Merge branch 'main' into update_crend (6ad5703c) · Commits · SA4 / Audio / IVAS BASOP

lib_com/bitstream.c

+23 −0

Original line number	Diff line number	Diff line
		@@ -2170,6 +2170,7 @@ uint16_t get_indice_1(
		* Reset the buffer of encoder indices
		-------------------------------------------------------------------/

		#ifndef IVAS_FLOAT_FIXED
		void reset_indices_enc(
		BSTR_ENC_HANDLE hBstr, /* i/o: encoder bitstream handle */
		const Word16 max_num_indices /* i : max number of indices */
		@@ -2187,6 +2188,28 @@ void reset_indices_enc(

		return;
		}
		#else
		void reset_indices_enc(
		BSTR_ENC_HANDLE hBstr, /* i/o: encoder bitstream handle */
		const Word16 max_num_indices /* i : max number of indices */
		)
		{
		Word16 i;

		hBstr->nb_bits_tot = 0;
		move16();
		hBstr->nb_ind_tot = 0;
		move16();

		FOR( i = 0; i < max_num_indices; i++ )
		{
		hBstr->ind_list[i].nb_bits = -1;
		move16();
		}

		return;
		}
		#endif

		/-------------------------------------------------------------------
		* reset_indices_dec()

lib_com/cldfb.c

+21 −19

Original line number	Diff line number	Diff line
		@@ -1950,7 +1950,9 @@ void analysisCldfbEncoder_ivas(

		void analysisCldfbEncoder_ivas_fx(
		Encoder_State st, / i/o: encoder state structure */
		const Word16 *timeIn,
		const Word32 timeIn, /q11*/
		Word16 timeInq,
		Word16 samplesToProcess,
		Word32 realBuffer[CLDFB_NO_COL_MAX][CLDFB_NO_CHANNELS_MAX],
		Word32 imagBuffer[CLDFB_NO_COL_MAX][CLDFB_NO_CHANNELS_MAX],
		Word16 realBuffer16[CLDFB_NO_COL_MAX][CLDFB_NO_CHANNELS_MAX],
		@@ -1965,7 +1967,6 @@ void analysisCldfbEncoder_ivas_fx(
		Word32 *ppBuf_Imag[CLDFB_NO_COL_MAX];
		Word16 *ppBuf_Real16[CLDFB_NO_COL_MAX];
		Word16 *ppBuf_Imag16[CLDFB_NO_COL_MAX];
		Word32 workBuffer[256];

		FOR( i = 0; i < CLDFB_NO_COL_MAX; i++ )
		{
		@@ -1974,13 +1975,14 @@ void analysisCldfbEncoder_ivas_fx(
		ppBuf_Real16[i] = &realBuffer16[i][0];
		ppBuf_Imag16[i] = &imagBuffer16[i][0];
		}
		cldfbAnalysis_ivas_fx( timeIn, ppBuf_Real, ppBuf_Imag, samplesToProcess, st->cldfbAnaEnc );

		cldfbAnalysisFiltering( st->cldfbAnaEnc, ppBuf_Real, ppBuf_Imag, scale, timeIn, 0, CLDFB_NO_COL_MAX, workBuffer );

		scale->lb_scale = sub( 16 + 5, timeInq );
		enerScale.lb_scale = negate( scale->lb_scale );
		enerScale.lb_scale16 = negate( scale->lb_scale );
		move16();
		move16();
		move16();

		AnalysisPostSpectrumScaling_Fx( st->cldfbAnaEnc, ppBuf_Real, ppBuf_Imag, ppBuf_Real16, ppBuf_Imag16, &enerScale.lb_scale16 );

