Encoder clean up in core-coder functions (dbfff24f) · Commits · SA4 / Audio / IVAS BASOP

lib_com/ivas_prot.h

+24 −5

Original line number	Diff line number	Diff line
		@@ -3223,22 +3223,41 @@ int16_t read_GR0(
		/----------------------------------------------------------------------------------
		* MCT prototypes
		----------------------------------------------------------------------------------/

		#ifdef IVAS_FLOAT_FIXED
		void ivas_mdct_core_whitening_enc_fx(
		CPE_ENC_HANDLE hCPE, /* i/o: CPE encoder structure */
		Word16 new_samples_fx[CPE_CHANNELS][L_INP], /* i : new samples */
		Word16 old_wsp_fx[CPE_CHANNELS][L_WSP], /* i : 12.8kHz weighted speech (for LTP */
		Word16 pitch_buf[CPE_CHANNELS][NB_SUBFR16k], /* o : floating pitch for each subframe */
		Word32 mdst_spectrum_long[CPE_CHANNELS], / o : buffer for MDST spectrum */
		int16_t tnsBits[CPE_CHANNELS][NB_DIV], /* o : buffer TNS bits */
		Word32 orig_spectrum_long[CPE_CHANNELS], / o : origingal spectrum w/o whitening */
		int16_t tnsSize[CPE_CHANNELS][NB_DIV], /* o : size of TNS */
		int16_t p_param[CPE_CHANNELS][NB_DIV], /* o : pointer to parameter array */
		BSTR_ENC_HANDLE hBstr, /* i/o: encoder bitstream handle */
		const int16_t mct_on, /* i : flag mct block (1) or stereo (0) */
		const int16_t nChannels, /* i : total number of coded channels */
		Word16 mdst_spectrum_e[CPE_CHANNELS][NB_DIV],
		Word16 orig_spectrum_e[CPE_CHANNELS][NB_DIV]
		);
		#else
		void ivas_mdct_core_whitening_enc(
		CPE_ENC_HANDLE hCPE, /* i/o: CPE encoder structure */
		Word16 new_samples_fx[CPE_CHANNELS][L_INP], /* i : new samples */
		Word16 old_wsp_fx[CPE_CHANNELS][L_WSP], /* i : 12.8kHz weighted speech (for LTP */
		Word16 pitch_buf[CPE_CHANNELS][NB_SUBFR16k], /* o : floating pitch for each subframe */
		float mdst_spectrum_long[CPE_CHANNELS], / o : buffer for MDST spectrum */
		Word32 mdst_spectrum_long[CPE_CHANNELS], / o : buffer for MDST spectrum */
		int16_t tnsBits[CPE_CHANNELS][NB_DIV], /* o : buffer TNS bits */
		float orig_spectrum_long[CPE_CHANNELS], / o : origingal spectrum w/o whitening */
		Word32 orig_spectrum_long[CPE_CHANNELS], / o : origingal spectrum w/o whitening */
		int16_t tnsSize[CPE_CHANNELS][NB_DIV], /* o : size of TNS */
		int16_t p_param[CPE_CHANNELS][NB_DIV], /* o : pointer to parameter array */
		BSTR_ENC_HANDLE hBstr, /* i/o: encoder bitstream handle */
		const int16_t mct_on, /* i : flag mct block (1) or stereo (0) */
		const int16_t nChannels /* i : total number of coded channels */
		const int16_t nChannels, /* i : total number of coded channels */
		Word16 mdst_spectrum_e[CPE_CHANNELS][NB_DIV],
		Word16 orig_spectrum_e[CPE_CHANNELS][NB_DIV]
		);

		#endif
		void ivas_mct_core_enc(
		const IVAS_FORMAT ivas_format, /* i : IVAS format */
		MCT_ENC_HANDLE hMCT, /* i/o: MCT encoder structure */

lib_enc/core_enc_init.c

+2 −2

Original line number	Diff line number	Diff line
		@@ -156,7 +156,7 @@ void init_coder_ace_plus_ivas_fx(
		{
		st->acelpFramesCount = 0;
		move16();
		st->prevTempFlatness_fx = 16 /1.0f Q4/;
		st->prevTempFlatness_fx = 128 /1.0f Q7/;
		move16();
		}

		@@ -405,7 +405,7 @@ void init_coder_ace_plus(
		#endif
		#ifdef IVAS_FLOAT_FIXED
		#ifdef MSAN_FIX
		st->prevTempFlatness_fx = 16; // Q4
		st->prevTempFlatness_fx = 128; // Q7
		#endif
		#endif
		}

