Merge branch 'main' into verification/ism1-nometadata (c431310a) · Commits · SA4 / Audio / IVAS BASOP

lib_com/enhancer_fx.c

+240 −1

Original line number	Diff line number	Diff line
		@@ -520,7 +520,7 @@ void enhancer_ivas_fx(
		{
		/* tmp = 0.150 * (1.0 + voice_fac) */
		/* 0.30=voiced, 0=unvoiced */
		tmp = mac_r( 0x10000000L, voice_fac, 4915 ); /Q15 /
		tmp = mac_r( 0x13333333L, voice_fac, 4915 ); /Q15 /
		}
		ELSE
		{
		@@ -581,6 +581,245 @@ void enhancer_ivas_fx(
		}
		}

		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;
		Word16 gain_code_hi;
		Word16 pit_sharp, tmp16;
		Word16 excp[L_SUBFR], sc;


		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
		*
		* - 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) */
		#ifdef BASOP_NOGLOB
		tmp = msu_r_sat( 0x40000000, voice_fac, 16384 ); /Q15 / /* 1=unvoiced, 0=voiced */
		#else
		tmp = msu_r( 0x40000000, voice_fac, 16384 ); /Q15 / /* 1=unvoiced, 0=voiced */
		#endif
		/* 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 ); /Q16 /
		L_tmp = L_min( L_tmp, gc_threshold ); /Q16 */
		}
		ELSE
		{
		L_tmp = Mult_32_16( norm_gain_code, 27536 ); /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 ) ) )
		{
		#ifdef BASOP_NOGLOB
		pit_sharp = shl_sat( gain_pit, 1 ); /* saturation can occur here Q14 -> Q15 */
		#else
		pit_sharp = shl( gain_pit, 1 ); /* saturation can occur here Q14 -> Q15 */
		#endif
		/* saturation takes care of "if (pit_sharp > 1.0) { pit_sharp=1.0; }" */
		IF( GT_16( pit_sharp, 16384 ) )
		{
		tmp16 = mult( pit_sharp, 8192 );
		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, voice_fac, 4915 ); /Q15 /
		}
		ELSE
		{
		/* tmp = 0.125 * (1.0 + voice_fac) */
		/* 0.25=voiced, 0=unvoiced */
		tmp = mac_r( 0x10000000L, 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 */
		#ifdef BASOP_NOGLOB
		L_tmp = L_shl_sat( L_mult( gain_code_hi, extract_h( L_tmp ) ), sc );
		pt_exc2[0] = msu_r_sat( L_tmp, -32768, pt_exc2[0] );
		move16();
		#else
		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();
		#endif
		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 );
		#ifdef BASOP_NOGLOB
		tmp16 = msu_r_sat( L_tmp, code[i - 1], tmp );
		L_tmp = L_shl_sat( L_mult( gain_code_hi, tmp16 ), sc );
		pt_exc2[i] = msu_r_sat( L_tmp, -32768, pt_exc2[i] );
		move16();
		#else
		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( L_tmp, -32768, pt_exc2[i] );
		#endif
		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 );
		#ifdef BASOP_NOGLOB
		pt_exc2[L_SUBFR - 1] = msu_r_sat( L_tmp, -32768, pt_exc2[L_SUBFR - 1] );
		move16();
		#else
		pt_exc2[L_SUBFR - 1] = msu_r( L_tmp, -32768, pt_exc2[L_SUBFR - 1] );
		move16();
		#endif
		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 ) )
		{
		FOR( i = 0; i < L_SUBFR; i++ )
		{
		/* excp[i] += pt_exc2[i] */
		#ifdef BASOP_NOGLOB
		excp[i] = add_sat( excp[i], pt_exc2[i] );
		#else
		excp[i] = add( excp[i], pt_exc2[i] );
		#endif
		move16();
		}
		agc2_fx( pt_exc2, excp, L_SUBFR );
		Copy( excp, pt_exc2, L_SUBFR );
		}
		}
		}
		}

		/---------------------------------------------------------
		* Enhancement of the excitation signal before synthesis
		---------------------------------------------------------/

lib_com/ivas_prot_fx.h

+14 −0

Original line number	Diff line number	Diff line
		@@ -1708,6 +1708,19 @@ void synchro_synthesis_fx(

