harmonized find_uv (1d29f81d) · Commits · SA4 / Audio / IVAS BASOP

lib_com/options.h

+1 −0

Original line number	Diff line number	Diff line
		@@ -109,6 +109,7 @@
		#define FIX_FLOAT_1501_UNIT_VALUE_IN_OMASA /* Nokia: Fix float issue 1501, uninitialized value in ivas_masa_combine_directions for OMASA */
		#define FIX_BASOP_2323_DIRAC_ENC_WRONG_INIT /* Nokia/FhG: basop issue 2323: Use correct init size */
		#define FIX_BASOP_2324_MISSING_SET_TO_ZERO /* Nokia: basop issue 2324: Fix issue by setting the exponent to zero where it should be */
		#define FIX_2349_HARM_FIND_UV /* VA: basop issue 2349 : harmonizing find_uv function */

		/* #################### End BE switches ################################## */

lib_enc/find_uv_fx.c

+48 −13

Original line number	Diff line number	Diff line
		@@ -82,7 +82,7 @@ static Word16 find_ener_decrease_fx( /* o : maximum
		*
		* Decision about coder type
		-------------------------------------------------------------------/

		#ifndef FIX_2349_HARM_FIND_UV
		Word16 find_uv_fx( /* o : coding type */
		Encoder_State st_fx, / i/o: encoder state structure */
		const Word16 T_op_fr, / i : pointer to adjusted fractional pitch (4 val.) Q6*/
		@@ -573,6 +573,9 @@ Word16 find_uv_fx( /* o : coding type
		* Decision about coder type
		-------------------------------------------------------------------/
		Word16 find_uv_ivas_fx( /* o : coding type */
		#else
		Word16 find_uv_fx( /* o : coding type */
		#endif
		Encoder_State st_fx, / i/o: encoder state structure */
		const Word16 T_op_fr, / i : pointer to adjusted fractional pitch (4 val.) Q6*/
		const Word16 voicing_fr, / i : refined correlation for each subframes Q15*/
		@@ -587,6 +590,7 @@ Word16 find_uv_ivas_fx( /* o : coding typ
		const Word16 last_core_orig, /* i : original last core Q0*/
		STEREO_CLASSIF_HANDLE hStereoClassif, /* i/o: stereo classifier structure */
		const Word16 Q_new,
		const Word16 shift,
		const Word16 q_hp_E )
		{
		Word16 coder_type, i;
		@@ -607,7 +611,16 @@ Word16 find_uv_ivas_fx( /* o : coding typ
		SC_VBR_ENC_HANDLE hSC_VBR = st_fx->hSC_VBR;
		Word16 Last_Resort;
		Word16 vadnoise;
		#ifdef FIX_2349_HARM_FIND_UV
		Word16 Q_in, loc_lp_noise;

		Q_in = Q_new;
		move16();
		if( st_fx->element_mode == EVS_MONO )
		{
		Q_in = sub( Q_new, 1 );
		}
		#endif
		IF( hSC_VBR != NULL )
		{
		Last_Resort = hSC_VBR->Last_Resort; /* Q0 */
		@@ -637,7 +650,7 @@ Word16 find_uv_ivas_fx( /* o : coding typ
		pt_enr_ssf = enr_ssf + 2 * NB_SSF;
		FOR( i = 0; i < 2 * ( NB_SSF + 1 ); i++ )
		{
		emaximum_fx( Q_new, pt_speech, SSF, pt_enr_ssf );
		emaximum_fx( Q_in, pt_speech, SSF, pt_enr_ssf );
		pt_speech += ( SSF / 2 );
		pt_enr_ssf++;
		}
		@@ -675,7 +688,7 @@ Word16 find_uv_ivas_fx( /* o : coding typ
		pt_enr_ssf1++;
		}

		IF( hStereoClassif != NULL )
		IF( hStereoClassif != NULL ) /* IVAS only */
		{
		IF( st_fx->idchan == 0 )
		{
		@@ -717,7 +730,12 @@ Word16 find_uv_ivas_fx( /* o : coding typ
		-----------------------------------------------------------------/

		dE3 = sub( Etot, extract_h( hNoiseEst->Etot_last_32fx ) ); /Q8/

		#ifdef FIX_2349_HARM_FIND_UV
		if ( st_fx->element_mode == EVS_MONO ) /* Overwrite DE3 in case of EVS */
		{
		dE3 = sub( Etot, hNoiseEst->Etot_last_fx ); /Q8/
		}
		#endif
		/-----------------------------------------------------------------
		* Energy decrease after spike (dE2)
		-----------------------------------------------------------------/
		@@ -806,7 +824,12 @@ Word16 find_uv_ivas_fx( /* o : coding typ
		}

