From 0d9fa201f91dc6977e4cec82201eeb6a5a86c854 Mon Sep 17 00:00:00 2001 From: Sandesh Venkatesh Date: Mon, 31 Mar 2025 15:05:24 +0530 Subject: [PATCH 1/2] Fix for 3GPP issue 1396: Stereo Encoder: BWE artefact at 48 kbps LTV Link #1396 --- lib_enc/ext_sig_ana_fx.c | 2 + lib_enc/ivas_mdct_core_enc_fx.c | 70 ++++++++++++++------------ lib_enc/ivas_stereo_mdct_core_enc_fx.c | 15 +++--- lib_enc/stat_enc.h | 1 + 4 files changed, 46 insertions(+), 42 deletions(-) diff --git a/lib_enc/ext_sig_ana_fx.c b/lib_enc/ext_sig_ana_fx.c index 087d892e3..53c8d080b 100644 --- a/lib_enc/ext_sig_ana_fx.c +++ b/lib_enc/ext_sig_ana_fx.c @@ -1385,6 +1385,8 @@ void core_signal_analysis_high_bitrate_ivas_fx( ProcessIGF_ivas_fx( st, N_MAX + L_MDCT_OVLP_MAX, hTcxEnc->spectrum_fx[frameno], hTcxEnc->spectrum_fx[frameno], &q_spectrum, powerSpec, powerSpec_e, transform_type[frameno] == TCX_20, frameno, 0, vad_hover_flag ); } } + st->hTcxEnc->spectrum_length = L_subframe; + move16(); } diff --git a/lib_enc/ivas_mdct_core_enc_fx.c b/lib_enc/ivas_mdct_core_enc_fx.c index f7da231a3..29ae0518c 100644 --- a/lib_enc/ivas_mdct_core_enc_fx.c +++ b/lib_enc/ivas_mdct_core_enc_fx.c @@ -1251,6 +1251,8 @@ void ivas_mdct_core_whitening_enc_fx( FOR( ch = 0; ch < CPE_CHANNELS; ch++ ) { st = sts[ch]; + st->hTcxEnc->spectrum_length = st->hTcxEnc->L_frameTCX; + move16(); IF( GE_16( add( imult1616( hCPE->cpe_id, CPE_CHANNELS ), ch ), nChannels ) ) { CONTINUE; @@ -1266,6 +1268,8 @@ void ivas_mdct_core_whitening_enc_fx( core_signal_analysis_high_bitrate_ivas_fx( new_samples_fx[ch] + L_INP_MEM, T_op[ch], NULL, NULL, st, tnsSize[ch], tnsBits[ch], param_core[ch], <pBits[ch], windowedSignal_fx[ch], st->L_frame, st->hTcxEnc->L_frameTCX, hCPE->last_element_mode, 0, mdst_spectrum_fx[ch], mdst_spectrum_e[ch], &Q_new, &q_windowedSignal[ch] ); + st->hTcxEnc->spectrum_length = s_max( st->hTcxEnc->spectrum_length, st->hTcxEnc->L_frameTCX ); + move16(); /* BWD in MDCT domain */ IF( NE_16( st->hTcxCfg->tcx_last_overlap_mode, TRANSITION_OVERLAP ) ) { @@ -1373,16 +1377,16 @@ void ivas_mdct_core_whitening_enc_fx( q_min = sub( Q31, q_min ); - q_com = s_min( L_norm_arr( hTcxEnc1->spectrum_fx[n], shr( hTcxEnc1->L_frameTCX, shift ) /*hTcxEnc1->L_frameTCX / nSubframes*/ ), L_norm_arr( hTcxEnc0->spectrum_fx[n], shr( hTcxEnc0->L_frameTCX, shift ) /*hTcxEnc0->L_frameTCX / nSubframes*/ ) ); - q_com = s_min( q_com, L_norm_arr( mdst_spectrum_fx[0][n], shr( hTcxEnc0->L_frameTCX, shift ) /*hTcxEnc0->L_frameTCX / nSubframes*/ ) ); - q_com = s_min( q_com, L_norm_arr( mdst_spectrum_fx[1][n], shr( hTcxEnc1->L_frameTCX, shift ) /*hTcxEnc1->L_frameTCX / nSubframes*/ ) ); + q_com = s_min( L_norm_arr( hTcxEnc1->spectrum_fx[n], shr( hTcxEnc1->spectrum_length, shift ) /*hTcxEnc1->L_frameTCX / nSubframes*/ ), L_norm_arr( hTcxEnc0->spectrum_fx[n], shr( hTcxEnc0->spectrum_length, shift ) /*hTcxEnc0->L_frameTCX / nSubframes*/ ) ); + q_com = s_min( q_com, L_norm_arr( mdst_spectrum_fx[0][n], shr( hTcxEnc0->spectrum_length, shift ) /*hTcxEnc0->L_frameTCX / nSubframes*/ ) ); + q_com = s_min( q_com, L_norm_arr( mdst_spectrum_fx[1][n], shr( hTcxEnc1->spectrum_length, shift ) /*hTcxEnc1->L_frameTCX / nSubframes*/ ) ); q_com = s_min( Q31, add( q_min, q_com ) ); - Scale_sig32( hTcxEnc0->spectrum_fx[n], shr( hTcxEnc0->L_frameTCX, shift ) /*hTcxEnc0->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, hTcxEnc0->spectrum_e[n] ) ) ); // q_com - Scale_sig32( hTcxEnc1->spectrum_fx[n], shr( hTcxEnc1->L_frameTCX, shift ) /*hTcxEnc1->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, hTcxEnc1->spectrum_e[n] ) ) ); // q_com - Scale_sig32( mdst_spectrum_fx[0][n], shr( hTcxEnc0->L_frameTCX, shift ) /*hTcxEnc0->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, mdst_spectrum_e[0][n] ) ) ); // q_com - Scale_sig32( mdst_spectrum_fx[1][n], shr( hTcxEnc1->L_frameTCX, shift ) /*hTcxEnc1->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, mdst_spectrum_e[1][n] ) ) ); // q_com + Scale_sig32( hTcxEnc0->spectrum_fx[n], shr( hTcxEnc0->spectrum_length, shift ) /*hTcxEnc0->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, hTcxEnc0->spectrum_e[n] ) ) ); // q_com + Scale_sig32( hTcxEnc1->spectrum_fx[n], shr( hTcxEnc1->spectrum_length, shift ) /*hTcxEnc1->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, hTcxEnc1->spectrum_e[n] ) ) ); // q_com + Scale_sig32( mdst_spectrum_fx[0][n], shr( hTcxEnc0->spectrum_length, shift ) /*hTcxEnc0->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, mdst_spectrum_e[0][n] ) ) ); // q_com + Scale_sig32( mdst_spectrum_fx[1][n], shr( hTcxEnc1->spectrum_length, shift ) /*hTcxEnc1->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, mdst_spectrum_e[1][n] ) ) ); // q_com const Word16 switchKernel = /* these 4 transform types can be applied: 0 = MDCT-IV, 1 = MDST-II, 2 = MDCT-II, 3 = MDST-IV */ kernel_switch_detect_fx( hTcxEnc0->spectrum_fx[n], hTcxEnc1->spectrum_fx[n], mdst_spectrum_fx[0][n], mdst_spectrum_fx[1][n], q_com, nSampCore / nSubframes, @@ -1470,11 +1474,11 @@ void ivas_mdct_core_whitening_enc_fx( q_min = s_max( mdst_spectrum_e[0][n], hTcxEnc0->spectrum_e[n] ); q_min = sub( Q31, q_min ); - q_com = s_min( L_norm_arr( hTcxEnc0->spectrum_fx[n], shr( hTcxEnc0->L_frameTCX, shift ) /*hTcxEnc0->L_frameTCX / nSubframes*/ ), L_norm_arr( mdst_spectrum_fx[0][n], shr( hTcxEnc0->L_frameTCX, shift ) /*hTcxEnc0->L_frameTCX / nSubframes*/ ) ); + q_com = s_min( L_norm_arr( hTcxEnc0->spectrum_fx[n], shr( hTcxEnc0->spectrum_length, shift ) /*hTcxEnc0->L_frameTCX / nSubframes*/ ), L_norm_arr( mdst_spectrum_fx[0][n], shr( hTcxEnc0->spectrum_length, shift ) /*hTcxEnc0->L_frameTCX / nSubframes*/ ) ); q_com = s_min( Q31, add( q_min, q_com ) ); - Scale_sig32( hTcxEnc0->spectrum_fx[n], shr( hTcxEnc0->L_frameTCX, shift ) /*hTcxEnc0->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, hTcxEnc0->spectrum_e[n] ) ) ); // q_com - Scale_sig32( mdst_spectrum_fx[0][n], shr( hTcxEnc0->L_frameTCX, shift ) /*hTcxEnc0->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, mdst_spectrum_e[0][n] ) ) ); // q_com + Scale_sig32( hTcxEnc0->spectrum_fx[n], shr( hTcxEnc0->spectrum_length, shift ) /*hTcxEnc0->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, hTcxEnc0->spectrum_e[n] ) ) ); // q_com + Scale_sig32( mdst_spectrum_fx[0][n], shr( hTcxEnc0->spectrum_length, shift ) /*hTcxEnc0->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, mdst_spectrum_e[0][n] ) ) ); // q_com speech = NULL; if ( NE_16( n, 1 ) ) @@ -1491,11 +1495,11 @@ void ivas_mdct_core_whitening_enc_fx( q_min = s_max( mdst_spectrum_e[1][n], hTcxEnc1->spectrum_e[n] ); q_min = sub( Q31, q_min ); - q_com = s_min( L_norm_arr( hTcxEnc1->spectrum_fx[n], shr( hTcxEnc1->L_frameTCX, shift ) /*hTcxEnc1->L_frameTCX / nSubframes*/ ), L_norm_arr( mdst_spectrum_fx[1][n], shr( hTcxEnc1->L_frameTCX, shift ) /*hTcxEnc1->L_frameTCX / nSubframes*/ ) ); + q_com = s_min( L_norm_arr( hTcxEnc1->spectrum_fx[n], shr( hTcxEnc1->spectrum_length, shift ) /*hTcxEnc1->L_frameTCX / nSubframes*/ ), L_norm_arr( mdst_spectrum_fx[1][n], shr( hTcxEnc1->spectrum_length, shift ) /*hTcxEnc1->L_frameTCX / nSubframes*/ ) ); q_com = s_min( Q31, add( q_min, q_com ) ); - Scale_sig32( hTcxEnc1->spectrum_fx[n], shr( hTcxEnc1->L_frameTCX, shift ) /*hTcxEnc1->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, hTcxEnc1->spectrum_e[n] ) ) ); // q_com - Scale_sig32( mdst_spectrum_fx[1][n], shr( hTcxEnc1->L_frameTCX, shift ) /*hTcxEnc1->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, mdst_spectrum_e[1][n] ) ) ); // q_com + Scale_sig32( hTcxEnc1->spectrum_fx[n], shr( hTcxEnc1->spectrum_length, shift ) /*hTcxEnc1->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, hTcxEnc1->spectrum_e[n] ) ) ); // q_com + Scale_sig32( mdst_spectrum_fx[1][n], shr( hTcxEnc1->spectrum_length, shift ) /*hTcxEnc1->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, mdst_spectrum_e[1][n] ) ) ); // q_com speech = NULL; if ( NE_16( n, 1 ) ) @@ -1532,7 +1536,7 @@ void ivas_mdct_core_whitening_enc_fx( move16(); /* length = max(nSampCore / (2 * NB_DIV), L_subframeTCX / (2 * NB_DIV), NB_DIV = 2 */ length = shr( s_max( nSampCore, L_subframeTCX ), 2 ); - len_sbfr = shr( hTcxEnc0->L_frameTCX, shift ); + len_sbfr = shr( hTcxEnc0->spectrum_length, shift ); assert( hTcxEnc0->L_frameTCX == hTcxEnc1->L_frameTCX ); exp_max = s_max( hTcxEnc0->spectrum_e[n], hTcxEnc1->spectrum_e[n] ); exp_max = s_max( exp_max, mdst_spectrum_e[0][n] ); @@ -1624,22 +1628,22 @@ void ivas_mdct_core_whitening_enc_fx( { Word16 length; length = shr( s_max( nSampCore, L_subframeTCX ), shift ); // max(/* nSampCore/nSubframes, L_subframeTCX/nSubframes) */ - Word16 offset2 = sub( shr( hTcxEnc0->L_frameTCX, shift ) /* hTcxEnc0->L_frameTCX/nSubframes */, length ); + Word16 offset2 = sub( shr( hTcxEnc0->spectrum_length, shift ) /* hTcxEnc0->L_frameTCX/nSubframes */, length ); exp_max = s_max( hTcxEnc0->spectrum_e[n], hTcxEnc1->spectrum_e[n] ); exp_max = s_max( exp_max, mdst_spectrum_e[0][n] ); exp_max = s_max( exp_max, mdst_spectrum_e[1][n] ); - exp_com = s_min( L_norm_arr( hTcxEnc1->spectrum_fx[n], shr( hTcxEnc1->L_frameTCX, shift ) /* hTcxEnc1->L_frameTCX/nSubframes */ ), L_norm_arr( hTcxEnc0->spectrum_fx[n], shr( hTcxEnc0->L_frameTCX, shift ) /* hTcxEnc0->L_frameTCX/nSubframes */ ) ); - exp_com = s_min( exp_com, L_norm_arr( mdst_spectrum_fx[0][n], shr( hTcxEnc0->L_frameTCX, shift ) /* hTcxEnc0->L_frameTCX/nSubframes */ ) ); - exp_com = s_min( exp_com, L_norm_arr( mdst_spectrum_fx[1][n], shr( hTcxEnc1->L_frameTCX, shift ) /* hTcxEnc1->L_frameTCX/nSubframes */ ) ); + exp_com = s_min( L_norm_arr( hTcxEnc1->spectrum_fx[n], shr( hTcxEnc1->spectrum_length, shift ) /* hTcxEnc1->L_frameTCX/nSubframes */ ), L_norm_arr( hTcxEnc0->spectrum_fx[n], shr( hTcxEnc0->spectrum_length, shift ) /* hTcxEnc0->L_frameTCX/nSubframes */ ) ); + exp_com = s_min( exp_com, L_norm_arr( mdst_spectrum_fx[0][n], shr( hTcxEnc0->spectrum_length, shift ) /* hTcxEnc0->L_frameTCX/nSubframes */ ) ); + exp_com = s_min( exp_com, L_norm_arr( mdst_spectrum_fx[1][n], shr( hTcxEnc1->spectrum_length, shift ) /* hTcxEnc1->L_frameTCX/nSubframes */ ) ); q_com = sub( s_min( Q31, add( sub( Q31, exp_max ), exp_com ) ), 6 ); exp_com = sub( Q31, q_com ); - Scale_sig32( hTcxEnc0->spectrum_fx[n], shr( hTcxEnc0->L_frameTCX, shift ) /* hTcxEnc0->L_frameTCX/nSubframes */, sub( hTcxEnc0->spectrum_e[n], exp_com ) ); // hTcxEnc0->spectrum_e - Scale_sig32( hTcxEnc1->spectrum_fx[n], shr( hTcxEnc1->L_frameTCX, shift ) /* hTcxEnc1->L_frameTCX/nSubframes */, sub( hTcxEnc1->spectrum_e[n], exp_com ) ); // hTcxEnc0->spectrum_e - Scale_sig32( mdst_spectrum_fx[0][n], shr( hTcxEnc0->L_frameTCX, shift ) /* hTcxEnc0->L_frameTCX/nSubframes */, sub( mdst_spectrum_e[0][n], exp_com ) ); // mdst_spectrum_e - Scale_sig32( mdst_spectrum_fx[1][n], shr( hTcxEnc1->L_frameTCX, shift ) /* hTcxEnc1->L_frameTCX/nSubframes */, sub( mdst_spectrum_e[1][n], exp_com ) ); // mdst_spectrum_e + Scale_sig32( hTcxEnc0->spectrum_fx[n], shr( hTcxEnc0->spectrum_length, shift ) /* hTcxEnc0->L_frameTCX/nSubframes */, sub( hTcxEnc0->spectrum_e[n], exp_com ) ); // hTcxEnc0->spectrum_e + Scale_sig32( hTcxEnc1->spectrum_fx[n], shr( hTcxEnc1->spectrum_length, shift ) /* hTcxEnc1->L_frameTCX/nSubframes */, sub( hTcxEnc1->spectrum_e[n], exp_com ) ); // hTcxEnc0->spectrum_e + Scale_sig32( mdst_spectrum_fx[0][n], shr( hTcxEnc0->spectrum_length, shift ) /* hTcxEnc0->L_frameTCX/nSubframes */, sub( mdst_spectrum_e[0][n], exp_com ) ); // mdst_spectrum_e + Scale_sig32( mdst_spectrum_fx[1][n], shr( hTcxEnc1->spectrum_length, shift ) /* hTcxEnc1->L_frameTCX/nSubframes */, sub( mdst_spectrum_e[1][n], exp_com ) ); // mdst_spectrum_e Word16 q_com_orig = q_com; move16(); @@ -1724,12 +1728,12 @@ void ivas_mdct_core_whitening_enc_fx( q_min = s_max( mdst_spectrum_e[ch][n], hTcxEncCh->spectrum_e[n] ); q_min = sub( Q31, q_min ); - q_com = s_min( L_norm_arr( mdst_spectrum_fx[ch][n], shr( hTcxEncCh->L_frameTCX, shift ) /*hTcxEncCh->L_frameTCX / nSubframes*/ ), L_norm_arr( hTcxEncCh->spectrum_fx[n], shr( hTcxEncCh->L_frameTCX, shift ) /*hTcxEncCh->L_frameTCX / nSubframes*/ ) ); + q_com = s_min( L_norm_arr( mdst_spectrum_fx[ch][n], shr( hTcxEncCh->spectrum_length, shift ) /*hTcxEncCh->L_frameTCX / nSubframes*/ ), L_norm_arr( hTcxEncCh->spectrum_fx[n], shr( hTcxEncCh->spectrum_length, shift ) /*hTcxEncCh->L_frameTCX / nSubframes*/ ) ); q_com = s_min( Q31, add( q_min, q_com ) ); - Scale_sig32( hTcxEncCh->spectrum_fx[n], shr( hTcxEncCh->L_frameTCX, shift ) /*hTcxEncCh->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, hTcxEncCh->spectrum_e[n] ) ) ); // q_com - Scale_sig32( mdst_spectrum_fx[ch][n], shr( hTcxEncCh->L_frameTCX, shift ) /*hTcxEncCh->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, mdst_spectrum_e[ch][n] ) ) ); // q_com + Scale_sig32( hTcxEncCh->spectrum_fx[n], shr( hTcxEncCh->spectrum_length, shift ) /*hTcxEncCh->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, hTcxEncCh->spectrum_e[n] ) ) ); // q_com + Scale_sig32( mdst_spectrum_fx[ch][n], shr( hTcxEncCh->spectrum_length, shift ) /*hTcxEncCh->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, mdst_spectrum_e[ch][n] ) ) ); // q_com speech = hTcxEncCh->speech_TCX; if ( n != 0 ) @@ -1769,7 +1773,7 @@ void