lib_enc/ivas_core_enc.c

+61 −3

Original line number	Diff line number	Diff line
		@@ -478,22 +478,80 @@ ivas_error ivas_core_enc(
		{
		IF( MCT_flag )
		{
		Word16 mdst_spectrum_e[CPE_CHANNELS][NB_DIV];
		Word16 orig_spectrum_e[CPE_CHANNELS][NB_DIV];
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		Word16 old_inp_16k_fx[CPE_CHANNELS][L_INP];
		Word16 old_wsp_fx[CPE_CHANNELS][L_WSP], q_fac[CPE_CHANNELS];
		Word16 old_wsp_fx[CPE_CHANNELS][L_WSP], q_fac[CPE_CHANNELS], q_fac1 = Q15;
		Word16 pitch_buf_fx_new[CPE_CHANNELS][NB_SUBFR16k]; /* Q6 */
		for ( i = 0; i < CPE_CHANNELS; i++ )
		{
		Encoder_State *st1 = sts[i];
		TCX_ENC_HANDLE hTcxEncCh = st1->hTcxEnc;
		Word16 nSubframes = ( st1->hTcxEnc->tcxMode == TCX_20 ) ? 1 : NB_DIV;
		for ( n = 0; n < nSubframes; n++ )
		{
		Word16 nl = (Word16) ( (float) ( hTcxEncCh->L_frameTCX / nSubframes ) * N_ZERO_MDCT_NS / FRAME_SIZE_NS );
		floatToFixed_arr( &hTcxEncCh->new_speech_TCX_flt[nl - ( hTcxEncCh->L_frameTCX / nSubframes )],
		&hTcxEncCh->new_speech_TCX[nl - ( hTcxEncCh->L_frameTCX / nSubframes )],
		0,
		2 * hTcxEncCh->L_frameTCX / nSubframes );

		IF( n == 0 )
		{
		floatToFixed_arr( &hTcxEncCh->speech_TCX_flt[-( sts[0]->hTcxCfg->tcx_mdct_window_min_lengthFB - 0 )],
		&hTcxEncCh->speech_TCX[-( sts[0]->hTcxCfg->tcx_mdct_window_min_lengthFB - 0 )],
		0,
		sts[0]->hTcxCfg->tcx_mdct_window_min_lengthFB );
		}
		}

		floatToFixed_arr( old_inp_16k[i], old_inp_16k_fx[i], 0, L_INP );
		q_fac[i] = Q_factor_arr( old_wsp[i], L_WSP );
		floatToFixed_arr( old_wsp[i], old_wsp_fx[i], q_fac[i], L_WSP );
		if ( st1->element_mode == IVAS_CPE_DFT )
		{
		q_fac1 = Q_factor_arr( st1->buf_wspeech_enc_flt, L_FRAME16k + L_SUBFR + L_FRAME16k + L_NEXT_MAX_16k + 320 );
		floatToFixed_arr( st1->buf_wspeech_enc_flt, st1->buf_wspeech_enc, q_fac1, L_FRAME16k + L_SUBFR + L_FRAME16k + L_NEXT_MAX_16k + 320 );
		st1->exp_buf_wspeech_enc = 15 - q_fac1;
		}
		else if ( st1->element_mode != IVAS_CPE_MDCT )
		{
		q_fac1 = Q_factor_arr( st1->buf_speech_enc_flt, L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k );
		floatToFixed_arr( st1->buf_speech_enc_flt, st1->buf_speech_enc, q_fac1, L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k );
		st1->exp_buf_speech_enc = 15 - q_fac1;
		floatToFixed_arr( st1->hTcxEnc->buf_speech_ltp_flt, st1->hTcxEnc->buf_speech_ltp, q_fac1, L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k );
		st1->hTcxEnc->exp_buf_speech_ltp = 15 - q_fac1;
		}
		floatToFixed_arr( st1->input_buff, st1->input_buff_fx, 0, L_FRAME48k + L_FRAME48k + NS2SA( 48000, DELAY_FIR_RESAMPL_NS ) );
		}
		#endif
		ivas_mdct_core_whitening_enc( hCPE, old_inp_16k_fx, old_wsp_fx, pitch_buf_fx_new, hMCT->p_mdst_spectrum_long[cpe_id], hMCT->tnsBits[cpe_id], hMCT->p_orig_spectrum_long[cpe_id],
		hMCT->tnsSize[cpe_id], hMCT->p_param[cpe_id], hMCT->hBstr, 1, hMCT->nchan_out_woLFE );
		ivas_mdct_core_whitening_enc_fx( hCPE, old_inp_16k_fx, old_wsp_fx, pitch_buf_fx_new, hMCT->p_mdst_spectrum_long_fx[cpe_id], hMCT->tnsBits[cpe_id], hMCT->p_orig_spectrum_long_fx[cpe_id],
		hMCT->tnsSize[cpe_id], hMCT->p_param[cpe_id], hMCT->hBstr, 1, hMCT->nchan_out_woLFE, mdst_spectrum_e, orig_spectrum_e );
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		for ( i = 0; i < CPE_CHANNELS; i++ )
		{
		Encoder_State *st1 = sts[i];
		Word16 nSubframes = ( st1->hTcxEnc->tcxMode == TCX_20 ) ? 1 : NB_DIV;
		if ( st1->element_mode == IVAS_CPE_DFT )
		{
		fixedToFloat_arr( st1->buf_wspeech_enc, st1->buf_wspeech_enc_flt, q_fac1, L_FRAME16k + L_SUBFR + L_FRAME16k + L_NEXT_MAX_16k + 320 );
		}
		else if ( st1->element_mode != IVAS_CPE_MDCT )
		{
		fixedToFloat_arr( st1->hTcxEnc->buf_speech_ltp, st1->hTcxEnc->buf_speech_ltp_flt, q_fac1, L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k );
		}
		fixedToFloat_arr( st1->input_buff_fx, st1->input_buff, 0, L_FRAME48k + L_FRAME48k + NS2SA( 48000, DELAY_FIR_RESAMPL_NS ) );

