Fix ISM HR (058b7ab2) · Commits · SA4 / Audio / IVAS BASOP

lib_com/options.h

+5 −2

Original line number	Diff line number	Diff line
		@@ -41,7 +41,7 @@

		/* ################### Start DEBUGGING switches ######################## */

		/#define DEBUGGING/ /* Allows debugging message to be printed out during runtime */
		#define DEBUGGING /* Allows debugging message to be printed out during runtime */
		#ifdef DEBUGGING
		/#define DBG_BITSTREAM_ANALYSIS/ /* Write bitstream with annotations to a text file */
		#define DEBUG_MODE_INFO /* Define to output most important parameters to the subdirectory "res/" */
		@@ -123,9 +123,12 @@
		#ifdef REUSE_EVS_ACELP_INOV
		#define REUSE_EVS_ACELP_TC // Replace TC by EVS TC
		#define REUSE_EVS_ACELP_GEN // Replace VOICE_GEN by EVS VOICE_GEN
		#define REUSE_EVS_GSC // Replace GSC by EVS GSC
		//#define REUSE_EVS_GSC // Replace GSC by EVS GSC
		//#define REUSE_EVS_GSC_part1 // Replace GSC by EVS GSC
		#define USE_NEW_TARGET /* Ittiam target computation allows for -> this seems not to give improvement on stv */



		#endif
		#endif

lib_enc/enc_gen_voic_fx.c

+2 −2

Original line number	Diff line number	Diff line
		@@ -190,7 +190,7 @@ void encod_gen_voic_fx(
		shift_wsp = add( Q_new, shift );
		#if defined USE_NEW_TARGET \|\| defined REUSE_EVS_ACELP_GEN
		test();
		if ( st_fx->element_mode > EVS_MONO )
		if ( st_fx->element_mode > EVS_MONO && LT_32( st_fx->core_brate, MIN_BRATE_AVQ_EXC ) )
		{
		shift_wsp = sub( shift_wsp, 1 );
		}
		@@ -225,7 +225,7 @@ void encod_gen_voic_fx(

		Copy( &res_fx[i_subfr_fx], &exc_fx[i_subfr_fx], L_SUBFR ); /Q_new/
		#ifdef USE_NEW_TARGET
		IF( st_fx->element_mode > EVS_MONO )
		IF( st_fx->element_mode > EVS_MONO && LT_32( st_fx->core_brate, MIN_BRATE_AVQ_EXC ) )
		{
		Word16 q_h1;
		find_targets_ivas_fx( speech_fx, hLPDmem->mem_syn, i_subfr_fx, &hLPDmem->mem_w0, p_Aq_fx,

lib_enc/enc_pit_exc_fx.c

+90 −23

Original line number	Diff line number	Diff line
		@@ -45,6 +45,7 @@
		/* RETURN ARGUMENTS : */
		/* _ None */
		/=======================================================================/

		void enc_pit_exc_fx(
		Encoder_State st_fx, / i/o: State structure */
		const Word16 speech, / i : Input speech Q_new-1*/
		@@ -69,6 +70,9 @@ void enc_pit_exc_fx(
		Word16 xn2[PIT_EXC_L_SUBFR]; /* Target vector for codebook search */
		Word16 h1[PIT_EXC_L_SUBFR + ( M + 1 )]; /* Impulse response vector */
		Word16 y1[PIT_EXC_L_SUBFR]; /* Filtered adaptive excitation */
		#ifdef REUSE_EVS_GSC_part1
		Word16 cn[PIT_EXC_L_SUBFR]; /* Filtered adaptive excitation */
		#endif
		Word16 code[2 * L_SUBFR]; /* Fixed codebook excitation */
		Word16 y2[2 * L_SUBFR]; /* Filtered algebraic excitation */
		Word16 voice_fac; /* Voicing factor */
		@@ -80,7 +84,9 @@ void enc_pit_exc_fx(
		Word16 g_corr[10]; /* ACELP correlation values + gain pitch */
		Word16 clip_gain, i; /* LSF clip gain and LP flag */
		const Word16 p_Aw, p_Aq; /* pointer to LP filter coefficient vector */
		#ifndef REUSE_EVS_GSC_part1
		Word16 cn1[L_SUBFR], cn; / (Used only when L_subfr == L_SUBFR) Target vector in residual domain */
		#endif
		Word16 pt_pitch; / pointer to floating pitch */
		Word16 L_subfr;
		Word16 cum_gpit, gpit_tmp;
		@@ -228,17 +234,30 @@ void enc_pit_exc_fx(
		/------------------------------------------------------------------
		* ACELP subframe loop
		------------------------------------------------------------------/
		#ifndef REUSE_EVS_GSC_part1
		cn = NULL;
		if ( EQ_16( L_subfr, L_SUBFR ) )
		{
		cn = cn1;
		move16();
		}
		#else
		set16_fx( cn, 0, PIT_EXC_L_SUBFR );
		#endif
		p_Aw = Aw;

		p_Aq = Aq;
		pt_pitch = pitch_buf; /* pointer to the pitch buffer */
		shift_wsp = add( Q_new, shift );