		@@ -2793,7 +2795,7 @@ static void cldfb_init_proto_and_twiddles_enc(
		#ifdef IVAS_FLOAT_FIXED
		hs->p_filter_sf = (Word16) 17036;
		hs->scale = cldfb_scale_2_5ms[0];
		hs->p_filter = cldfb_protoFilter_2_5ms[0];
		hs->p_filter = CLDFB80_10_fx;
		#endif
		}
		else if ( hs->prototype == CLDFB_PROTOTYPE_5_00MS )
		@@ -2808,7 +2810,7 @@ static void cldfb_init_proto_and_twiddles_enc(
		hs->rot_vec_syn_delay_im = rot_vec_delay_im_LDQMF;
		#ifdef IVAS_FLOAT_FIXED
		hs->p_filter_sf = (Word16) 15388;
		hs->p_filter = cldfb_protoFilter_5_0ms[0];
		hs->p_filter = LDQMF_10_fx;
		hs->scale = cldfb_scale_5_0ms[0];
		#endif
		}
		@@ -2843,7 +2845,7 @@ static void cldfb_init_proto_and_twiddles_enc(
		hs->rot_vec_syn_delay_im = NULL;
		#ifdef IVAS_FLOAT_FIXED
		hs->p_filter_sf = (Word16) 17051;
		hs->p_filter = cldfb_protoFilter_2_5ms[1];
		hs->p_filter = CLDFB80_16_fx;
		hs->scale = cldfb_scale_2_5ms[1];
		#endif
		}
		@@ -2859,7 +2861,7 @@ static void cldfb_init_proto_and_twiddles_enc(
		hs->rot_vec_syn_delay_im = rot_vec_delay_im_LDQMF;
		#ifdef IVAS_FLOAT_FIXED
		hs->p_filter_sf = (Word16) 15388;
		hs->p_filter = cldfb_protoFilter_5_0ms[1];
		hs->p_filter = LDQMF_16_fx;
		hs->scale = cldfb_scale_5_0ms[1];
		#endif
		}
		@@ -2900,7 +2902,7 @@ static void cldfb_init_proto_and_twiddles_enc(
		hs->rot_vec_syn_delay_im = NULL;
		#ifdef IVAS_FLOAT_FIXED
		hs->p_filter_sf = (Word16) 17050;
		hs->p_filter = cldfb_protoFilter_2_5ms[2];
		hs->p_filter = CLDFB80_20_fx;
		hs->scale = cldfb_scale_2_5ms[2];
		#endif
		}
		@@ -2916,7 +2918,7 @@ static void cldfb_init_proto_and_twiddles_enc(
		hs->rot_vec_syn_delay_im = rot_vec_delay_im_LDQMF;
		#ifdef IVAS_FLOAT_FIXED
		hs->p_filter_sf = (Word16) 15390;
		hs->p_filter = cldfb_protoFilter_5_0ms[2];
		hs->p_filter = LDQMF_20_fx;
		hs->scale = cldfb_scale_5_0ms[2];
		#endif
		}
		@@ -2958,7 +2960,7 @@ static void cldfb_init_proto_and_twiddles_enc(
		#ifdef IVAS_FLOAT_FIXED
		hs->p_filter_sf = (Word16) 17051;
		hs->scale = cldfb_scale_2_5ms[6];
		hs->p_filter = cldfb_protoFilter_2_5ms[6];
		hs->p_filter = CLDFB80_30_fx;
		#endif
		}
		else if ( hs->prototype == CLDFB_PROTOTYPE_5_00MS )
		@@ -2974,7 +2976,7 @@ static void cldfb_init_proto_and_twiddles_enc(
		#ifdef IVAS_FLOAT_FIXED
		hs->p_filter_sf = (Word16) 15388;
		hs->scale = cldfb_scale_5_0ms[6];
		hs->p_filter = cldfb_protoFilter_5_0ms[6];
		hs->p_filter = LDQMF_30_fx;
		#endif
		}
		break;
		@@ -3008,7 +3010,7 @@ static void cldfb_init_proto_and_twiddles_enc(
		hs->rot_vec_syn_delay_im = NULL;
		#ifdef IVAS_FLOAT_FIXED
		hs->p_filter_sf = (Word16) 17050;
		hs->p_filter = cldfb_protoFilter_2_5ms[3];
		hs->p_filter = CLDFB80_32_fx;
		hs->scale = cldfb_scale_2_5ms[3];
		#endif
		}
		@@ -3025,7 +3027,7 @@ static void cldfb_init_proto_and_twiddles_enc(
		#ifdef IVAS_FLOAT_FIXED
		hs->p_filter_sf = (Word16) 15392;
		hs->scale = cldfb_scale_5_0ms[3];
		hs->p_filter = cldfb_protoFilter_5_0ms[3];
		hs->p_filter = LDQMF_32_fx;
		#endif
		}
		break;
		@@ -3067,7 +3069,7 @@ static void cldfb_init_proto_and_twiddles_enc(
		hs->p_filter_sf = (Word16) 17051;
		hs->q_scale = norm_s( (Word16) CLDFB80_40_SCALE );
		hs->scale = cldfb_scale_2_5ms[4];
		hs->p_filter = cldfb_protoFilter_2_5ms[4];
		hs->p_filter = CLDFB80_40_fx;
		#endif
		}
		else if ( hs->prototype == CLDFB_PROTOTYPE_5_00MS )
		@@ -3086,7 +3088,7 @@ static void cldfb_init_proto_and_twiddles_enc(
		hs->rot_vec_syn_delay_re_fx = rot_vec_delay_re_LDQMF_fx;
		hs->rot_vec_syn_delay_im_fx = rot_vec_delay_im_LDQMF_fx;
		hs->p_filter_sf = (Word16) 15391;
		hs->p_filter = cldfb_protoFilter_5_0ms[4];
		hs->p_filter = LDQMF_40_fx;
		hs->scale = cldfb_scale_5_0ms[4];
		#endif
		}
		@@ -3131,7 +3133,7 @@ static void cldfb_init_proto_and_twiddles_enc(
		hs->rot_vec_syn_delay_re_fx = NULL;
		hs->rot_vec_syn_delay_im_fx = NULL;
		hs->p_filter_sf = (Word16) 17051;
		hs->p_filter = cldfb_protoFilter_2_5ms[5];
		hs->p_filter = CLDFB80_60_fx;
		hs->scale = cldfb_scale_2_5ms[5];
		#endif
		}
		@@ -3151,7 +3153,7 @@ static void cldfb_init_proto_and_twiddles_enc(
		hs->rot_vec_syn_delay_re_fx = rot_vec_delay_re_LDQMF_fx;
		hs->rot_vec_syn_delay_im_fx = rot_vec_delay_im_LDQMF_fx;
		hs->p_filter_sf = (Word16) 15391;
		hs->p_filter = cldfb_protoFilter_5_0ms[5];
		hs->p_filter = LDQMF_60_fx;
		hs->scale = cldfb_scale_5_0ms[5];
		#endif
		}

