Merge branch 'swb_tbe_bwe_dec_fxd' into 'main' (077e7b2c) · Commits · SA4 / Audio / IVAS BASOP

lib_com/ivas_prot_fx.h

+44 −0

Original line number	Diff line number	Diff line
		@@ -1529,4 +1529,48 @@ void synchonize_channels_mdct_sid_fx(
		Decoder_State sts[CPE_CHANNELS], / i/o: decoder state structure */
		const Word16 n /* i : channel number */
		);

		void ivas_interpolate_3_over_1_allpass_fx(
		const Word16 input_fx, / i : input signal / / Q_input */
		const Word16 len, /* i : number of input samples */
		Word16 out_fx, / o : output signal / / Q_input */
		Word16 mem_fx / i/o: memory / / Q_input */
		);

		void ivas_wb_tbe_dec_fx(
		Decoder_State st_fx, / i/o: decoder state structure */
		const Word16 coder_type, /* i : coding type */
		Word32 bwe_exc_extended, / i : bandwidth extended exciatation 2Q_exc/
		const Word16 Q_exc,
		const Word16 voice_factors[], /* i : voicing factors */
		Word16 synth, / o : WB synthesis/final synthesis */
		Word16 *Q_synth );

		void ivas_GenShapedWBExcitation_fx(
		Word16 excSHB, / o : synthesized shaped shb exctiation Q_bwe_exc*/
		const Word16 lpc_shb, / i : lpc coefficients Q12*/
		Word16 exc4kWhtnd, / o : whitened synthesized shb excitation Q_bwe_exc*/
		Word32 mem_csfilt, / i/o : memory Q_bwe_exc+16*/
		Word16 mem_genSHBexc_filt_down1, / i/o : memory Q_bwe_exc*/
		Word16 mem_genSHBexc_filt_down2, / i/o : memory Q_bwe_exc*/
		Word16 mem_genSHBexc_filt_down3, / i/o : memory Q_bwe_exc*/
		Word16 state_lpc_syn, / i/o : memory Q_bwe_exc*/
		const Word16 coder_type, /* i : coding type */
		const Word16 bwe_exc_extended, / i : bwidth extended exciatation Q_bwe_exc*/
		const Word16 Q_bwe_exc,
		Word16 bwe_seed[], /* i/o : random number generator seed */
		const Word16 voice_factors[], /* i : voicing factor Q15*/
		const Word16 uv_flag, /* i : unvoiced flag */
		const Word16 igf_flag );

		Word16 ivas_wb_bwe_dec_fx(
		Decoder_State st_fx, / i/o: decoder state structure */
		const Word16 output[], /* i : suntehsis @ internal Fs */
		Word16 synth_fx, / i/o: ACELP core synthesis/final synthesis */
		Word16 hb_synth_fx, / o : SHB synthesis/final synthesis */
		const Word16 use_cldfb_for_dft, /* i : flag to use of CLDFB for DFT Stereo */
		const Word16 output_frame, /* i : frame length */
		Word16 voice_factors_fx, / i : voicing factors */
		const Word16 pitch_buf_fx[], /* i : pitch buffer */
		Word16 *Qpost );
		#endif

lib_com/modif_fs_fx.c

+94 −0

Original line number	Diff line number	Diff line
		@@ -1111,6 +1111,100 @@ void interpolate_3_over_2_allpass_fx(
		* Interpolate 3/1 using allpass iir polyphase filter. Delay 4 samples @48k
		-------------------------------------------------------------------/

		#ifdef IVAS_FLOAT_FIXED
		void ivas_interpolate_3_over_1_allpass_fx(
		const Word16 input_fx, / i : input signal / / Q_input */
		const Word16 len, /* i : number of input samples */
		Word16 out_fx, / o : output signal / / Q_input */
		Word16 mem_fx / i/o: memory / / Q_input */
		)
		{
		/* mem of current frame would be stored in Qinput, so the next call to this function shoulf have Q_mem parameter set to prev_Q_input */
		Word16 i;
		Word16 Vu[2], Vm[2], Vl[2]; /* Outputs of three cascaded allpass stages (upper, middle, and lower) */
		Word16 *out1;
		Word16 mem_temp;
		const Word16 *filt_coeff_fx = allpass_poles_3_ov_2;

		out1 = &out_fx[0];