		// ivas_dirac_output_synthesis_cov
		void ivas_dirac_dec_output_synthesis_cov_param_mc_collect_slot_fx(
		#ifdef FIX_835_PARAMMC_BUFFER_VALUES
		Word32 RealBuffer_fx, / i : input channel filter bank samples (real part) */
		Word16 RealBuffer_e, /* i : exponent input channel filter bank samples (real part)*/
		Word32 ImagBuffer_fx, / i : input channel filter bank samples (imaginary part */
		Word16 ImagBuffer_e, /* i : exponent input channel filter bank samples (real part)*/
		Word32 cx_fx[PARAM_MC_MAX_TRANSPORT_CHANS * PARAM_MC_MAX_TRANSPORT_CHANS], /* o : accumulated input covariance (real part) */
		Word16 cx_e, / i : exponent for accumulated input covariance (real part) */
		Word32 cx_imag_fx[PARAM_MC_MAX_TRANSPORT_CHANS * PARAM_MC_MAX_TRANSPORT_CHANS], /* o : accumulated input covariance (imaginary part) */
		Word16 cx_imag_e, / i : exponent accumulated input covariance (imag part) */
		PARAM_MC_DEC_HANDLE hParamMC, /* i : handle to Parametric MC state */
		const Word16 param_band, /* i : parameter band */
		const Word16 nchan_in /* i : number of input channels */
		#else
		Word32 RealBuffer_fx, / i : input channel filter bank samples (real part) */
		Word16 RealBuffer_e, /* i : exponent input channel filter bank samples (real part)*/
		Word32 ImagBuffer_fx, / i : input channel filter bank samples (imaginary part */
		@@ -1718,6 +1731,7 @@ void ivas_dirac_dec_output_synthesis_cov_param_mc_collect_slot_fx(
		Word16 cx_imag_e, / i : exponent accumulated input covariance (imag part) */
		PARAM_MC_DEC_HANDLE hParamMC, /* i : handle to Parametric MC state */
		const Word16 nchan_in /* i : number of input channels */
		#endif
		);

		void configureFdCngDec_ivas_fx(

lib_com/ivas_spar_com.c

+1 −1

Original line number	Diff line number	Diff line
		@@ -3531,7 +3531,7 @@ void ivas_get_spar_md_from_dirac_fx(
		/SPAR from DirAC/
		set32_fx( response_avg_fx, 0, MAX_OUTPUT_CHANNELS );

		IF( GE_16( n_ts, 1 ) )
		IF( GT_16( n_ts, 1 ) )
		{
		ivas_dirac_dec_get_response_fx( extract_l( L_shr( azi_dirac_fx[band][i_ts], Q22 ) ), extract_l( L_shr( ele_dirac_fx[band][i_ts], Q22 ) ), response_avg_fx, order, Q30 );
		}

lib_com/options.h

+2 −0

Original line number	Diff line number	Diff line
		@@ -169,6 +169,8 @@


		#define FIX_854_HILBERT_SCALING /* VA: reduce lost of precision due to unnecessary scaling, reduce a lot the 2 kHz tone */
		#define FIX_856_EXTRACT_L /* VA: Fix undesirable wrap-around */
		#define FIX_835_PARAMMC_BUFFER_VALUES /* FhG: issue 835: wide range of buffer values for cx in ParamMC */
		/* ################## End DEVELOPMENT switches ######################### */

		/* clang-format on */

lib_com/prot_fx.h

+33 −0

Original line number	Diff line number	Diff line
		@@ -5193,6 +5193,24 @@ 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 */
		@@ -6416,6 +6434,21 @@ void gain_dec_lbr_fx(
		const Word16 L_subfr /* i : subfr lenght */
		);

		void gain_dec_lbr_ivas_fx(
		Decoder_State st_fx, / i/o: decoder state structure */
		const Word16 coder_type, /* i : coding type */
		const Word16 i_subfr, /* i : subframe index */
		const Word16 code_fx, / i : algebraic excitation Q9 */
		Word16 gain_pit_fx, / o : quantized pitch gain Q14*/
		Word32 gain_code_fx, / o : quantized codebook gain Q16*/
		Word16 gain_inov_fx, / o : gain of the innovation (used for normalization) Q12*/
		Word32 norm_gain_code_fx, / o : norm. gain of the codebook excitation Q16*/
		Word32 gc_mem[], /* i/o: gain_code from previous subframes */
		Word16 gp_mem[] /* i/o: gain_pitch from previous subframes */
		,
		const Word16 L_subfr /* i : subfr lenght */
		);

		void lp_gain_updt_fx(
		const Word16 i_subfr, /* i : subframe number Q0 */
		const Word16 gain_pit, /* i : Decoded gain pitch Q14 */