ivas_mdct_core_whitening_enc_fx( FOR( n = 0; n < nSubframes; n++ ) { q_min = s_max( q_min, hTcxEnc->spectrum_e[n] ); - q_com = s_min( q_com, L_norm_arr( hTcxEnc->spectrum_fx[n], shr( hTcxEnc->L_frameTCX, shift ) /*hTcxEnc->L_frameTCX / nSubframes*/ ) ); + q_com = s_min( q_com, L_norm_arr( hTcxEnc->spectrum_fx[n], shr( hTcxEnc->spectrum_length, shift ) /*hTcxEnc->L_frameTCX / nSubframes*/ ) ); } } q_min = sub( Q31, q_min ); @@ -1791,7 +1795,7 @@ void ivas_mdct_core_whitening_enc_fx( } FOR( n = 0; n < nSubframes; n++ ) { - Scale_sig32( hTcxEnc->spectrum_fx[n], shr( hTcxEnc->L_frameTCX, shift ) /*hTcxEnc->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, hTcxEnc->spectrum_e[n] ) ) ); // q_com + Scale_sig32( hTcxEnc->spectrum_fx[n], shr( hTcxEnc->spectrum_length, shift ) /*hTcxEnc->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, hTcxEnc->spectrum_e[n] ) ) ); // q_com hTcxEnc->spectrum_e[n] = sub( Q31, q_com ); move16(); } @@ -1825,8 +1829,8 @@ void ivas_mdct_core_whitening_enc_fx( q_min = s_max( q_min, hTcxEnc->spectrum_e[n] ); q_min = s_max( q_min, mdst_spectrum_e[ch][n] ); - q_com = s_min( q_com, L_norm_arr( mdst_spectrum_fx[ch][n], shr( hTcxEnc->L_frameTCX, shift ) /*hTcxEnc->L_frameTCX / nSubframes*/ ) ); - q_com = s_min( q_com, L_norm_arr( hTcxEnc->spectrum_fx[n], shr( hTcxEnc->L_frameTCX, shift ) /*hTcxEnc->L_frameTCX / nSubframes*/ ) ); + q_com = s_min( q_com, L_norm_arr( mdst_spectrum_fx[ch][n], shr( hTcxEnc->spectrum_length, shift ) /*hTcxEnc->L_frameTCX / nSubframes*/ ) ); + q_com = s_min( q_com, L_norm_arr( hTcxEnc->spectrum_fx[n], shr( hTcxEnc->spectrum_length, shift ) /*hTcxEnc->L_frameTCX / nSubframes*/ ) ); } } q_min = sub( Q31, q_min ); @@ -1849,8 +1853,8 @@ void ivas_mdct_core_whitening_enc_fx( } FOR( n = 0; n < nSubframes; n++ ) { - Scale_sig32( hTcxEnc->spectrum_fx[n], shr( hTcxEnc->L_frameTCX, shift ) /*hTcxEnc->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, hTcxEnc->spectrum_e[n] ) ) ); // q_com - Scale_sig32( mdst_spectrum_fx[ch][n], shr( hTcxEnc->L_frameTCX, shift ) /*hTcxEnc->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, mdst_spectrum_e[ch][n] ) ) ); // q_com + Scale_sig32( hTcxEnc->spectrum_fx[n], shr( hTcxEnc->spectrum_length, shift ) /*hTcxEnc->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, hTcxEnc->spectrum_e[n] ) ) ); // q_com + Scale_sig32( mdst_spectrum_fx[ch][n], shr( hTcxEnc->spectrum_length, shift ) /*hTcxEnc->L_frameTCX / nSubframes*/, sub( q_com, sub( Q31, mdst_spectrum_e[ch][n] ) ) ); // q_com hTcxEnc->spectrum_e[n] = sub( Q31, q_com ); move16(); mdst_spectrum_e[ch][n] = sub( Q31, q_com ); @@ -2137,7 +2141,7 @@ void