Merge remote-tracking branch 'origin' into... (030fcff4) · Commits · SA4 / Audio / IVAS BASOP

.gitlab-ci.yml

+1 −2

Original line number	Diff line number	Diff line
		@@ -274,12 +274,11 @@ stages:

		.rules-pytest-to-ref-short:
		rules:
		- if: $PYTEST_MLD_SHORT # Set by scheduled pipeline
		- if: $CI_PIPELINE_SOURCE == 'web' && $MANUAL_PIPELINE_TYPE == "pytest-compare"
		- if: $CI_PIPELINE_SOURCE == 'push' && $CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH
		- if: $CI_PIPELINE_SOURCE == 'push'
		when: never
		- if: $CI_PIPELINE_SOURCE == 'schedule'
		when: never

		.rules-pytest-to-input-short:
		rules:

lib_com/bits_alloc_fx.c

+1 −2

Original line number	Diff line number	Diff line
		@@ -408,7 +408,7 @@ static Word16 BITS_ALLOC_adjust_acelp_fixed_cdk(
		return bitsused;
		}

		/#ifdef IVAS_CODE Below basop operators are missing /

		/-------------------------------------------------------------------
		* fcb_table()
		*
		@@ -3271,4 +3271,3 @@ Word16 set_ACELP_flag_IVAS(
		return 0;
		}
		}
		/#endif IVAS_CODE/

lib_com/est_tilt_fx.c

+13 −22

Original line number	Diff line number	Diff line
		@@ -40,18 +40,12 @@ Word16 est_tilt_fx( /* o : tilt of the code
		const Word32 gain_code, /* i : algebraic code gain Q16 */
		Word16 voice_fac, / o : voicing factor Q15 */
		const Word16 Q_exc /* i : Scaling factor of excitation Q0 */
		#ifdef ADD_LRTD
		,
		const Word16 L_subfr /* i : Sub frame length */
		#endif
		)
		{
		Word16 i, tmp, exp, ener1, exp1, ener2, exp2;
		Word32 L_tmp;
		Word16 tilt_code;
		#ifdef ADD_LRTD
		PMT( "FIX POINT NEED to be adapted for 16 kHz frame length " )
		#endif

		ener1 = extract_h( Dot_product12( exc, exc, L_SUBFR, &exp1 ) );
		exp1 = sub( exp1, add( Q_exc, Q_exc ) );
		L_tmp = L_mult( gain_pit, gain_pit ); /* energy of pitch excitation */
		@@ -117,26 +111,23 @@ Word16 est_tilt_fx( /* o : tilt of the code
		/* RETURN ARGUMENTS : */
		/* _ (Word16) tolt_code : tilt of the code Q15 */
		/=======================================================================/
		Word16 est_tilt_ivas_fx( /* o : tilt of the code Q15 */

		/* o : tilt of the code Q15 */
		Word16 est_tilt_ivas_fx(
		const Word16 exc, / i : adaptive excitation vector Qx */
		const Word16 gain_pit, /* i : adaptive gain Q14 */
		const Word16 code, / i : algebraic excitation vector Q9 */
		const Word32 gain_code, /* i : algebraic code gain Q16 */
		Word16 voice_fac, / o : voicing factor Q15 */
		const Word16 Q_exc /* i : Scaling factor of excitation Q0 */
		#if 1 // def ADD_LRTD
		,
		const Word16 Q_exc, /* i : Scaling factor of excitation Q0 */
		const Word16 L_subfr, /* i : Sub frame length */
		const Word16 flag_tilt /* i : flag for special tilt */
		#endif
		)
		{
		Word16 i, tmp, exp, ener1, exp1, ener2, exp2;
		Word32 L_tmp;
		Word16 tilt_code;
		#ifdef ADD_LRTD
		PMT( "FIX POINT NEED to be adapted for 16 kHz frame length " )
		#endif

		ener1 = extract_h( Dot_product12( exc, exc, L_subfr, &exp1 ) );
		exp1 = sub( exp1, add( Q_exc, Q_exc ) );
		L_tmp = L_mult( gain_pit, gain_pit ); /* energy of pitch excitation */

lib_com/fd_cng_com_fx.c

+0 −5

Original line number	Diff line number	Diff line
		@@ -566,11 +566,6 @@ void minimum_statistics(
		Word16 msPeriodogBuf, / i/o: Buffer of periodograms (energies) */
		Word16 msPeriodogBufPtr, / i/o: Counter */
		HANDLE_FD_CNG_COM hFdCngCom /* i/o: FD_CNG structure containing all buffers and variables */
		#ifdef IVAS_CODE_CNG
		,
		const Word16 enc_dec, /* i : encoder/decoder indicator */
		const Word16 element_mode /* i : IVAS element mode type */
		#endif
		)
		{
		Word16 i, j, k, s, s1, s2, s3;

lib_com/frame_ener.c

deleted100644 → 0

+0 −333

Original line number	Diff line number	Diff line
		/******************************************************************************************************