		FOR( i = 0; i < len; i++ )
		{
		/* Upper branch */
		/Vu[0] = mem[0] + filt_coeff[0] ( input_fx_temp[i] - mem[1] );
		Vu[1] = mem[1] + filt_coeff[1] * ( Vu[0] - mem[2] );
		mem[3] = mem[2] + filt_coeff[2] * ( Vu[1] - mem[3] );*/

		Vu[0] = add_sat( mem_fx[0], mult_r( filt_coeff_fx[0], sub_sat( input_fx[i], mem_fx[1] ) ) );
		move16(); /* all Vu's in : Q_current*/
		Vu[1] = add_sat( mem_fx[1], mult_r( filt_coeff_fx[1], sub_sat( Vu[0], mem_fx[2] ) ) );
		move16();
		mem_fx[3] = add_sat( mem_fx[2], mult_r( filt_coeff_fx[2], sub_sat( Vu[1], mem_fx[3] ) ) );
		move16();


		mem_fx[1] = Vu[0];
		move16();
		mem_fx[2] = Vu[1];
		move16();
		*out1++ = mem_fx[3];
		move16();

		/* Middle branch */
		/* Vm[0] = mem[0] + filt_coeff[3] * (input[i]-mem[4]);
		Vm[1] = mem[4] + filt_coeff[4] * (Vm[0]-mem[5]);
		mem[6] = mem[5] + filt_coeff[5] * (Vm[1]-mem[6]); */
		Vm[0] = add_sat( mem_fx[0], mult_r( filt_coeff_fx[3], sub_sat( input_fx[i], mem_fx[4] ) ) );
		move16();
		Vm[1] = add_sat( mem_fx[4], mult_r( filt_coeff_fx[4], sub_sat( Vm[0], mem_fx[5] ) ) );
		move16();
		mem_fx[6] = add_sat( mem_fx[5], mult_r( filt_coeff_fx[5], sub_sat( Vm[1], mem_fx[6] ) ) );
		move16();

		mem_fx[4] = Vm[0];
		move16();
		mem_fx[5] = Vm[1];
		move16();
		*out1++ = mem_fx[6];
		move16();

		/* Lower branch */
		/* Vl[0] = mem[0] + filt_coeff[6] * (input[i]-mem[7]);
		Vl[1] = mem[7] + filt_coeff[7] * (Vl[0]-mem[8]);
		mem[9] = mem[8] + filt_coeff[8] * (Vl[1]-mem[9]); */
		Vl[0] = add_sat( mem_fx[0], mult_r( filt_coeff_fx[6], sub_sat( input_fx[i], mem_fx[7] ) ) );
		move16();
		Vl[1] = add_sat( mem_fx[7], mult_r( filt_coeff_fx[7], sub_sat( Vl[0], mem_fx[8] ) ) );
		move16();
		mem_fx[9] = add_sat( mem_fx[8], mult_r( filt_coeff_fx[8], sub_sat( Vl[1], mem_fx[9] ) ) );
		move16();

		mem_fx[0] = input_fx[i];
		move16();
		mem_fx[7] = Vl[0];
		move16();
		mem_fx[8] = Vl[1];
		move16();
		*out1++ = mem_fx[9];
		move16();
		}
		/LPF/
		FOR( i = 0; i < len * 3; i++ )
		{
		mem_temp = out_fx[i];
		move16();
		out_fx[i] = sub_sat( mult_r( 18768 /0.57276865021499168f Q15/, add_sat( mem_fx[12], mem_fx[11] ) ), mult_r( 2425 /0.074004974641176793f Q15/, add_sat( mem_temp, mem_fx[10] ) ) );
		mem_fx[10] = mem_fx[11];
		move16();
		mem_fx[11] = mem_fx[12];
		move16();
		mem_fx[12] = mem_temp;
		move16();
		}
		return;
		}
		#endif

		void interpolate_3_over_1_allpass_fx(
		const Word16 input_fx, / i : input signal / / Q_input */
		const Word16 len, /* i : number of input samples */