		E_min_th = L_shl( E_MIN_IVAS_FX_Q31, sub( q_hp_E, Q31 ) );

		#ifdef FIX_2349_HARM_FIND_UV
		if ( st_fx->element_mode == EVS_MONO ) /* Overwrite E_min_th in case of EVS */
		{
		E_min_th = L_shl( E_MIN_FX, Q_new );
		}
		#endif
		test();
		test();
		test();
		@@ -836,16 +859,22 @@ Word16 find_uv_ivas_fx( /* o : coding typ
		{
		ee1_th = 544; /8.5f Q6/
		move16();

		loc_lp_noise = extract_h( st_fx->lp_noise_32fx );
		#ifdef FIX_2349_HARM_FIND_UV
		if ( st_fx->element_mode == EVS_MONO ) /* Overwrite E_min_th in case of EVS */
		{
		loc_lp_noise = st_fx->lp_noise_fx;
		}
		#endif
		/* SC-VBR - determine the threshold on relative energy as a function of lp_noise */
		IF( st_fx->input_bwidth != NB )
		{
		/relE_thres = 0.700f st->lp_noise - 33.5f; (lp_noise in Q8, constant Q8<<16) */
		L_tmp = L_mac( -562036736 /* 33.5f in Q24 /, 22938 / 0.7 in Q15 */, extract_h( st_fx->lp_noise_32fx ) ); // Q24
		L_tmp = L_mac( -562036736 /* 33.5f in Q24 /, 22938 / 0.7 in Q15 */, loc_lp_noise ); // Q24
		IF( Last_Resort == 0 )
		{
		/relE_thres = 0.650f st->lp_noise - 33.5f; (lp_noise in Q8, constant Q8<<16)*/
		L_tmp = L_mac( -562036736 /* 33.5f in Q24 /, 21299 / 0.650f in Q15 */, extract_h( st_fx->lp_noise_32fx ) ); // Q24
		L_tmp = L_mac( -562036736 /* 33.5f in Q24 /, 21299 / 0.650f in Q15 */, loc_lp_noise ); // Q24
		}
		relE_thres = round_fx( L_tmp );
		}
		@@ -853,7 +882,7 @@ Word16 find_uv_ivas_fx( /* o : coding typ
		{

		/relE_thres = 0.60f st->lp_noise - 28.2f; (lp_noise in Q8, constant Q8<<16)*/
		L_tmp = L_mac( -473117491 /* 28.2f in Q24 /, 19661 / 0.6f in Q15 */, extract_h( st_fx->lp_noise_32fx ) ); // Q24
		L_tmp = L_mac( -473117491 /* 28.2f in Q24 /, 19661 / 0.6f in Q15 */, loc_lp_noise ); // Q24
		relE_thres = round_fx( L_tmp );
		}
		relE_thres = s_max( relE_thres, -6400 /* -25.0f in Q8 / ); / Q8 */
		@@ -889,7 +918,7 @@ Word16 find_uv_ivas_fx( /* o : coding typ
		test();
		test();
		test();
		IF( ( ( LT_16( add_sat( mean_voi3, corr_shift ), add( 22282 /* 0.68 in Q15 /, mean_voi3_offset ) ) ) && / normalized correlation low */
		if( ( ( LT_16( add_sat( mean_voi3, corr_shift ), add( 22282 /* 0.68 in Q15 /, mean_voi3_offset ) ) ) && / normalized correlation low */
		( LT_16( add_sat( st_fx->voicing_fx[2], corr_shift ), 25887 /* 0.79 in Q15 / ) ) && / normalized correlation low on look-ahead - onset detection */
		( LT_32( ee[0], 640 /* 10.0f in Q6 / ) ) && ( GT_32( hp_E[0], E_min_th ) ) && / energy concentrated in high frequencies provided that some energy is present in HF... */
		( LT_32( ee[1], ee1_th ) ) && ( GT_32( hp_E[1], E_min_th ) ) && /* ... biased towards look-ahead to detect onsets */
		@@ -917,7 +946,9 @@ Word16 find_uv_ivas_fx( /* o : coding typ
		test();
		test();
		if ( ( ( LT_16( add_sat( mean_voi3, corr_shift ), add( 22774 /* 0.695f in Q15 /, mean_voi3_offset ) ) ) && / normalized correlation low */
		/( LT_16( add_o( st_fx->voicing_fx[2], corr_shift, &Overflow ), 25887 ) ) && / /* normalized correlation low on look-ahead - onset detection */
		#ifdef FIX_2349_HARM_FIND_UV
		( LT_16( add_sat( st_fx->voicing_fx[2], corr_shift ), 25887 ) && st_fx->element_mode == EVS_MONO ) && /* normalized correlation low on look-ahead - onset detection */
		#endif
		( LT_32( ee[0], 397 /* 6.2f in Q6 / ) ) && ( GT_32( hp_E[0], E_min_th ) ) && / energy concentrated in high frequencies provided that some energy is present in HF... */
		( LT_32( ee[1], 397 /* 6.2f in Q16 / ) ) && ( GT_32( hp_E[1], E_min_th ) ) && / ... biased towards look-ahead to detect onsets */
		( tmp_offset_flag == 0 ) && /* Take care of voiced offsets */
		@@ -1037,9 +1068,13 @@ Word16 find_uv_ivas_fx( /* o : coding typ
		move16();
		/* Current frame cannot be compressed to pack the partial redundancy;*/