		(C) 2022-2025 IVAS codec Public Collaboration with portions copyright Dolby International AB, Ericsson AB,
		Fraunhofer-Gesellschaft zur Foerderung der angewandten Forschung e.V., Huawei Technologies Co. LTD.,
		Koninklijke Philips N.V., Nippon Telegraph and Telephone Corporation, Nokia Technologies Oy, Orange,
		Panasonic Holdings Corporation, Qualcomm Technologies, Inc., VoiceAge Corporation, and other
		contributors to this repository. All Rights Reserved.

		This software is protected by copyright law and by international treaties.
		The IVAS codec Public Collaboration consisting of Dolby International AB, Ericsson AB,
		Fraunhofer-Gesellschaft zur Foerderung der angewandten Forschung e.V., Huawei Technologies Co. LTD.,
		Koninklijke Philips N.V., Nippon Telegraph and Telephone Corporation, Nokia Technologies Oy, Orange,
		Panasonic Holdings Corporation, Qualcomm Technologies, Inc., VoiceAge Corporation, and other
		contributors to this repository retain full ownership rights in their respective contributions in
		the software. This notice grants no license of any kind, including but not limited to patent
		license, nor is any license granted by implication, estoppel or otherwise.

		Contributors are required to enter into the IVAS codec Public Collaboration agreement before making
		contributions.

		This software is provided "AS IS", without any express or implied warranties. The software is in the
		development stage. It is intended exclusively for experts who have experience with such software and
		solely for the purpose of inspection. All implied warranties of non-infringement, merchantability
		and fitness for a particular purpose are hereby disclaimed and excluded.

		Any dispute, controversy or claim arising under or in relation to providing this software shall be
		submitted to and settled by the final, binding jurisdiction of the courts of Munich, Germany in
		accordance with the laws of the Federal Republic of Germany excluding its conflict of law rules and
		the United Nations Convention on Contracts on the International Sales of Goods.

		*******************************************************************************************************/

		/*====================================================================================
		EVS Codec 3GPP TS26.443 Nov 04, 2021. Version 12.14.0 / 13.10.0 / 14.6.0 / 15.4.0 / 16.3.0
		====================================================================================*/

		#include <stdint.h>
		#include "options.h"
		#include <math.h>
		#include "cnst.h"
		#include "prot.h"
		#include "wmc_auto.h"
		#include "prot_fx.h"
		/----------------------------------------------------------------------------------
		* fer_energy()
		*
		* Estimation of pitch-synchronous (voiced sounds) or half-frame energy
		----------------------------------------------------------------------------------/

		#ifndef IVAS_FLOAT_FIXED
		void fer_energy(
		const int16_t L_frame, /* i : frame length */
		const int16_t clas, /* i : frame classification */
		const float synth, / i : synthesized speech at Fs = 12k8 Hz */
		const float pitch, /* i : pitch period */
		float enr, / o : pitch-synchronous or half_frame energy */
		const int16_t offset /* i : speech pointer offset (0 or L_frame) */
		)
		{
		int16_t len;
		const float *pt_synth;

		if ( clas == VOICED_CLAS \|\| clas == ONSET \|\| clas == SIN_ONSET ) /* Voiced or Onset current frame */
		{
		len = (int16_t) ( pitch + 0.5f ); /* pitch value */

		pt_synth = synth;
		if ( offset != 0 )
		{
		pt_synth = synth + L_frame - len;
		}

		emaximum( pt_synth, len, enr ); /* pitch synchronous E */
		}
		else
		{
		pt_synth = synth;
		if ( offset != 0 )
		{
		pt_synth = synth + L_frame / 2;
		}

		*enr = dotp( pt_synth, pt_synth, L_frame / 2 );
		*enr /= (float) ( L_frame / 2 );
		}
		return;
		}
		#endif

		void fer_energy_fx(
		const Word16 L_frame, /* i : frame length */
		const Word16 clas, /* i : frame classification */
		const Word32 synth, / i : synthesized speech at Fs = 12k8 Hz Q(q_synth) */
		const Word16 q_synth, /* i : synthesized speech at Fs = 12k8 Hz */
		const Word16 pitch, /* i : pitch period Q0 */
		Word32 enr, / o : pitch-synchronous or half_frame energy Q0 */
		const Word16 offset /* i : speech pointer offset (0 or L_frame) */
		)
		{
		Word16 len, shift, exp;
		const Word32 *pt_synth;
		Word16 enr_tmp, i;
		Word64 W_tmp;

		test();
		test();
		IF( EQ_16( clas, VOICED_CLAS ) \|\| EQ_16( clas, ONSET ) \|\| EQ_16( clas, SIN_ONSET ) ) /* Voiced or Onset current frame */
		{
		len = ( pitch ); /* pitch value */
		move16();

		pt_synth = synth;
		IF( offset != 0 )
		{
		pt_synth = synth + sub( L_frame, len );
		}

		emaximum_32fx( q_synth, pt_synth, len, enr ); /* pitch synchronous E */
		}
		ELSE
		{
		pt_synth = synth;
		IF( offset != 0 )
		{
		pt_synth = synth + shr( L_frame, 1 );
		}