lib_com/swb_tbe_com_fx.c

+289 −0

Original line number	Diff line number	Diff line
		@@ -915,6 +915,295 @@ void flip_spectrum_and_decimby4_fx(
		/--------------------------------------------------------------------------/
		/* CALLED FROM : */
		/==========================================================================/
		#ifdef IVAS_FLOAT_FIXED
		void ivas_GenShapedWBExcitation_fx(
		Word16 excSHB, / o : synthesized shaped shb exctiation Q_bwe_exc*/
		const Word16 lpc_shb, / i : lpc coefficients Q12*/
		Word16 exc4kWhtnd, / o : whitened synthesized shb excitation Q_bwe_exc*/
		Word32 mem_csfilt, / i/o : memory Q_bwe_exc+16*/
		Word16 mem_genSHBexc_filt_down1, / i/o : memory Q_bwe_exc*/
		Word16 mem_genSHBexc_filt_down2, / i/o : memory Q_bwe_exc*/
		Word16 mem_genSHBexc_filt_down3, / i/o : memory Q_bwe_exc*/
		Word16 state_lpc_syn, / i/o : memory Q_bwe_exc*/
		const Word16 coder_type, /* i : coding type */
		const Word16 bwe_exc_extended, / i : bwidth extended exciatation Q_bwe_exc*/
		const Word16 Q_bwe_exc,
		Word16 bwe_seed[], /* i/o : random number generator seed */
		const Word16 voice_factors[], /* i : voicing factor Q15*/
		const Word16 uv_flag, /* i : unvoiced flag */
		const Word16 igf_flag )
		{
		Word16 i, j, k;
		Word16 wht_fil_mem[LPC_WHTN_ORDER_WB];
		Word16 lpc_whtn[LPC_WHTN_ORDER_WB + 1];
		Word16 R_h[LPC_WHTN_ORDER_WB + 2], R_l[LPC_WHTN_ORDER_WB + 2];
		Word16 Q_R;
		Word16 excTmp[L_FRAME16k];
		Word16 excTmp2[L_FRAME16k / 4];
		Word16 excTmp2_frac[L_FRAME16k / 4];
		Word16 exc4k[L_FRAME16k / 4];
		Word16 exc4k_frac[L_FRAME16k / 4];
		Word32 exc4k_32[L_FRAME16k / 4];
		Word32 pow1, pow22;
		Word16 scale;
		Word32 excNoisyEnv[L_FRAME16k / 4];
		Word16 csfilt_num2[1] = { 1638 }; /* Q15*/
		Word16 neg_csfilt_den2[2] = { -32768, 31457 }; /* Q15 */
		Word32 L_tmp, Ltemp1, Ltemp2;
		Word16 temp1, temp2, exp;
		Word32 Lmax;
		Word16 max_val, n1, n2, sc;
		Word32 LepsP[LPC_WHTN_ORDER_WB + 1];
		Word16 tmp_vfac;
		Word16 avg_voice_fac;
		#ifdef BASOP_NOGLOB_DECLARE_LOCAL
		Flag Overflow = 0;
		#endif

		/0.25fsum_f(voice_factors, NB_SUBFR)*/
		L_tmp = L_mult( voice_factors[0], 8192 );
		FOR( i = 1; i < NB_SUBFR; i++ )
		{
		L_tmp = L_mac( L_tmp, voice_factors[i], 8192 );
		}
		avg_voice_fac = round_fx( L_tmp );

		test();
		test();
		test();
		test();
		IF( igf_flag != 0 && ( EQ_16( coder_type, VOICED ) \|\| GT_16( avg_voice_fac, 11469 ) ) ) /Q15 -> 0.35f/
		{
		csfilt_num2[0] = 6554;
		move16(); /Q15 -> 0.2f/
		neg_csfilt_den2[1] = 26214;
		move16(); /Q15 -> 0.8f/
		}
		ELSE IF( igf_flag != 0 && ( EQ_16( coder_type, UNVOICED ) \|\| LT_16( avg_voice_fac, 6654 ) ) ) /Q15 -> 0.2f/
		{
		csfilt_num2[0] = 328;
		move16(); /Q15 -> 0.01f/
		neg_csfilt_den2[1] = 32440;
		move16(); /Q15 -> 0.99f/
		}
		set16_fx( wht_fil_mem, 0, LPC_WHTN_ORDER_WB );
		Decimate_allpass_steep_fx( bwe_exc_extended, mem_genSHBexc_filt_down1, L_FRAME32k, excTmp );
		flip_spectrum_and_decimby4_fx( excTmp, exc4k, L_FRAME16k, mem_genSHBexc_filt_down2, mem_genSHBexc_filt_down3, 0 );

