From d307078616b59d33498ac589a1055838c9ae11c0 Mon Sep 17 00:00:00 2001 From: Sandesh Venkatesh Date: Fri, 28 Mar 2025 12:43:47 +0530 Subject: [PATCH] Fix for 3GPP issue 1407: BASOP encoder: Distortion Stereo DTX 24.4 kbps, high MLD Link #1407 --- lib_enc/ivas_cpe_enc_fx.c | 53 ++++++++++++++--------------- lib_enc/ivas_stereo_dft_enc_fx.c | 17 +++------ lib_enc/ivas_stereo_dft_td_itd_fx.c | 36 ++++++++++++++------ 3 files changed, 55 insertions(+), 51 deletions(-) diff --git a/lib_enc/ivas_cpe_enc_fx.c b/lib_enc/ivas_cpe_enc_fx.c index 4e7706baf..be0afc1a4 100644 --- a/lib_enc/ivas_cpe_enc_fx.c +++ b/lib_enc/ivas_cpe_enc_fx.c @@ -127,6 +127,7 @@ ivas_error ivas_cpe_enc_fx( Word16 e_old_wsp[CPE_CHANNELS], q_old_wsp; move16(); // Q_new move16(); // Q_new + Word16 q_com; error = IVAS_ERR_OK; move32(); @@ -649,35 +650,31 @@ ivas_error ivas_cpe_enc_fx( hCPE->hStereoDft->Spd_R_smooth_fx_e = sub( hCPE->hStereoDft->Spd_R_smooth_fx_e, temp ); move16(); - shift = getScaleFactor16( sts[1]->old_input_signal_fx, input_frame ); - Scale_sig( sts[1]->old_input_signal_fx, input_frame, shift ); /* sts[1]->q_old_inp + shift */ - sts[1]->q_old_inp = add( sts[1]->q_old_inp, shift ); - move16(); - shift = getScaleFactor16( sts[1]->input_fx, input_frame ); - Copy_Scale_sig_32_16( sts[1]->input32_fx, sts[1]->input_fx, input_frame, sub( add( sts[1]->q_inp, shift ), sts[1]->q_inp32 ) ); /* sts[1]->q_inp + shift */ - sts[1]->q_inp = add( sts[1]->q_inp, shift ); - move16(); - Scale_sig( sts[1]->input_fx, input_frame, sub( s_min( sts[1]->q_inp, sts[1]->q_old_inp ), sts[1]->q_inp ) ); /* min( sts[1]->q_inp, sts[1]->q_old_inp) */ - Scale_sig( sts[1]->old_input_signal_fx, input_frame, sub( s_min( sts[1]->q_old_inp, sts[1]->q_inp ), sts[1]->q_old_inp ) ); /* min( sts[1]->q_old_inp, sts[1]->q_inp) */ - sts[1]->q_inp = s_min( sts[1]->q_inp, sts[1]->q_old_inp ); - move16(); - sts[1]->q_old_inp = sts[1]->q_inp; - move16(); - shift = getScaleFactor16( sts[0]->old_input_signal_fx, input_frame ); - Scale_sig( sts[0]->old_input_signal_fx, input_frame, shift ); /* sts[0]->q_old_inp + shift */ - sts[0]->q_old_inp = add( sts[0]->q_old_inp, shift ); - move16(); - shift = getScaleFactor16( sts[0]->input_fx, input_frame ); - Copy_Scale_sig_32_16( sts[0]->input32_fx, sts[0]->input_fx, input_frame, sub( add( sts[0]->q_inp, shift ), sts[0]->q_inp32 ) ); /* sts[0]->q_inp + shift */ - sts[0]->q_inp = add( sts[0]->q_inp, shift ); - move16(); - Scale_sig( sts[0]->input_fx, input_frame, sub( s_min( sts[0]->q_inp, sts[0]->q_old_inp ), sts[0]->q_inp ) ); /* min( sts[1]->q_inp, sts[1]->q_old_inp) */ - Scale_sig( sts[0]->old_input_signal_fx, input_frame, sub( s_min( sts[0]->q_old_inp, sts[0]->q_inp ), sts[0]->q_old_inp ) ); /* min( sts[1]->q_old_inp, sts[1]->q_inp) */ - sts[0]->q_inp = s_min( sts[0]->q_inp, sts[0]->q_old_inp ); - move16(); - sts[0]->q_old_inp = sts[0]->q_inp; - move16(); + FOR( n = 0; n < CPE_CHANNELS; n++ ) + { + q_com = sub( add( L_norm_arr( sts[n]->input32_fx, input_frame ), sts[n]->q_inp32 ), 16 ); + q_com = s_min( q_com, add( norm_arr( sts[n]->old_input_signal_fx, input_frame ), sts[n]->q_old_inp ) ); + q_com = s_min( q_com, add( norm_arr( hCPE->input_mem_fx[n], hCPE->hStereoDft->dft_ovl ), hCPE->q_input_mem[n] ) ); + + if ( EQ_16( q_com, Q15 ) ) + { + q_com = 0; + move16(); + } + + Copy_Scale_sig_32_16( sts[n]->input32_fx, sts[n]->input_fx, input_frame, sub( q_com, sts[n]->q_inp32 ) ); + sts[n]->q_inp = q_com; + move16(); + + scale_sig( sts[n]->old_input_signal_fx, input_frame, sub( q_com, sts[n]->q_old_inp ) ); + sts[n]->q_old_inp = q_com; + move16(); + + scale_sig( hCPE->input_mem_fx[n], hCPE->hStereoDft->dft_ovl, sub( q_com, hCPE->q_input_mem[n] ) ); + hCPE->q_input_mem[n] = q_com; + move16(); + } stereo_dft_hybrid_ITD_flag_fx( hCPE->hStereoDft->hConfig, input_Fs, hCPE->hStereoDft->hItd->hybrid_itd_max ); diff --git a/lib_enc/ivas_stereo_dft_enc_fx.c b/lib_enc/ivas_stereo_dft_enc_fx.c index cebbc2619..01318af28 100644 --- a/lib_enc/ivas_stereo_dft_enc_fx.c +++ b/lib_enc/ivas_stereo_dft_enc_fx.c @@ -1064,23 +1064,14 @@ void stereo_dft_enc_analyze_fx( Word16 q_shift, guarded_bits; guarded_bits = find_guarded_bits_fx( NFFT ); - q_shift = sub( getScaleFactor32( pDFT_L, NFFT ), guarded_bits ); - q_shift = s_min( q_shift, sub( getScaleFactor32( pDFT_R, NFFT ), guarded_bits ) ); - FOR( Word16 j = 0; j < NFFT; j++ ) - { - pDFT_L[j] = L_shl( pDFT_L[j], q_shift ); - move32(); - } + q_shift = sub( L_norm_arr( pDFT_L, NFFT ), guarded_bits ); + scale_sig32( pDFT_L, NFFT, q_shift ); DFT_e[0] = sub( DFT_e[0], q_shift ); move16(); - FOR( Word16 j = 0; j < NFFT; j++ ) - { - pDFT_R[j] = L_shl( pDFT_R[j], q_shift ); - move32(); - } - + q_shift = sub( L_norm_arr( pDFT_R, NFFT ), guarded_bits ); + scale_sig32( pDFT_R, NFFT, q_shift ); DFT_e[1] = sub( DFT_e[1], q_shift ); move16(); diff --git a/lib_enc/ivas_stereo_dft_td_itd_fx.c b/lib_enc/ivas_stereo_dft_td_itd_fx.c index df4fe3edd..569a5690a 100644 --- a/lib_enc/ivas_stereo_dft_td_itd_fx.c +++ b/lib_enc/ivas_stereo_dft_td_itd_fx.c @@ -380,15 +380,6 @@ void stereo_td_itd_fx( move16(); set16_fx( shift_input, 0, input_frame ); - FOR( n = 0; n < CPE_CHANNELS; n++ ) - { - - Scale_sig( &input_mem[n][0], dft_ovl, sts[n]->q_inp - q_input_mem[n] ); // Q(sts[n]->q_inp) - // Scale_sig(&sts[n]->old_input_signal_fx, 1965, sts[n]->q_inp - sts[n]->q_old_inp); - // sts[n]->q_old_inp = sts[n]->q_inp; - q_input_mem[n] = sts[n]->q_inp; - move16(); - } IF( EQ_16( sts[0]->element_mode, IVAS_CPE_MDCT ) ) { FOR( n = 0; n < CPE_CHANNELS; n++ ) @@ -540,7 +531,7 @@ void stereo_td_itd_mdct_stereo_fx( const Word16 input_frame /* i : frame length */ ) { - Word16 i; + Word16 i, n, q_com; Word32 bin_nrgL_fx[STEREO_DFT_N_32k_ENC]; Word16 bin_nrgL_e[STEREO_DFT_N_32k_ENC]; Word32 bin_nrgR_fx[STEREO_DFT_N_32k_ENC]; @@ -584,6 +575,31 @@ void stereo_td_itd_mdct_stereo_fx( /*call ITD function*/ stereo_dft_enc_compute_itd_fx( hCPE, DFT_fx[0], DFT_tmp_e[0], DFT_fx[1], DFT_tmp_e[1], STEREO_DFT_OFFSET, input_frame, vad_flag_dtx, vad_hover_flag, bin_nrgL_fx, bin_nrgL_e, bin_nrgR_fx, bin_nrgR_e ); + FOR( n = 0; n < CPE_CHANNELS; n++ ) + { + q_com = add( norm_arr( hCPE->hCoreCoder[n]->input_fx, input_frame ), hCPE->hCoreCoder[n]->q_inp ); + q_com = s_min( q_com, add( norm_arr( hCPE->hCoreCoder[n]->old_input_signal_fx, input_frame ), hCPE->hCoreCoder[n]->q_old_inp ) ); + q_com = s_min( q_com, add( norm_arr( hCPE->input_mem_fx[n], hStereoMdct->hDft_ana->dft_ovl ), hCPE->q_input_mem[n] ) ); + + if ( EQ_16( q_com, Q15 ) ) + { + q_com = 0; + move16(); + } + + scale_sig( hCPE->hCoreCoder[n]->input_fx, input_frame, sub( q_com, hCPE->hCoreCoder[n]->q_inp ) ); + hCPE->hCoreCoder[n]->q_inp = q_com; + move16(); + + scale_sig( hCPE->hCoreCoder[n]->old_input_signal_fx, input_frame, sub( q_com, hCPE->hCoreCoder[n]->q_old_inp ) ); + hCPE->hCoreCoder[n]->q_old_inp = q_com; + move16(); + + scale_sig( hCPE->input_mem_fx[n], hStereoMdct->hDft_ana->dft_ovl, sub( q_com, hCPE->q_input_mem[n] ) ); + hCPE->q_input_mem[n] = q_com; + move16(); + } + /* Time Domain ITD compensation using extrapolation */ stereo_td_itd_fx( hStereoMdct->hItd, NULL, NULL, 1, hStereoMdct->hDft_ana->dft_ovl, hCPE->hCoreCoder, input_frame, hCPE->input_mem_fx, hCPE->q_input_mem ); } -- GitLab