		W_tmp = 0;
		move64();
		FOR( i = 0; i < L_frame / 2; i++ )
		{
		W_tmp = W_add( W_tmp, W_mult0_32_32( pt_synth[i], pt_synth[i] ) ); // Q = q_synth * 2
		}
		shift = W_norm( W_tmp );
		W_tmp = W_shl( W_tmp, shift ); // Q = q_synth * 2 + shift
		enr = W_extract_h( W_tmp ); // Q = q_synth 2 + shift - 32
		move32();

		enr_tmp = BASOP_Util_Divide3216_Scale( enr, shr( L_frame, 1 ) /Q0*/, &exp );
		enr = L_shr( L_deposit_l( enr_tmp ), sub( sub( sub( shift, 32 ), exp ), 1 ) ); /Q0*/
		move32();
		}
		return;
		}


		/----------------------------------------------------------------------------------
		* frame_ener()
		*
		* Estimation of pitch-synchronous (voiced) or mean half-frame (unvoiced) energy
		----------------------------------------------------------------------------------/
		Word16 frame_ener_fx(
		const Word16 L_frame, /* i : length of the frame */
		const Word16 clas, /* i : frame classification */
		const Word16 synth, / i : synthesized speech at Fs = 12k8 Hz Q_new */
		const Word16 pitch, /* i : pitch period Q0 */
		Word32 enr_q, / o : pitch-synchronous or half_frame energy Q0 */
		const Word16 offset, /* i : speech pointer offset (0 or L_FRAME) */
		const Word16 Q_new, /* i : Scaling factor */
		Word16 shift, /* i : Shift need to obtain 12 bits vectors */
		const Word16 enc /* i : Encoder/decoder */
		)
		{
		Word16 len, exp_enrq, exp_tmp, pos;
		Word16 i;
		const Word16 *pt_synth;
		Word32 Ltmp;

		exp_enrq = 0;
		move16();
		test();
		test();
		IF( ( EQ_16( clas, VOICED_CLAS ) ) \|\| ( EQ_16( clas, ONSET ) ) \|\| ( EQ_16( clas, SIN_ONSET ) ) ) /* current frame is voiced */
		{
		/* current frame is voiced */
		len = pitch;
		move16(); /* pitch value at the end of frame */
		pt_synth = synth;
		if ( offset != 0 )
		{
		pt_synth = synth + sub( L_frame, len );
		}
		emaximum_fx( Q_new, pt_synth, len, enr_q );
		IF( enc != 0 )
		{
		exp_enrq = norm_l( *enr_q );
		enr_q = L_shl( enr_q, exp_enrq );
		move32();
		exp_enrq = sub( exp_enrq, 2 );
		}
		}
		ELSE
		{
		/* current frame is unvoiced */
		Word16 L_frame2, exp2, enr_q_tmp;

		L_frame2 = shr( L_frame, 1 );
		pos = 0;
		move16();

		if ( offset != 0 )
		{
		pos = sub( L_frame, L_frame2 );
		}
		Ltmp = L_mult_sat( synth[pos], synth[pos] ); /2 Qnew + 1*/
		FOR( i = 1; i < L_frame2; i++ )
		{
		Ltmp = L_mac_sat( Ltmp, synth[pos + i], synth[pos + i] ); /2 Qnew + 1*/
		}
		test();
		IF( EQ_32( Ltmp, MAX_32 ) \|\| enc != 0 )
		{
		/* scale down when overflow occurs */
		*enr_q = Energy_scale( synth + pos, L_frame2, shift, &exp_enrq );
		move32();
		}
		ELSE
		{
		shift = 0;
		move16();
		/* Normalize acc in Q31 (energy already calculated) */
		pos = norm_l( Ltmp );
		Ltmp = L_shl( Ltmp, pos );
		exp_enrq = sub( 30, pos ); /* exponent = 0..30 */
		*enr_q = Ltmp;
		move32();
		}

		/* enr2 = 1.0f/L_FRAME2 * dot_product(synth, synth, L_FRAME2) */
		exp_enrq = sub( exp_enrq, shl( shift, 1 ) );