		IF( uv_flag )
		{
		create_random_vector_fx( exc4kWhtnd, L_FRAME16k / 4, bwe_seed );
		IF( LT_16( Q_bwe_exc, 5 ) )
		{

		FOR( i = 0; i < L_FRAME16k / 4; i++ )
		{
		exc4kWhtnd[i] = shl_r( exc4kWhtnd[i], sub( Q_bwe_exc, 5 ) ); /Q(Q_bwe_exc)/Q5(if Q_bwe_exc > 5) /
		}
		}
		}
		ELSE
		{
		autocorr_fx( exc4k, LPC_WHTN_ORDER_WB + 1, R_h, R_l, &Q_R,
		L_FRAME16k / 4, win_flatten_4k_fx, 0, 1 );

		/* Ensure R[0] isn't zero when entering Levinson Durbin */
		R_l[0] = s_max( R_l[0], 1 );
		move16();
		FOR( i = 1; i <= LPC_WHTN_ORDER_WB; i++ )
		{
		L_tmp = Mpy_32( R_h[i], R_l[i], wac_h[i - 1], wac_l[i - 1] );
		L_Extract( L_tmp, &R_h[i], &R_l[i] );
		}

		E_LPC_lev_dur( R_h, R_l, lpc_whtn, LepsP, LPC_WHTN_ORDER_WB, NULL );

		Copy_Scale_sig( lpc_whtn, lpc_whtn, LPC_WHTN_ORDER_WB + 1, sub( norm_s( lpc_whtn[0] ), 2 ) );

		fir_fx( exc4k, lpc_whtn, exc4kWhtnd, wht_fil_mem, L_FRAME16k / 4,
		LPC_WHTN_ORDER_WB, 0, 3 );

		/* Ensure pow1 is greater than zero when computing normalization */
		max_val = 0;
		FOR( i = 0; i < L_FRAME16k / 4; i++ )
		{
		excTmp2[i] = abs_s( exc4kWhtnd[i] );
		move16(); /* Q_bwe_exc */
		max_val = s_max( max_val, excTmp2[i] );
		move16();
		}

		IF( max_val == 0 )
		{
		pow1 = 1;
		move16();
		n1 = 0;
		move16();
		}
		ELSE
		{
		n1 = norm_s( max_val );
		FOR( i = 0; i < L_FRAME16k / 4; i++ )
		{
		#ifdef BASOP_NOGLOB
		excTmp2_frac[i] = shl_o( excTmp2[i], n1, &Overflow ); // Q_bwe_exc + n1
		#else
		excTmp2_frac[i] = shl( excTmp2[i], n1 );
		#endif
		move16(); /* Q14 */
		}
		n1 = sub( sub( 14, n1 ), Q_bwe_exc );
		pow1 = 1;
		FOR( i = 0; i < L_FRAME16k / 4; i++ )
		{
		#ifdef BASOP_NOGLOB
		L_tmp = L_mult_o( excTmp2_frac[i], excTmp2_frac[i], &Overflow ); /* Q29 */
		pow1 = L_add_o( pow1, L_shr( L_tmp, 10 ), &Overflow ); /* Q22 */
		#else
		L_tmp = L_mult( excTmp2_frac[i], excTmp2_frac[i] ); /* Q29 */
		pow1 = L_add( pow1, L_shr( L_tmp, 7 ) ); /* Q22 */
		#endif
		}
		}

