Merge remote-tracking branch 'origin/main' into 3gpp_issue_1348_fix (84ecc488) · Commits · SA4 / Audio / IVAS BASOP

lib_com/enhancer_fx.c

+8 −213

Original line number	Diff line number	Diff line
		@@ -299,10 +299,11 @@ void enhancer_fx(
		/* RETURN ARGUMENTS : */
		/* _ None */
		/======================================================================================/

		void enhancer_ivas_fx(
		const Word16 codec_mode, /* i : flag indicating Codec Mode */
		const Word32 core_brate, /* i : decoder bitrate */
		const Word16 cbk_index, /* i : */
		const Word16 cbk_index,
		const Word16 Opt_AMR_WB, /* i : flag indicating AMR-WB IO mode */
		const Word16 coder_type, /* i : coder type */
		const Word16 i_subfr, /* i : subframe number */
		@@ -322,9 +323,11 @@ void enhancer_ivas_fx(
		Word16 tmp, fac, *pt_exc2;
		Word16 i;
		Word32 L_tmp;
		Word32 L_tmp1, L_tmp2;
		Word16 gain_code_hi;
		Word16 pit_sharp, tmp16;
		Word16 excp[L_SUBFR], sc;
		Word64 w_temp;


		pit_sharp = gain_pit;
		@@ -400,215 +403,6 @@ void enhancer_ivas_fx(

		phase_dispersion_fx( norm_gain_code, gain_pit, code, i, dm_fx );

		/*------------------------------------------------------------
		* noise enhancer
		*
		* - Enhance excitation on noise. (modify gain of code)
		* If signal is noisy and LPC filter is stable, move gain
		* of code 1.5 dB toward gain of code threshold.
		* This decreases by 3 dB noise energy variation.
		-----------------------------------------------------------/

		/* tmp = 0.5f * (1.0f - voice_fac) */
		tmp = msu_r_sat( 0x40000000 /0.5.Q31/, voice_fac, 16384 /0.5.Q14/ ); /Q15 / /* 1=unvoiced, 0=voiced */
		/* fac = stab_fac * tmp */
		fac = mult( stab_fac, tmp ); /Q15/

		IF( LT_32( norm_gain_code, *gc_threshold ) )
		{
		L_tmp = Madd_32_16( norm_gain_code, norm_gain_code, 6226 /0.19.Q15/ ); /Q16 /
		L_tmp = L_min( L_tmp, gc_threshold ); /Q16 */
		}
		ELSE
		{
		L_tmp = Mult_32_16( norm_gain_code, 27536 /0.84.Q15/ ); /Q16 /
		L_tmp = L_max( L_tmp, gc_threshold ); /Q16 */
		}
		*gc_threshold = L_tmp;
		move32(); /Q16 /

		/* gain_code = (fac * tmp) + (1.0 - fac) * gain_code ==> fac * (tmp - gain_code) + gain_code */
		L_tmp = L_sub( L_tmp, norm_gain_code ); /Q16 /
		norm_gain_code = Madd_32_16( norm_gain_code, L_tmp, fac ); /Q16 /

		/* gain_code = gain_inov - Inverse the normalization /
		L_tmp = Mult_32_16( norm_gain_code, gain_inov ); /Q13/ /* gain_inov in Q12 */

		sc = 6;
		move16();

		gain_code_hi = round_fx( L_shl( L_tmp, add( Q_exc, 3 ) ) ); /* in Q_exc */

		/------------------------------------------------------------
		* pitch enhancer
		*
		* - Enhance excitation on voiced. (HP filtering of code)
		* On voiced signal, filtering of code by a smooth fir HP
		* filter to decrease energy of code at low frequency.
		------------------------------------------------------------/
		test();
		IF( !Opt_AMR_WB && EQ_16( coder_type, UNVOICED ) )
		{
		/* Copy(code, exc2, L_SUBFR) */
		FOR( i = 0; i < L_SUBFR; i++ )
		{
		pt_exc2[i] = round_fx( L_shl( L_mult( gain_code_hi, code[i] ), sc ) ); /Q0 / /* code in Q12 (Q9 for encoder) */
		move16();
		}
		}
		ELSE
		{
		test();
		test();
		IF( Opt_AMR_WB && ( EQ_32( core_brate, ACELP_8k85 ) \|\| EQ_32( core_brate, ACELP_6k60 ) ) )
		{
		pit_sharp = shl_sat( gain_pit, 1 ); /* saturation can occur here Q14 -> Q15 */
		/* saturation takes care of "if (pit_sharp > 1.0) { pit_sharp=1.0; }" */
		IF( GT_16( pit_sharp, 16384 /0.5.Q15/ ) )
		{
		tmp16 = mult( pit_sharp, 8192 /0.25.Q15/ );
		FOR( i = 0; i < L_SUBFR; i++ )
		{
		/* excp[i] = pt_exc2[i] * pit_sharp * 0.25 */
		excp[i] = mult_r( pt_exc2[i], tmp16 );
		move16();
		}
		}
		}