ivas_mdct_core_whitening_enc_fx( { q_min = s_max( q_min, hTcxEnc->spectrum_e[n] ); - q_com = s_min( q_com, L_norm_arr( hTcxEnc->spectrum_fx[n], shr( hTcxEnc->L_frameTCX, shift ) /*hTcxEnc->L_frameTCX / nSubframes*/ ) ); + q_com = s_min( q_com, L_norm_arr( hTcxEnc->spectrum_fx[n], shr( hTcxEnc->spectrum_length, shift ) /*hTcxEnc->L_frameTCX / nSubframes*/ ) ); } } q_min = sub( Q31, q_min ); @@ -2161,7 +2165,7 @@ void ivas_mdct_core_whitening_enc_fx( FOR( n = 0; n < nSubframes; n++ ) { - Scale_sig32( hTcxEnc->spectrum_fx[n], shr( hTcxEnc->L_frameTCX, shift ) /*hTcxEnc->L_frameTCX / nSubframes*/, sub( q_com, sub( 31, hTcxEnc->spectrum_e[n] ) ) ); // q_com + Scale_sig32( hTcxEnc->spectrum_fx[n], shr( hTcxEnc->spectrum_length, shift ) /*hTcxEnc->L_frameTCX / nSubframes*/, sub( q_com, sub( 31, hTcxEnc->spectrum_e[n] ) ) ); // q_com hTcxEnc->spectrum_e[n] = sub( 31, q_com ); move16(); } diff --git a/lib_enc/ivas_stereo_mdct_core_enc_fx.c b/lib_enc/ivas_stereo_mdct_core_enc_fx.c index 3becb30d9..b32e8a561 100644 --- a/lib_enc/ivas_stereo_mdct_core_enc_fx.c +++ b/lib_enc/ivas_stereo_mdct_core_enc_fx.c @@ -363,11 +363,11 @@ void stereo_mdct_core_enc_fx( move16(); FOR( ch = 0; ch < CPE_CHANNELS; ch++ ) { - length = sts[ch]->hTcxEnc->L_frameTCX; + length = sts[ch]->hTcxEnc->spectrum_length; move16(); - if ( NE_16( sts[ch]->hTcxEnc->tcxMode, TCX_20 ) ) + IF( NE_16( sts[ch]->hTcxEnc->tcxMode, TCX_20 ) ) { - length = shr( sts[ch]->hTcxEnc->L_frameTCX, 1 ); + length = shr( length, 1 ); } hdrm_min = s_min( hdrm_min, L_norm_arr( sts[ch]->hTcxEnc->spectrum_fx[0], length ) ); @@ -384,15 +384,12 @@ void stereo_mdct_core_enc_fx( FOR( ch = 0; ch < CPE_CHANNELS; ch++ ) { - Word16 n_sb = NB_DIV; + length = sts[ch]->hTcxEnc->spectrum_length; move16(); - if ( EQ_16( sts[ch]->hTcxEnc->tcxMode, TCX_20 ) ) + IF( NE_16( sts[ch]->hTcxEnc->tcxMode, TCX_20 ) ) { - n_sb = 1; - move16(); + length = shr( length, 1 ); } - length = idiv1616( sts[ch]->hTcxEnc->L_frameTCX, n_sb ); /* Q0 */ - FOR( k = 0; k <= ( ( sts[ch]->core == TCX_20_CORE ) ? 1 : NB_DIV ) - 1; k++ ) { Scale_sig32( sts[ch]->hTcxEnc->spectrum_fx[k], length, sub( q_spec, sub( Q31, sts[ch]->hTcxEnc->spectrum_e[k] ) ) ); /* q_spec */ diff --git a/lib_enc/stat_enc.h b/lib_enc/stat_enc.h index 5305a8bdc..3977c4054 100644 --- a/lib_enc/stat_enc.h +++ b/lib_enc/stat_enc.h @@ -1211,6 +1211,7 @@ typedef struct tcx_enc_structure Word32 spectrum_long_fx[N_MAX]; /* MDCT output for a long block. Points to spectrum */ Word16 spectrum_long_e; /* MDCT output for a long block. Points to spectrum */ Word16 q_spectrum_long_fx; + Word16 spectrum_length; } TCX_ENC_DATA, *TCX_ENC_HANDLE; typedef struct