		FOR( i = 0; i < L_FRAME16k / 4; i++ )
		{
		#ifdef BASOP_NOGLOB
		excNoisyEnv[i] = L_add_o( *mem_csfilt, L_mult_o( csfilt_num2[0], excTmp2[i], &Overflow ), &Overflow );
		#else
		excNoisyEnv[i] = L_add( *mem_csfilt, L_mult( csfilt_num2[0], excTmp2[i] ) );
		#endif
		move32(); /* Q_bwe_exc+16 */
		*mem_csfilt = Mult_32_16( excNoisyEnv[i], neg_csfilt_den2[1] );
		move32(); /* Q_bwe_exc+16 */
		}

		create_random_vector_fx( exc4k, L_FRAME16k / 4, bwe_seed );

		/* Ensure pow22 is greater than zero when computing normalization */
		Lmax = 0;
		FOR( i = 0; i < L_FRAME16k / 4; i++ )
		{
		exc4k_32[i] = Mult_32_16( excNoisyEnv[i], exc4k[i] );
		move32(); /* Q_bwe_exc+6 */
		Lmax = L_max( Lmax, L_abs( exc4k_32[i] ) );
		}

		IF( Lmax == 0 )
		{
		pow22 = 1;
		move16();
		n2 = 0;
		move16();
		set16_fx( exc4k_frac, 0, L_FRAME16k / 4 );
		}
		ELSE
		{
		n2 = norm_l( Lmax );
		FOR( i = 0; i < L_FRAME16k / 4; i++ )
		{
		#ifdef BASOP_NOGLOB
		exc4k_frac[i] = extract_h( L_shl_o( exc4k_32[i], n2, &Overflow ) ); /* Q(14-n2) */
		#else
		exc4k_frac[i] = extract_h( L_shl( exc4k_32[i], n2 ) ); /* Q(14-n2) */
		#endif
		}
		n2 = 30 - n2 - ( Q_bwe_exc + 6 );
		pow22 = 1;
		FOR( i = 0; i < L_FRAME16k / 4; i++ )
		{
		#ifdef BASOP_NOGLOB
		L_tmp = L_mult_o( exc4k_frac[i], exc4k_frac[i], &Overflow ); /* Q29 */
		pow22 = L_add_o( pow22, L_shr( L_tmp, 10 ), &Overflow ); /* Q22 */
		#else
		L_tmp = L_mult( exc4k_frac[i], exc4k_frac[i] ); /* Q29 */
		pow22 = L_add( pow22, L_shr( L_tmp, 7 ) ); /* Q22 */
		#endif
		}
		}

		test();
		test();
		IF( EQ_16( coder_type, UNVOICED ) \|\| ( igf_flag != 0 && LT_16( avg_voice_fac, 6654 ) ) )
		{
		L_tmp = root_a_over_b_fx( pow1, sub( 19, shl( n1, 1 ) ), pow22, sub( 19, shl( n2, 1 ) ), &exp );
		#ifdef BASOP_NOGLOB
		scale = round_fx_o( L_shl_o( L_tmp, exp, &Overflow ), &Overflow ); /Q15 /
		#else
		scale = round_fx( L_shl( L_tmp, exp ) ); /Q15 /
		#endif
		sc = sub( add( n2, Q_bwe_exc ), 14 );
		FOR( i = 0; i < L_FRAME16k / 4; i++ )
		{
		#ifdef BASOP_NOGLOB
		exc4kWhtnd[i] = round_fx_o( L_shl_o( L_mult_o( exc4k_frac[i], scale, &Overflow ), sc, &Overflow ), &Overflow ); /* Q_bwe_exc+n2-10+16+ Q_bwe_exc + n2 -14 -16 = //Q_bwe_exc */
		#else
		exc4kWhtnd[i] = round_fx( L_shl( L_mult( exc4k_frac[i], scale ), sc ) ); /* Q_bwe_exc+n2-10+16+ Q_bwe_exc + n2 -14 -16 = //Q_bwe_exc */
		#endif
		}
		}
		ELSE
		{
		sc = sub( add( n2, Q_bwe_exc ), 14 ); /* Q_bwe_exc+n2-14*/

		k = 0;
		FOR( i = 0; i < 4; i++ )
		{
		test();
		IF( igf_flag != 0 && EQ_16( coder_type, VOICED ) )
		{
		/tmp_vfac = 2voice_factors[i];
		tmp_vfac = min(1, tmp_vfac);*/
		#ifdef BASOP_NOGLOB
		tmp_vfac = shl_o( voice_factors[i], 1, &Overflow );
		#else
		BASOP_SATURATE_WARNING_OFF_EVS
		tmp_vfac = shl( voice_factors[i], 1 );
		BASOP_SATURATE_WARNING_ON_EVS
		#endif
		}
		ELSE
		{
		tmp_vfac = voice_factors[i];
		move16();
		}