		fixedToFloat_arrL( st1->hTcxEnc->spectrum_fx[0], st1->hTcxEnc->spectrum[0], 31 - st1->hTcxEnc->spectrum_e[0], st1->hTcxEnc->L_frameTCX / nSubframes );
		fixedToFloat_arrL( hMCT->p_orig_spectrum_long_fx[cpe_id][i], hMCT->p_orig_spectrum_long[cpe_id][i], 31 - orig_spectrum_e[i][0], N_TCX10_MAX );
		fixedToFloat_arrL( hMCT->p_mdst_spectrum_long_fx[cpe_id][i], hMCT->p_mdst_spectrum_long[cpe_id][i], 31 - mdst_spectrum_e[i][0], N_TCX10_MAX );
		if ( nSubframes == NB_DIV )
		{
		fixedToFloat_arrL( st1->hTcxEnc->spectrum_fx[1], st1->hTcxEnc->spectrum[1], 31 - st1->hTcxEnc->spectrum_e[1], st1->hTcxEnc->L_frameTCX / nSubframes );
		fixedToFloat_arrL( hMCT->p_orig_spectrum_long_fx[cpe_id][i] + N_TCX10_MAX, hMCT->p_orig_spectrum_long[cpe_id][i] + N_TCX10_MAX, 31 - orig_spectrum_e[i][1], N_TCX10_MAX );
		fixedToFloat_arrL( hMCT->p_mdst_spectrum_long_fx[cpe_id][i] + N_TCX10_MAX, hMCT->p_mdst_spectrum_long[cpe_id][i] + N_TCX10_MAX, 31 - mdst_spectrum_e[i][1], N_TCX10_MAX );
		}
		fixedToFloat_arr( old_wsp_fx[i], old_wsp[i], q_fac[i], L_WSP );
		}
		#endif

lib_enc/ivas_core_pre_proc_front.c

+0 −27

Original line number	Diff line number	Diff line
		@@ -2588,33 +2588,6 @@ ivas_error pre_proc_front_ivas_fx(
		/* Switch to ACELP for non-harmonic transient signals */
		else if ( ( ( element_mode >= IVAS_CPE_DFT && element_brate <= IVAS_16k4 ) \|\| ( element_mode == IVAS_SCE && element_brate < SCE_SMC_THR ) ) && ( loc_harm[0] != 1 ) && smc_dec == MUSIC )
		{
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		IF( EQ_16( element_mode, IVAS_SCE ) )
		{
		Word16 q1 = Q7;
		Word16 q2 = Q7;
		st->hTranDet->subblockEnergies.subblockNrg_e = sub( 31, q1 );
		st->hTranDet->subblockEnergies.accSubblockNrg_e = sub( 31, q2 );
		move16();
		move16();
		move16();
		move16();
		}
		ELSE IF( EQ_16( element_mode, IVAS_CPE_DFT ) )
		{
		FOR( i = 0; i < CPE_CHANNELS; i++ )
		{
		Word16 q1 = Q7;
		Word16 q2 = Q7;
		hCPE->hCoreCoder[i]->hTranDet->subblockEnergies.subblockNrg_e = sub( 31, q1 );
		hCPE->hCoreCoder[i]->hTranDet->subblockEnergies.accSubblockNrg_e = sub( 31, q2 );
		move16();
		move16();
		move16();
		move16();
		}
		}
		#endif
		IF( EQ_16( element_mode, IVAS_SCE ) )
		{
		if ( transient_analysis_ivas_fx( st->hTranDet, st->hNoiseEst->cor_map_fx, Q15, st->hNoiseEst->multi_harm_limit_fx, Q9 ) )

lib_enc/ivas_cpe_enc.c

+1 −1

Original line number	Diff line number	Diff line
		@@ -459,7 +459,7 @@ ivas_error ivas_cpe_enc_fx(
		RunTransientDetection_ivas_fx( sts[n]->input_fx, input_frame, sts[n]->hTranDet, q_inp ); // Note q of sts[n]->input_fx changes inside function
		}

		currFlatness_fx[n] = GetTCXAvgTemporalFlatnessMeasure_ivas_fx( (const TransientDetection *) sts[n]->hTranDet, NSUBBLOCKS, 0 ); // Q4
		currFlatness_fx[n] = GetTCXAvgTemporalFlatnessMeasure_ivas_fx( (const TransientDetection *) sts[n]->hTranDet, NSUBBLOCKS, 0 ); // Q7
		move16();
		}