		IF( EQ_16( L_frame, L_FRAME16k ) )
		{
		/* tmp = 0.150 * (1.0 + voice_fac) */
		/* 0.30=voiced, 0=unvoiced */
		tmp = mac_r( 0x13333333L /0.150.Q31/, voice_fac, 4915 /0.150.Q15/ ); /Q15 /
		}
		ELSE
		{
		/* tmp = 0.125 * (1.0 + voice_fac) */
		/* 0.25=voiced, 0=unvoiced */
		tmp = mac_r( 0x10000000L /0.125.Q31/, voice_fac, 4096 ); /Q15 /
		}

		/*-----------------------------------------------------------------
		* Do a simple noncasual "sharpening": effectively an FIR
		* filter with coefs [-tmp 1.0 -tmp] where tmp=0...0.25.
		* This is applied to code and add_fxed to exc2
		-----------------------------------------------------------------/
		/* pt_exc2[0] += code[0] - tmp * code[1] */
		L_tmp = L_deposit_h( code[0] ); /* if Enc :Q9 * Q15 -> Q25 */
		L_tmp = L_msu( L_tmp, code[1], tmp ); /* Q12 * Q15 -> Q28 */
		L_tmp = L_shl( L_mult( gain_code_hi, extract_h( L_tmp ) ), sc );
		pt_exc2[0] = msu_r( L_tmp, -32768, pt_exc2[0] );
		move16(); /* in Q_exc */

		FOR( i = 1; i < L_SUBFR - 1; i++ )
		{
		/* pt_exc2[i] += code[i] - tmp * code[i-1] - tmp * code[i+1] */
		L_tmp = L_msu( -32768, code[i], -32768 );
		L_tmp = L_msu( L_tmp, code[i + 1], tmp );
		tmp16 = msu_r( L_tmp, code[i - 1], tmp );
		L_tmp = L_shl( L_mult( gain_code_hi, tmp16 ), sc );
		pt_exc2[i] = msu_r_sat( L_tmp, -32768, pt_exc2[i] );
		move16(); /* in Q_exc */
		}

		/* pt_exc2[L_SUBFR-1] += code[L_SUBFR-1] - tmp * code[L_SUBFR-2] */
		L_tmp = L_deposit_h( code[L_SUBFR - 1] ); /Q28 /
		L_tmp = L_msu( L_tmp, code[L_SUBFR - 2], tmp ); /Q28 /
		L_tmp = L_shl( L_mult( gain_code_hi, extract_h( L_tmp ) ), sc );
		pt_exc2[L_SUBFR - 1] = msu_r( L_tmp, -32768, pt_exc2[L_SUBFR - 1] );
		move16(); /* in Q_exc */
		test();
		test();
		IF( Opt_AMR_WB && ( EQ_32( core_brate, ACELP_8k85 ) \|\| EQ_32( core_brate, ACELP_6k60 ) ) )
		{
		IF( GT_16( pit_sharp, 16384 /0.5.Q14/ ) )
		{
		FOR( i = 0; i < L_SUBFR; i++ )
		{
		/* excp[i] += pt_exc2[i] */
		excp[i] = add( excp[i], pt_exc2[i] );
		move16();
		}
		agc2_fx( pt_exc2, excp, L_SUBFR );
		Copy( excp, pt_exc2, L_SUBFR );
		}
		}
		}
		}

		void enhancer_ivas_fx2(
		const Word32 core_brate, /* i : decoder bitrate */
		const Word16 Opt_AMR_WB, /* i : flag indicating AMR-WB IO mode */
		const Word16 coder_type, /* i : coder type */
		const Word16 i_subfr, /* i : subframe number */
		const Word16 L_frame, /* i : frame size */
		const Word16 voice_fac, /* i : subframe voicing estimation Q15 */
		const Word16 stab_fac, /* i : LP filter stablility measure Q15 */
		Word32 norm_gain_code, /* i : normalised innovative cb. gain Q16 */
		const Word16 gain_inov, /* i : gain of the unscaled innovation Q12 */
		Word32 gc_threshold, / i/o: gain code threshold Q16 */
		Word16 code, / i/o: innovation Q12 */
		Word16 exc2, / i/o: adapt. excitation/total exc. Q_exc*/
		const Word16 gain_pit, /* i : quantized pitch gain Q14 */
		struct dispMem_fx dm_fx, / i/o: phase dispersion algorithm memory */
		const Word16 Q_exc /* i : Q of the excitation */
		)
		{
		Word16 tmp, fac, *pt_exc2;
		Word16 i;
		Word32 L_tmp;
		Word32 L_tmp1, L_tmp2;
		Word16 gain_code_hi;
		Word16 pit_sharp, tmp16;
		Word16 excp[L_SUBFR], sc;
		Word64 w_temp;


		pit_sharp = gain_pit;
		move16(); /* to remove gcc warning */
		pt_exc2 = exc2 + i_subfr;