		Ltemp1 = root_a_fx( L_deposit_h( tmp_vfac ), 31, &exp );
		#ifdef BASOP_NOGLOB
		temp1 = round_fx_o( L_shl_o( Ltemp1, exp, &Overflow ), &Overflow ); /* Q15 */
		#else
		temp1 = round_fx( L_shl( Ltemp1, exp ) ); /* Q15 */
		#endif
		L_tmp = Mult_32_16( pow1, sub( 32767, tmp_vfac ) ); /* Q22*/
		Ltemp2 = root_a_over_b_fx( L_tmp, sub( 19, shl( n1, 1 ) ), pow22, sub( 19, shl( n2, 1 ) ), &exp );
		#ifdef BASOP_NOGLOB
		temp2 = round_fx_o( L_shl_o( Ltemp2, exp, &Overflow ), &Overflow ); /* Q15 */
		#else
		temp2 = round_fx( L_shl( Ltemp2, exp ) ); /* Q15 */
		#endif
		FOR( j = 0; j < L_FRAME16k / 16; j++ )
		{
		#ifdef BASOP_NOGLOB
		L_tmp = L_mult_o( temp1, exc4kWhtnd[k], &Overflow ); /* Q(16+Q_bwe_exc) */
		L_tmp = L_add_o( L_tmp, L_shl_o( L_mult_o( temp2, exc4k_frac[k], &Overflow ), sc, &Overflow ), &Overflow ); /* Q(16+Q_bwe_exc) */
		exc4kWhtnd[k] = round_fx_o( L_tmp, &Overflow ); /* Q_bwe_exc */
		#else
		L_tmp = L_mult( temp1, exc4kWhtnd[k] ); /* Q(16+Q_bwe_exc) */
		L_tmp = L_add( L_tmp, L_shl( L_mult( temp2, exc4k_frac[k] ), sc ) ); /* Q(16+Q_bwe_exc) */
		exc4kWhtnd[k] = round_fx( L_tmp ); /* Q_bwe_exc */
		#endif
		k++;
		}
		}
		}
		}

		Syn_filt_s( 0, lpc_shb, LPC_SHB_ORDER_WB, exc4kWhtnd, excSHB, L_FRAME16k / 4, state_lpc_syn, 1 );

		return;
		}
		#endif

		void GenShapedWBExcitation_fx(
		Word16* excSHB, /* o : synthesized shaped shb exctiation Q_bwe_exc*/
		const Word16* lpc_shb, /* i : lpc coefficients Q12*/

lib_dec/init_dec.c

+12 −0

Original line number	Diff line number	Diff line
		@@ -374,6 +374,13 @@ ivas_error init_decoder(
		}

		td_bwe_dec_init( st->hBWE_TD, st->extl, st->output_Fs );

		#ifdef IVAS_FLOAT_FIXED
		st->prev_Q_bwe_exc = 31;
		st->prev_Qx = 0;
		st->prev_ener_fx_Q = 31;
		st->prev_frame_pow_exp = 0;
		#endif
		}
		else
		{
		@@ -395,6 +402,11 @@ ivas_error init_decoder(
		}

		fd_bwe_dec_init_flt( st->hBWE_FD );
		#ifdef IVAS_FLOAT_FIXED
		st->hBWE_FD->old_wtda_swb_fx_exp = 0;
		st->hBWE_FD->mem_imdct_exp_fx = 0;
		st->prev_Q_synth = 0;
		#endif
		}
		else
		{

lib_dec/ivas_core_dec.c

+171 −0

File changed.

Preview size limit exceeded, changes collapsed.