TransientDetection -- GitLab From 6f6165506aab955a2072277cd41e566794c17ec0 Mon Sep 17 00:00:00 2001 From: Markus Multrus Date: Wed, 2 Apr 2025 01:15:47 +0200 Subject: [PATCH 2/2] simplify initial patch --- lib_enc/ext_sig_ana_fx.c | 2 -- lib_enc/ivas_mdct_core_enc_fx.c | 7 +++---- lib_enc/stat_enc.h | 2 +- 3 files changed, 4 insertions(+), 7 deletions(-) diff --git a/lib_enc/ext_sig_ana_fx.c b/lib_enc/ext_sig_ana_fx.c index 53c8d080b..087d892e3 100644 --- a/lib_enc/ext_sig_ana_fx.c +++ b/lib_enc/ext_sig_ana_fx.c @@ -1385,8 +1385,6 @@ void core_signal_analysis_high_bitrate_ivas_fx( ProcessIGF_ivas_fx( st, N_MAX + L_MDCT_OVLP_MAX, hTcxEnc->spectrum_fx[frameno], hTcxEnc->spectrum_fx[frameno], &q_spectrum, powerSpec, powerSpec_e, transform_type[frameno] == TCX_20, frameno, 0, vad_hover_flag ); } } - st->hTcxEnc->spectrum_length = L_subframe; - move16(); } diff --git a/lib_enc/ivas_mdct_core_enc_fx.c b/lib_enc/ivas_mdct_core_enc_fx.c index 29ae0518c..e27ec8c11 100644 --- a/lib_enc/ivas_mdct_core_enc_fx.c +++ b/lib_enc/ivas_mdct_core_enc_fx.c @@ -1251,8 +1251,9 @@ void ivas_mdct_core_whitening_enc_fx( FOR( ch = 0; ch < CPE_CHANNELS; ch++ ) { st = sts[ch]; - st->hTcxEnc->spectrum_length = st->hTcxEnc->L_frameTCX; - move16(); + init_tcx_enc_info_fx( st, &L_subframe, &L_subframeTCX, &tcx_subframe_coded_lines ); + st->hTcxEnc->spectrum_length = L_subframeTCX; + IF( GE_16( add( imult1616( hCPE->cpe_id, CPE_CHANNELS ), ch ), nChannels ) ) { CONTINUE; @@ -1268,8 +1269,6 @@ void ivas_mdct_core_whitening_enc_fx( core_signal_analysis_high_bitrate_ivas_fx( new_samples_fx[ch] + L_INP_MEM, T_op[ch], NULL, NULL, st, tnsSize[ch], tnsBits[ch], param_core[ch], <pBits[ch], windowedSignal_fx[ch], st->L_frame, st->hTcxEnc->L_frameTCX, hCPE->last_element_mode, 0, mdst_spectrum_fx[ch], mdst_spectrum_e[ch], &Q_new, &q_windowedSignal[ch] ); - st->hTcxEnc->spectrum_length = s_max( st->hTcxEnc->spectrum_length, st->hTcxEnc->L_frameTCX ); - move16(); /* BWD in MDCT domain */ IF( NE_16( st->hTcxCfg->tcx_last_overlap_mode, TRANSITION_OVERLAP ) ) { diff --git a/lib_enc/stat_enc.h b/lib_enc/stat_enc.h index 3977c4054..c6176709b 100644 --- a/lib_enc/stat_enc.h +++ b/lib_enc/stat_enc.h @@ -1211,7 +1211,7 @@ typedef struct tcx_enc_structure Word32 spectrum_long_fx[N_MAX]; /* MDCT output for a long block. Points to spectrum */ Word16 spectrum_long_e; /* MDCT output for a long block. Points to spectrum */ Word16 q_spectrum_long_fx; - Word16 spectrum_length; + Word16 spectrum_length; /* corresponds to L_frameTCX, used for scaling of MDCT/MDST buffers */ } TCX_ENC_DATA, *TCX_ENC_HANDLE; typedef struct TransientDetection -- GitLab