lib_com/cnst.h

+7 −0

Original line number	Diff line number	Diff line
		@@ -745,12 +745,14 @@ enum
		#define PIT_UP_SAMP6 6
		#define PIT_L_INTERPOL6_2 17
		#define PIT_FIR_SIZE6_2 ( PIT_UP_SAMP6 * PIT_L_INTERPOL6_2 + 1 )
		#define E_MIN_Q11_FX 7 /* minimum allowable energy in Q11*/
		#define E_MIN 0.0035f /* minimum allowable energy */
		#define E_MIN_Q11_FX 7 /* minimum allowable energy in Q11*/
		#define STEP_DELTA 0.0625f /* quantization step for tilt compensation of gaussian cb. excitation */
		#define GAMMA_EV 0.92f /* weighting factor for core synthesis error weighting */
		#define FORMANT_SHARPENING_NOISE_THRESHOLD 21.0f /* lp_noise level above which formant sharpening is deactivated */
		#define E_MIN_FX 1 /* QSCALE (Q7)*/
		#define E_MIN_IVAS_FX 1835 /* (Q19) (E12) */
		#define STEP_DELTA_FX 11
		#define FORMANT_SHARPENING_NOISE_THRESHOLD_FX 5376 /* 21 (!8)lp_noise level above which formant sharpening is deactivated - at this level most of 20 dB SNR office noisy speech still uses sharpening */

		@@ -1010,6 +1012,7 @@ typedef enum
		#define BWD_COUNT_MAX 100 /* maximum value of BWD counter */
		#define BWD_N_BINS_MAX 13 /* maximum number of BWD bins */
		#define BWS_TRAN_PERIOD 5 /* BWS - number of frames for transition period */
		#define ONE_BY_BWS_TRAN_PERIOD_Q15 6554

		#define PREEMPH_FAC_FLT 0.68f /* preemphasis factor at 12.8kHz */
		#define PREEMPH_FAC_16k_FLT 0.72f
		@@ -1953,10 +1956,14 @@ typedef enum _DCTTYPE
		#define ATT_NSEG 32 /* strong attack detection - number of time blocks */

		#define TOD_NSPEC 80 /* number of spectral bins of the tonal detector */
		#define TOD_NSPEC_INV_Q31 26843546 /* inverse of number of spectral bins of the tonal detector */
		#define TOD_THR_MASS 0.86f /* initial value for the adaptive threshold of the tonal detector */
		#define P2A_FACT 0.9f /* long-term averaging factor for peak-to-average ratio */
		#define P2A_FACT_FX_Q15 29491 /* long-term averaging factor for peak-to-average ratio */
		#define THR_P2A_HIGH 95.0f /* higher threshold to detect strongly peaky signals at low bitrates*/
		#define THR_P2A 80.0f /* lower threshold to detect strongly peaky signals at higher bitrates */
		#define THR_P2A_HIGH_FX 95 /* Q0, higher threshold to detect strongly peaky signals at low bitrates*/
		#define THR_P2A_FX 80 /* Q0, lower threshold to detect strongly peaky signals at higher bitrates */

