are */
/* calculated recursively. */
/********************************************************************/
void silk_P_Ana_calc_energy_st3(
SKP_int32 energies_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ],/* (O) 3 DIM energy array */
const SKP_int16 signal[], /* I vector to calc energy in */
SKP_int start_lag, /* I lag offset to search around */
SKP_int sf_length, /* I length of one 5 ms subframe */
SKP_int nb_subfr, /* I number of subframes */
SKP_int complexity /* I Complexity setting */
)
{
const SKP_int16 *target_ptr, *basis_ptr;
SKP_int32 energy;
SKP_int k, i, j, lag_counter;
SKP_int nb_cbk_search, delta, idx, cbk_size, lag_diff;
SKP_int32 scratch_mem[ SCRATCH_SIZE ];
const SKP_int8 *Lag_range_ptr, *Lag_CB_ptr;
SKP_assert( complexity >= SILK_PE_MIN_COMPLEX );
SKP_assert( complexity <= SILK_PE_MAX_COMPLEX );
if( nb_subfr == PE_MAX_NB_SUBFR ){
Lag_range_ptr = &silk_Lag_range_stage3[ complexity ][ 0 ][ 0 ];
Lag_CB_ptr = &silk_CB_lags_stage3[ 0 ][ 0 ];
nb_cbk_search = silk_nb_cbk_searchs_stage3[ complexity ];
cbk_size = PE_NB_CBKS_STAGE3_MAX;
} else {
SKP_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1);
Lag_range_ptr = &silk_Lag_range_stage3_10_ms[ 0 ][ 0 ];
Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ];
nb_cbk_search = PE_NB_CBKS_STAGE3_10MS;
cbk_size = PE_NB_CBKS_STAGE3_10MS;
}
target_ptr = &signal[ SKP_LSHIFT( sf_length, 2 ) ];
for( k = 0; k < nb_subfr; k++ ) {
lag_counter = 0;
/* Calculate the energy for first lag */
basis_ptr = target_ptr - ( start_lag + matrix_ptr( Lag_range_ptr, k, 0, 2 ) );
energy = silk_inner_prod_aligned( basis_ptr, basis_ptr, sf_length );
SKP_assert( energy >= 0 );
scratch_mem[ lag_counter ] = energy;
lag_counter++;
lag_diff = ( matrix_ptr( Lag_range_ptr, k, 1, 2 ) - matrix_ptr( Lag_range_ptr, k, 0, 2 ) + 1 );
for( i = 1; i < lag_diff; i++ ) {
/* remove part outside new window */
energy -= SKP_SMULBB( basis_ptr[ sf_length - i ], basis_ptr[ sf_length - i ] );
SKP_assert( energy >= 0 );
/* add part that comes into window */
energy = SKP_ADD_SAT32( energy, SKP_SMULBB( basis_ptr[ -i ], basis_ptr[ -i ] ) );
SKP_assert( energy >= 0 );
SKP_assert( lag_counter < SCRATCH_SIZE );
scratch_mem[ lag_counter ] = energy;
lag_counter++;
}
delta = matrix_ptr( Lag_range_ptr, k, 0, 2 );
for( i = 0; i < nb_cbk_search; i++ ) {
/* Fill out the 3 dim array that stores the correlations for */
/* each code_book vector for each start lag */
idx = matrix_ptr( Lag_CB_ptr, k, i, cbk_size ) - delta;
for( j = 0; j < PE_NB_STAGE3_LAGS; j++ ) {
SKP_assert( idx + j < SCRATCH_SIZE );
SKP_assert( idx + j < lag_counter );
energies_st3[ k ][ i ][ j ] = scratch_mem[ idx + j ];
SKP_assert( energies_st3[ k ][ i ][ j ] >= 0 );
}
}
target_ptr += sf_length;
}
}
SKP_int32 silk_P_Ana_find_scaling(
const SKP_int16 *signal,
const SKP_int signal_length,
const SKP_int sum_sqr_len
)
{
SKP_int32 nbits, x_max;
x_max = silk_int16_array_maxabs( signal, signal_length );
if( x_max < SKP_int16_MAX ) {
/* Number of bits needed for the sum of the squares */
nbits = 32 - silk_CLZ32( SKP_SMULBB( x_max, x_max ) );
} else {
/* Here we don't know if x_max should have been SKP_int16_MAX + 1, so we expect the worst case */
nbits = 30;
}
nbits += 17 - silk_CLZ16( sum_sqr_len );
/* Without a guarantee of saturation, we need to keep the 31st bit free */
if( nbits < 31 ) {
return 0;
} else {
return( nbits - 30 );
}
}