		/------------------------------------------------------------
		* Phase dispersion to enhance noise at low bit rate
		------------------------------------------------------------/

		i = 2;
		move16(); /* no dispersion */
		IF( Opt_AMR_WB )
		{
		IF( LE_32( core_brate, ACELP_6k60 ) )
		{
		i = 0;
		move16(); /* high dispersion */
		}
		ELSE IF( LE_32( core_brate, ACELP_8k85 ) )
		{
		i = 1;
		move16(); /* low dispersion */
		}
		}
		ELSE IF( NE_16( coder_type, UNVOICED ) )

		{
		test();
		test();
		test();
		test();
		IF( LE_32( core_brate, ACELP_7k20 ) )
		{
		i = 0;
		move16(); /* high dispersion */
		}
		ELSE IF( ( EQ_16( coder_type, GENERIC ) \|\| EQ_16( coder_type, TRANSITION ) \|\| EQ_16( coder_type, AUDIO ) \|\| coder_type == INACTIVE ) && LE_32( core_brate, ACELP_9k60 ) )
		{
		i = 1;
		move16(); /* low dispersion */
		}
		}

		phase_dispersion_fx( norm_gain_code, gain_pit, code, i, dm_fx );

		/*------------------------------------------------------------
		* noise enhancer
		*
		@@ -656,7 +450,8 @@ void enhancer_ivas_fx2(
		* filter to decrease energy of code at low frequency.
		------------------------------------------------------------/
		test();
		IF( !Opt_AMR_WB && EQ_16( coder_type, UNVOICED ) )
		test();
		IF( !Opt_AMR_WB && EQ_16( codec_mode, MODE1 ) && EQ_16( coder_type, UNVOICED ) )
		{
		/* Copy(code, exc2, L_SUBFR) */
		FOR( i = 0; i < L_SUBFR; i++ )

lib_com/prot_fx.h

+2 −20

Original line number	Diff line number	Diff line
		@@ -5641,24 +5641,6 @@ void enhancer_ivas_fx(
		const Word16 Q_exc /* i : Q of the excitation */
		);

		void enhancer_ivas_fx2(
		const Word32 core_brate, /* i : decoder bitrate */
		const Word16 Opt_AMR_WB, /* i : flag indicating AMR-WB IO mode */
		const Word16 coder_type, /* i : coder type */
		const Word16 i_subfr, /* i : subframe number */
		const Word16 L_frame, /* i : frame size */
		const Word16 voice_fac, /* i : subframe voicing estimation Q15 */
		const Word16 stab_fac, /* i : LP filter stablility measure Q15 */
		Word32 norm_gain_code, /* i : normalised innovative cb. gain Q16 */
		const Word16 gain_inov, /* i : gain of the unscaled innovation Q12 */
		Word32 gc_threshold, / i/o: gain code threshold Q16 */
		Word16 code, / i/o: innovation Q12 */
		Word16 exc2, / i/o: adapt. excitation/total exc. Q_exc*/
		const Word16 gain_pit, /* i : quantized pitch gain Q14 */
		struct dispMem_fx dm_fx, / i/o: phase dispersion algorithm memory */
		const Word16 Q_exc /* i : Q of the excitation */
		);

		Word16 E_UTIL_enhancer(
		Word16 voice_fac, /* i : subframe voicing estimation Q15 */
		Word16 stab_fac, /* i : LP filter stability measure Q15 */

lib_dec/dec_gen_voic_fx.c

+2 −2

Original line number	Diff line number	Diff line
		@@ -344,7 +344,7 @@ ivas_error decod_gen_voic_fx(
		}
		ELSE
		{
		enhancer_ivas_fx2( st_fx->core_brate, 0, st_fx->coder_type, i_subfr_fx, L_frame, voice_fac_fx, st_fx->stab_fac_fx,
		enhancer_ivas_fx( MODE1, st_fx->core_brate, -1, 0, st_fx->coder_type, i_subfr_fx, L_frame, voice_fac_fx, st_fx->stab_fac_fx,
		norm_gain_code_fx, gain_inov_fx, &st_fx->gc_threshold_fx, code_fx, exc2_fx, gain_pit_fx, &( st_fx->dm_fx ), st_fx->Q_exc );
		}
		}

lib_dec/ivas_dirac_dec_fx.c

+21 −21

Original line number	Diff line number	Diff line
		@@ -2217,27 +2217,6 @@ void ivas_dirac_dec_render_sf_fx(
		move16();
		move16();