		/----------------------------------------------------------------------------------
		* LD music post-filter constants

lib_com/fd_cng_com_fx.c

+142 −0

Original line number	Diff line number	Diff line
		@@ -10,6 +10,7 @@
		#include "rom_basop_util.h"
		#include "rom_com.h"
		#include "prot_fx.h"
		#include "prot_fx_enc.h"
		#include "ivas_prot_fx.h"

		#define DELTA_SHIFT 2
		@@ -2471,6 +2472,147 @@ void AnalysisSTFT_fx(
		}
		#endif // IVAS_FLOAT_FIXED

		#ifdef IVAS_FLOAT_FIXED
		/*-------------------------------------------------------------------
		* SynthesisSTFT_enc_ivas_fx()
		*
		* STFT synthesis filterbank
		-------------------------------------------------------------------/

		void SynthesisSTFT_enc_ivas_fx(
		Word32 fftBuffer, / i : pointer to FFT bins */
		Word16 fftBufferExp, /* i : exponent of FFT bins */
		Word16 timeDomainOutput, / o : pointer to time domain signal */
		Word16 olapBuffer, / i/o : pointer to overlap buffer */
		const PWord16 olapWin, / i : pointer to overlap window */
		Word16 tcx_transition,
		HANDLE_FD_CNG_COM hFdCngCom, /* i/o : pointer to FD_CNG structure containing all buffers and variables */
		Word16 gen_exc,
		Word16 Q_new, / i : Q of generated exc_cng */
		const Word16 element_mode, /* i : element mode */
		const Word16 nchan_out /* i : number of output channels */
		)
		{
		Word16 i, len, scale, tmp;
		Word16 len2, len3, len4;
		Word16 buf[M + 1 + L_FRAME16k];


		/* Perform IFFT */
		scale = 0;
		move16();
		BASOP_rfft( fftBuffer, hFdCngCom->fftlen, &scale, 1 );
		fftBufferExp = add( fftBufferExp, scale );
		hFdCngCom->fftBuffer_exp = fftBufferExp;
		move16();

		fftBufferExp = add( fftBufferExp, hFdCngCom->fftlenShift );

		/* Perform overlap-add */
		/* Handle overlap in P/S domain for stereo */
		test();
		test();
		IF( ( EQ_16( element_mode, IVAS_CPE_TD ) \|\| EQ_16( element_mode, IVAS_CPE_DFT ) ) && EQ_16( nchan_out, 2 ) )
		{
		Copy( olapBuffer + 3 * hFdCngCom->frameSize / 4 - ( M + 1 ), buf, hFdCngCom->frameSize + M + 1 );
		set16_fx( olapBuffer, 0, hFdCngCom->fftlen );
		}
		ELSE
		{
		Copy( olapBuffer + hFdCngCom->frameSize, olapBuffer, hFdCngCom->frameSize );
		set16_fx( olapBuffer + hFdCngCom->frameSize, 0, hFdCngCom->frameSize );
		}
		len2 = shr( hFdCngCom->fftlen, 2 );
		len4 = shr( hFdCngCom->fftlen, 3 );
		len3 = add( len2, len4 );
		len = add( hFdCngCom->frameSize, len4 );
		IF( tcx_transition )
		{
		FOR( i = 0; i < len; i++ )
		{
		olapBuffer[i] = round_fx_sat( L_shl_sat( fftBuffer[i], sub( fftBufferExp, 15 ) ) );
		move16();
		}
		}
		ELSE
		{
		FOR( i = 0; i < len4; i++ )
		{
		olapBuffer[i + 1 * len4] = add_sat( olapBuffer[i + 1 * len4], mult_r( round_fx_sat( L_shl_sat( fftBuffer[i + 1 * len4], sub( fftBufferExp, 15 ) ) ), olapWin[i].v.im ) );
		move16();
		olapBuffer[i + 2 * len4] = add_sat( olapBuffer[i + 2 * len4], mult_r( round_fx_sat( L_shl_sat( fftBuffer[i + 2 * len4], sub( fftBufferExp, 15 ) ) ), olapWin[len4 - 1 - i].v.re ) );
		move16();
		}
		FOR( i = len3; i < len; i++ )
		{
		olapBuffer[i] = round_fx_sat( L_shl_sat( fftBuffer[i], sub( fftBufferExp, 15 ) ) );
		move16();
		}
		}