		IF( NE_16( st_fx->rf_mode, st_fx->Opt_RF_ON ) )
		IF( NE_16( st_fx->rf_mode, st_fx->Opt_RF_ON ) ) /* can happen only for EVS-BE modes*/
		{
		#ifdef FIX_2349_HARM_FIND_UV
		core_coder_mode_switch_fx( st_fx, st_fx->last_total_brate, 0, shift );
		#else
		core_coder_mode_switch_ivas_fx( st_fx, st_fx->last_total_brate, 0 );
		#endif
		}
		}
		}

lib_enc/ivas_core_pre_proc_front_fx.c

+5 −1

Original line number	Diff line number	Diff line
		@@ -1292,9 +1292,13 @@ ivas_error pre_proc_front_ivas_fx(

		find_tilt_ivas_fx( fr_bands_fx, fr_bands_fx_q, bckr_temp, scale, ee_fx, st->pitch, st->voicing_fx, lf_E_fx, q_lf_E_fx,
		corr_shift_fx, st->input_bwidth, st->max_band, hp_E_fx, MODE1, &( st->bckr_tilt_lt ), st->Opt_SC_VBR );

		#ifdef FIX_2349_HARM_FIND_UV
		st->coder_type = find_uv_fx( st, pitch_fr_fx, voicing_fr_fx, inp_12k8_fx, ee_fx, &dE1X_fx, corr_shift_fx, *relE_fx, extract_h( Etot_fx ), hp_E_fx,
		&flag_spitch, last_core_orig, hStereoClassif, Q_new /q_inp_12k8*/, shift, fr_bands_fx_q ); // Q0
		#else
		st->coder_type = find_uv_ivas_fx( st, pitch_fr_fx, voicing_fr_fx, inp_12k8_fx, ee_fx, &dE1X_fx, corr_shift_fx, *relE_fx, extract_h( Etot_fx ), hp_E_fx,
		&flag_spitch, last_core_orig, hStereoClassif, Q_new /q_inp_12k8*/, fr_bands_fx_q ); // Q0
		#endif
		Copy_Scale_sig_16_32_no_sat( st->lgBin_E_fx, st->Bin_E_fx, L_FFT / 2, sub( st->q_Bin_E, Q7 ) );

		/-----------------------------------------------------------------

lib_enc/pre_proc_fx.c

+5 −1

Original line number	Diff line number	Diff line
		@@ -520,9 +520,13 @@ void pre_proc_fx(
		find_tilt_fx( fr_bands, hNoiseEst->bckr_fx, ee, st->pitch, st->voicing_fx, lf_E, corr_shift, st->input_bwidth,
		st->max_band, hp_E, st->codec_mode, *Q_new, &( st->bckr_tilt_lt ), st->Opt_SC_VBR );

		#ifdef FIX_2349_HARM_FIND_UV
		st->coder_type = find_uv_fx( st, pitch_fr, voicing_fr, inp_12k8, ee, NULL, corr_shift, relE, *Etot, hp_E,
		&flag_spitch, last_core_orig, NULL, Q_new, shift, NULL ); // Q0
		#else
		st->coder_type = find_uv_fx( st, pitch_fr, voicing_fr, inp_12k8, ee,
		corr_shift, relE, Etot, hp_E, Q_new, &flag_spitch, *shift, last_core_orig );

		#endif
		/-----------------------------------------------------------------
		* channel aware mode configuration *
		-----------------------------------------------------------------/