		#if defined USE_NEW_TARGET \|\| defined REUSE_EVS_ACELP_GEN
		test();
		if ( st_fx->element_mode > EVS_MONO )
		{
		shift_wsp = sub( shift_wsp, 1 );
		}
		#endif
		//#undef USE_NEW_TARGET
		FOR( i_subfr = 0; i_subfr < st_fx->L_frame; i_subfr += L_subfr )
		{

		@@ -250,11 +269,25 @@ void enc_pit_exc_fx(
		----------------------------------------------------------------/
		Copy( &res[i_subfr], &exc[i_subfr], L_subfr ); /* Q_new */
		/* condition on target (compared to float) has been put outside the loop */
		#ifdef USE_NEW_TARGET
		IF( st_fx->element_mode > EVS_MONO )
		{
		Word16 q_h1;
		find_targets_ivas_fx( speech, hGSCEnc->mem_syn_tmp_fx, i_subfr, &hGSCEnc->mem_w0_tmp_fx, p_Aq,
		res, L_subfr, p_Aw, st_fx->preemph_fac, xn, cn, h1 );

		q_h1 = sub( 14, norm_s( h1[0] ) );
		Copy_Scale_sig( h1, h2, L_subfr, sub( 11, q_h1 ) ); /Q11/
		Scale_sig( h1, L_subfr, add( sub( 14, q_h1 ), shift ) );
		}
		ELSE
		#endif
		{
		find_targets_fx( speech, hLPDmem->mem_syn, i_subfr, &hLPDmem->mem_w0, p_Aq,
		res, L_subfr, p_Aw, st_fx->preemph_fac, xn, cn, h1 );
		Copy_Scale_sig( h1, h2, L_subfr, -2 ); /* Q13 */
		Scale_sig( h1, L_subfr, add( 1, shift ) ); /* set h1[] in Q14 with scaling for convolution */

		}
		/* scaling of xn[] to limit dynamic at 12 bits */
		Scale_sig( xn, L_subfr, shift ); /* Q_new - 1 + shift */

		@@ -320,7 +353,7 @@ void enc_pit_exc_fx(
		push_indice( hBstr, IND_PIT_IDX, pit_idx, 4 );
		}
		}
		else if ( use_fcb == 2 )
		ELSE IF ( EQ_16( use_fcb, 2 ) )
		{
		/-----------------------------------------------------------------
		* Innovation encoding
		@@ -358,7 +391,17 @@ void enc_pit_exc_fx(

		Lgcode = L_shl_sat( gain_code, Q_new ); /* scaled gain_code with Qnew -> Q16*/
		gcode16 = round_fx_sat( Lgcode );
		#ifdef REUSE_EVS_GSC_part1
		IF( EQ_16( use_fcb, 2 ) )
		{
		hLPDmem->tilt_code = est_tilt_ivas_fx( &exc[i_subfr], gain_pit, code, gain_code, &voice_fac, Q_new, L_subfr, 0 ); // Q15 /* Q15 */ //TV this has to be reviewed as saturations are most likely

		move16();
		}
		ELSE IF( use_fcb > 0 )
		#else
		IF( use_fcb != 0 )
		#endif
		{
		hLPDmem->tilt_code = Est_tilt2( &exc[i_subfr], gain_pit, code, Lgcode, &voice_fac, shift ); /* Q15 */
		move16();
		@@ -371,6 +414,30 @@ void enc_pit_exc_fx(
		/-----------------------------------------------------------------
		* Update memory of the weighting filter
		-----------------------------------------------------------------/
		#ifdef USE_NEW_TARGET
		IF( st_fx->element_mode > EVS_MONO )
		{
		IF( use_fcb != 0 )
		{
		Ltmp = L_mult( gcode16, y2[L_subfr - 1] );
		Ltmp = L_shl_o( Ltmp, add( 5, shift ), &Overflow ); // Q_new+14+shift
		Ltmp = L_negate( Ltmp );
		Ltmp = L_mac_o( Ltmp, xn[L_subfr - 1], 16384, &Overflow ); // Q_new-1+15+shift
		Ltmp = L_msu_o( Ltmp, y1[L_subfr - 1], gain_pit, &Overflow ); // Q_new-1+15+shift
		Ltmp = L_shl_o( Ltmp, sub( 1, shift ), &Overflow ); // Q_new+15
		hGSCEnc->mem_w0_tmp_fx = round_fx_o( Ltmp, &Overflow ); /Q_new-1 /
		}
		ELSE
		{
		Ltmp = L_mult( xn[L_subfr - 1], 16384 ); // Q_new-1+15+shift
		Ltmp = L_msu_sat( Ltmp, y1[L_subfr - 1], gain_pit ); // Q_new-1+15+shift
		Ltmp = L_shl_sat( Ltmp, sub( 1, shift ) ); // Q_new+15
		hGSCEnc->mem_w0_tmp_fx = round_fx_sat( Ltmp ); /Q_new-1 /
		}
		}
		ELSE
		#endif
		{
		IF( use_fcb != 0 )
		{
		Ltmp = L_mult( gcode16, y2[L_subfr - 1] ); /* Q10 + Q_new */
		@@ -390,7 +457,7 @@ void enc_pit_exc_fx(
		hLPDmem->mem_w0 = round_fx_sat( Ltmp ); /Q_new-1 /
		move16();
		}