		FOR( i = 0; i < len4; i++ )
		{
		olapBuffer[i + 5 * len4] = mult_r( round_fx_sat( L_shl_sat( fftBuffer[i + 5 * len4], sub( fftBufferExp, 15 ) ) ), olapWin[i].v.re );
		move16();
		olapBuffer[i + 6 * len4] = mult_r( round_fx_sat( L_shl_sat( fftBuffer[i + 6 * len4], sub( fftBufferExp, 15 ) ) ), olapWin[len4 - 1 - i].v.im );
		move16();
		}

		len = add( len, len2 );
		FOR( i = len; i < hFdCngCom->fftlen; i++ )
		{
		olapBuffer[i] = 0;
		move16();
		}

		/* Get time-domain signal */
		FOR( i = 0; i < hFdCngCom->frameSize; i++ )
		{
		timeDomainOutput[i] = mult_r( olapBuffer[i + len4], hFdCngCom->fftlenFac );
		move16();
		}
		/* Generate excitation */
		test();
		test();
		IF( ( EQ_16( element_mode, IVAS_CPE_TD ) \|\| EQ_16( element_mode, IVAS_CPE_DFT ) ) && EQ_16( nchan_out, 2 ) )
		{
		FOR( i = 0; i < hFdCngCom->frameSize / 2; i++ )
		{
		buf[i + ( M + 1 )] = add( buf[i + ( M + 1 )], olapBuffer[i + hFdCngCom->frameSize / 4] );
		move16();
		}

		FOR( i = 0; i < M + 1 + hFdCngCom->frameSize; i++ )
		{
		buf[i] = mult_r( buf[i], hFdCngCom->fftlenFac );
		move16();
		}
		}
		ELSE
		{
		FOR( i = 0; i < M + 1 + hFdCngCom->frameSize; i++ )
		{
		buf[i] = mult_r( olapBuffer[i + len4 - M - 1], hFdCngCom->fftlenFac );
		move16();
		}
		tmp = buf[0];
		move16();
		}
		IF( EQ_16( gen_exc, 1 ) )
		{

		E_UTIL_f_preemph2( sub( *Q_new, 1 ), buf + 1, PREEMPH_FAC, M + hFdCngCom->frameSize, &tmp );
		Residu3_fx( hFdCngCom->A_cng, buf + 1 + M, hFdCngCom->exc_cng, hFdCngCom->frameSize, 1 );
		}
		IF( EQ_16( gen_exc, 2 ) )
		{
		*Q_new = E_UTIL_f_preemph3( buf + 1, PREEMPH_FAC, M + hFdCngCom->frameSize, &tmp, 1 );
		move16();
		Residu3_fx( hFdCngCom->A_cng, buf + 1 + M, hFdCngCom->exc_cng, hFdCngCom->frameSize, 1 );
		}
		}
		#endif

		/*-------------------------------------------------------------------
		* SynthesisSTFT()
		*

lib_com/hq_tools_fx.c

+2 −2

Original line number	Diff line number	Diff line
		@@ -431,7 +431,7 @@ void apply_noisefill_HQ_fx(
		{
		cb_buff[i] = CodeBook_mod[cb_pos++];
		move16();
		L_E_cb_vec = L_mac0( L_E_cb_vec, cb_buff[i], cb_buff[i] ); /Q24 (12+12) /
		L_E_cb_vec = L_mac0_sat( L_E_cb_vec, cb_buff[i], cb_buff[i] ); /Q24 (12+12) /

		if ( GE_16( cb_pos, cb_size ) )
		{
		@@ -446,7 +446,7 @@ void apply_noisefill_HQ_fx(
		{
		cb_buff[i] = CodeBook[cb_pos++];
		move16();
		L_E_cb_vec = L_mac0( L_E_cb_vec, cb_buff[i], cb_buff[i] ); /Q24 (12+12) /
		L_E_cb_vec = L_mac0_sat( L_E_cb_vec, cb_buff[i], cb_buff[i] ); /Q24 (12+12) /

		if ( GE_16( cb_pos, cb_size ) )
		{