		IF( enc != 0 )
		{
		assert( L_frame == 256 \|\| L_frame == 320 );

		exp_tmp = add( shl( Q_new, 1 ), -2 + 7 ); /* L_subfr == L_SUBFR */
		exp_enrq = sub( exp_enrq, exp_tmp );
		exp_enrq = sub( 31, exp_enrq );

		IF( EQ_16( L_frame, 320 ) )
		{
		enr_q = Mult_32_16( enr_q, 26214 ); /x 0.8 to get /160/
		move32();
		i = norm_l( *enr_q );
		enr_q = L_shl( enr_q, i );
		move32();
		exp_enrq = add( i, exp_enrq );
		}
		}
		ELSE
		{
		exp_enrq = sub( exp_enrq, add( Q_new, Q_new ) );
		enr_q_tmp /Q30 exp2+exp_enrq/ = BASOP_Util_Divide3216_Scale( enr_q /Q31/, L_frame2 /Q0*/, &exp2 );
		enr_q = L_shr( L_deposit_l( enr_q_tmp ), sub( 30, add( exp2, exp_enrq ) ) ); /Q0*/
		move32();
		enr_q = L_add( enr_q, 1 );
		move32();
		exp_enrq = 0;
		move16();
		}
		}

		return exp_enrq;
		}

		/------------------------------------------------------------------------
		* frame_energy()
		*
		* Compute pitch-synchronous energy at the frame end
		------------------------------------------------------------------------/
		Word16 frame_energy_fx( /* o : Frame energy in Q8 */
		Word16 L_frame,
		const Word16 pitch, / i : pitch values for each subframe Q6 */
		const Word16 speech, / i : pointer to speech signal for E computation Q_syn*/
		const Word16 lp_speech, /* i : long term active speech energy average Q8 */
		Word16 frame_ener, / o : pitch-synchronous energy at frame end Q8 */
		const Word16 Q_syn /* i : Synthesis scaling */
		)
		{
		Word32 Ltmp;
		const Word16 *pt1;
		Word16 tmp16, exp1, exp2, tmp1, tmp2;
		Word16 len, enern;
		#ifdef BASOP_NOGLOB_DECLARE_LOCAL
		Flag Overflow = 0;
		move32();
		#endif


		/* len = (0.5f * (pitch[2]/64.0 + pitch[3]/64.0) + 0.5f) */
		len = mult_r( add_o( pitch[2], pitch[3], &Overflow ), 256 );

		if ( LT_16( len, L_SUBFR ) )
		{
		len = shl( len, 1 );
		}
		pt1 = speech + sub( L_frame, len );

		/* frame_ener = 10.0f log10(dot_product(pt1, pt1, len) / (float)len) */

		tmp1 = norm_s( len );
		tmp2 = shl( len, tmp1 );
		tmp1 = sub( 15, tmp1 );

		Ltmp = Dot_productSq16HQ( 0, pt1, len, &exp1 );
		exp1 = sub( exp1, shl( Q_syn, 1 ) );
		exp1 = sub( exp1, 1 ); /* compensation of leftshift caused by mac operation in dot_productSq16HQ */
		tmp16 = BASOP_Util_Divide3216_Scale( Ltmp, len, &exp2 );

		exp1 = add( exp1, exp2 );
		exp1 = add( exp1, 1 ); /* compensate result of division Q-1 */


		tmp2 = norm_s( tmp16 );
		Ltmp = L_shl( L_deposit_h( tmp16 ), tmp2 ); /Q16, (exp1-tmp2) = Q31, exp1-tmp2+15/

		Ltmp = BASOP_Util_Log2( Ltmp ); /Q(31-6) = Q25/
		exp1 = sub( 15 + exp1, tmp2 );

		/add ld(2^exp1)=exp1 but check format, first/
		tmp16 = sub( sub( 15, norm_s( exp1 ) ), 5 ); /factor to shift Ltmp and exp1 with (shr) to avoid overflows when adding/
		Ltmp = L_shr_o( Ltmp, tmp16, &Overflow ); /Q25, tmp16/
		exp2 = shr( exp1, tmp16 ); /Q0 , tmp16/
		Ltmp = L_add_o( Ltmp, L_shl( L_deposit_l( exp2 ), 25 ), &Overflow ); /Q25, tmp16, normalized/

		/make 10log10 out of log2*/
		Ltmp = Mpy_32_16_1( Ltmp, LG10 ); /Q25,tmp16 Q13 = Q23, tmp16*/
		frame_ener = extract_h( L_shl_o( Ltmp, add( tmp16, 1 ), &Overflow ) ); /Q8*/
		move16();
		enern = sub_o( frame_ener, lp_speech, &Overflow ); /Q8*/

		return enern;
		}