		}
		/-----------------------------------------------------------------
		* Construct adaptive part of the excitation
		* Save the non-enhanced excitation for FEC_exc
		@@ -534,7 +601,7 @@ void enc_pit_exc_fx(
		#endif
		gpit = round_fx( L_mac( L_mult( 3277, gpit ), 29491, cum_gpit ) ); /Q15/
		}

		#ifndef REUSE_EVS_GSC_part1
		void enc_pit_exc_ivas_fx(
		Encoder_State st_fx, / i/o: State structure */
		const Word16 speech, / i : Input speech Q_new-1*/
		@@ -790,7 +857,6 @@ void enc_pit_exc_ivas_fx(
		{
		push_indice( hBstr, IND_LP_FILT_SELECT, lp_select, 1 );
		}

		st_fx->hSpMusClas->lowrate_pitchGain = mac_r( L_mult( 29491 /* 0.9 in Q15 /, st_fx->hSpMusClas->lowrate_pitchGain ), 3277 / 0.1 in Q15 /, gain_pit ); / Q14 */

		gpit_tmp = gain_pit;
		@@ -1038,3 +1104,4 @@ void enc_pit_exc_ivas_fx(
		#endif
		gpit = round_fx( L_mac( L_mult( 3277, gpit ), 29491, cum_gpit ) ); /Q15/
		}
		#endif
		No newline at end of file

lib_enc/enc_tran_fx.c

+2 −2

Original line number	Diff line number	Diff line
		@@ -147,7 +147,7 @@ Word16 encod_tran_fx(
		shift_wsp = add( Q_new, shift );
		#if defined USE_NEW_TARGET \|\| defined REUSE_EVS_ACELP_TC
		test();
		if ( st_fx->element_mode > EVS_MONO )
		if ( st_fx->element_mode > EVS_MONO && LT_32( st_fx->core_brate, MIN_BRATE_AVQ_EXC ) )
		{
		shift_wsp = sub( shift_wsp, 1 );
		}
		@@ -165,7 +165,7 @@ Word16 encod_tran_fx(

		Copy( &res_fx[i_subfr], &exc_fx[i_subfr], L_SUBFR ); /* Q_new */
		#ifdef USE_NEW_TARGET
		IF( st_fx->element_mode > EVS_MONO )
		IF( st_fx->element_mode > EVS_MONO && LT_32( st_fx->core_brate, MIN_BRATE_AVQ_EXC ) )
		{
		Word16 q_h1;
		find_targets_ivas_fx( speech_fx, hLPDmem->mem_syn, i_subfr, &hLPDmem->mem_w0, p_Aq,

lib_enc/gs_enc_fx.c

+5 −2

Original line number	Diff line number	Diff line
		@@ -596,10 +596,13 @@ void encod_audio_ivas_fx(
		Es_pred_enc_fx( &Es_pred, &indice, st_fx->L_frame, res, st_fx->voicing_fx, nb_bits, 0, Q_new );
		push_indice( hBstr, IND_ES_PRED, indice, nb_bits );
		}

		#ifndef REUSE_EVS_GSC_part1
		enc_pit_exc_ivas_fx( st_fx, speech, Aw, Aq, Es_pred, res, synth, exc, &T0_tmp,
		&T0_frac_tmp, pitch_buf, nb_subfr, &hGSCEnc->lt_gpitch_fx, &saved_bit_pos, tdm_Pitch_reuse_flag, tdm_Pri_pitch_buf, Q_new, shift );

		#else
		enc_pit_exc_fx( st_fx, speech, Aw, Aq, Es_pred, res, synth, exc, &T0_tmp,
		&T0_frac_tmp, pitch_buf, nb_subfr, &hGSCEnc->lt_gpitch_fx, &saved_bit_pos, tdm_Pitch_reuse_flag, tdm_Pri_pitch_buf, Q_new, shift );
		#endif
		/---------------------------------------------------------------
		* DCT transform
		---------------------------------------------------------------/