		FOR( i = 0; i < MAX_OUTPUT_CHANNELS; i++ )
		{
		FOR( Word16 j = 0; j < MAX_PARAM_SPATIAL_SUBFRAMES; j++ )
		{
		set32_fx( Cldfb_RealBuffer_fx[i][j], 0, CLDFB_NO_CHANNELS_MAX );
		set32_fx( Cldfb_ImagBuffer_fx[i][j], 0, CLDFB_NO_CHANNELS_MAX );
		}
		}

		#ifndef OPT_BIN_RENDERER_V2
		FOR( i = 0; i < BINAURAL_CHANNELS; i++ )
		{
		FOR( Word16 j = 0; j < MAX_PARAM_SPATIAL_SUBFRAMES; j++ )
		{
		set32_fx( Cldfb_RealBuffer_Binaural_fx[i][j], 0, CLDFB_NO_CHANNELS_MAX );
		set32_fx( Cldfb_RealBuffer_Binaural_fx[i][j], 0, CLDFB_NO_CHANNELS_MAX );
		}
		}
		#endif /* OPT_BIN_RENDERER_V2 */

		/* local copies of azi, ele, diffuseness */
		Word16 azimuth[CLDFB_NO_CHANNELS_MAX];
		Word16 elevation[CLDFB_NO_CHANNELS_MAX];
		@@ -2288,6 +2267,27 @@ void ivas_dirac_dec_render_sf_fx(
		push_wmops( "ivas_dirac_dec_render" );

		/* Initialize aux buffers */

		FOR( i = 0; i < MAX_OUTPUT_CHANNELS; i++ )
		{
		FOR( Word16 j = 0; j < MAX_PARAM_SPATIAL_SUBFRAMES; j++ )
		{
		set32_fx( Cldfb_RealBuffer_fx[i][j], 0, CLDFB_NO_CHANNELS_MAX );
		set32_fx( Cldfb_ImagBuffer_fx[i][j], 0, CLDFB_NO_CHANNELS_MAX );
		}
		}

		#ifndef OPT_BIN_RENDERER_V2
		FOR( i = 0; i < BINAURAL_CHANNELS; i++ )
		{
		FOR( Word16 j = 0; j < MAX_PARAM_SPATIAL_SUBFRAMES; j++ )
		{
		set32_fx( Cldfb_RealBuffer_Binaural_fx[i][j], 0, CLDFB_NO_CHANNELS_MAX );
		set32_fx( Cldfb_ImagBuffer_Binaural_fx[i][j], 0, CLDFB_NO_CHANNELS_MAX );
		}
		}
		#endif /* OPT_BIN_RENDERER_V2 */

		hDirAC = st_ivas->hDirAC;
		hDirACRend = st_ivas->hDirACRend;
		hSpatParamRendCom = st_ivas->hSpatParamRendCom;

lib_dec/ivas_mct_dec_fx.c

+2 −3

Original line number	Diff line number	Diff line
		@@ -85,7 +85,7 @@ ivas_error ivas_mct_dec_fx(
		set16_fx( x_len[1], 0, NB_DIV );
		Decoder_State **sts;
		Word32 *p_output_orig_fx[2];
		Word32 synth_32_fx[CPE_CHANNELS][L_FRAME_PLUS];
		Word32 synth_fx_32[CPE_CHANNELS][L_FRAME_PLUS];
		Word16 synth_fx[CPE_CHANNELS][L_FRAME_PLUS]; //(Q_synth)
		Word32 ivas_total_brate;
		ivas_error error;
		@@ -168,7 +168,7 @@ ivas_error ivas_mct_dec_fx(
		FOR( n = 0; n < CPE_CHANNELS; n++ )
		{
		p_output_orig_fx[n] = output_fx[n];
		output_fx[n] = synth_32_fx[n];
		output_fx[n] = synth_fx_32[n];
		}
		}

		@@ -390,7 +390,6 @@ ivas_error ivas_mct_dec_fx(
		/----------------------------------------------------------------
		* CoreCoder Post-processing and updates
		----------------------------------------------------------------/
		Word32 synth_fx_32[CPE_CHANNELS][L_FRAME48k];

		FOR( n = 0; n < CPE_CHANNELS; n++ )
		{