Core coder and stereo path sub-funcs conversion and integration (dce4d2e3) · Commits · SA4 / Audio / IVAS BASOP

lib_com/cnst.h

+1 −0

Original line number	Diff line number	Diff line
		@@ -752,6 +752,7 @@ enum
		#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 */

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/ivas_prot.h

+23 −23

Original line number	Diff line number	Diff line
		@@ -2644,11 +2644,11 @@ void unclr_classifier_dft(
		CPE_ENC_HANDLE hCPE /* i/o: CPE encoder structure */
		);
		#endif

		#ifndef IVAS_FLOAT_FIXED
		void unclr_classifier_td(
		CPE_ENC_HANDLE hCPE /* i/o: CPE encoder structure */
		);

		#endif
		#ifdef IVAS_FLOAT_FIXED
		void xtalk_classifier_dft_fx(
		CPE_ENC_HANDLE hCPE, /* i/o: CPE encoder structure */
		@@ -2661,11 +2661,11 @@ void xtalk_classifier_dft(
		const int16_t itd, /* i : ITD from DFT stereo - used as a feature */
		const float gcc_phat[] /* i : GPHAT cross-channel correlation function */
		);

		#ifndef IVAS_FLOAT_FIXED
		void xtalk_classifier_td(
		CPE_ENC_HANDLE hCPE /* i/o: CPE encoder structure */
		);

		#endif

		/----------------------------------------------------------------------------------
		* TD Stereo prototypes
		@@ -2691,7 +2691,7 @@ void stereo_tdm_prep_dwnmx (
		const float input1, / i : right channel input */
		const int16_t input_frame /* i : frame lenght */
		);

		#ifndef IVAS_FLOAT_FIXED
		int16_t stereo_tdm_ener_analysis(
		const int16_t ivas_format, /* i : IVAS format */
		CPE_ENC_HANDLE hCPE, /* i : CPE structure */
		@@ -2699,7 +2699,7 @@ int16_t stereo_tdm_ener_analysis(
		int16_t tdm_SM_or_LRTD_Pri, / o : channel combination scheme flag in TD stereo OR LRTD primary channel */
		int16_t tdm_ratio_idx_SM / o : TDM ratio index for SM mode */
		);

		#endif
		void stereo_tdm_downmix(
		STEREO_TD_ENC_DATA_HANDLE hStereoTD, /* i : TD stereo IVAS encoder structure */
		float Left_in, / i/o: Left channel -> Primary channel */
		@@ -3661,7 +3661,7 @@ ivas_error front_vad_fx(
		CPE_ENC_HANDLE hCPE, /* i/o: CPE encoder structure, nullable */
		Encoder_State st, / i/o: encoder state structure */
		const ENCODER_CONFIG_HANDLE hEncoderConfig, /* i : configuration structure */
		FRONT_VAD_ENC_HANDLE hFrontVads, / i/o: front-VAD handles */
		FRONT_VAD_ENC_HANDLE hFrontVads, / i/o: FrontVad handles */
		const int16_t MCT_flag, /* i : hMCT handle allocated (1) or not (0) */
		const int16_t input_frame, /* i : frame length */
		int16_t vad_flag_dtx[], /* o : HE-SAD flag with additional DTX HO */
		@@ -3673,8 +3673,8 @@ ivas_error front_vad_fx(
		float band_energies_LR[2 * NB_BANDS], /* o : energy in critical bands without minimum noise floor E_MIN*/
		float PS_out, / o : energy spectrum */
		float Bin_E_out, / o : log-energy spectrum of the current frame */
		Word16 Qinp
		);
		Word16 Q_inp ,
		Word16 Q_add );
		#endif
		ivas_error front_vad_spar(
		SPAR_ENC_HANDLE hSpar, /* i/o: SPAR encoder structure */

lib_com/ivas_prot_fx.h

+20 −0

Original line number	Diff line number	Diff line
		@@ -2045,6 +2045,16 @@ void masa_compensate_two_dir_energy_ratio_index_fx(
		const Word16 hodirac_flag /* i : flag to indicate HO-DirAC mode */
		);

		#ifdef IVAS_FLOAT_FIXED
		int16_t stereo_tdm_ener_analysis_fx(
		const Word16 ivas_format, /* i : IVAS format */
		CPE_ENC_HANDLE hCPE, /* i : CPE structure */
		const Word16 input_frame, /* i : Number of samples */
		Word16 tdm_SM_or_LRTD_Pri, / o : channel combination scheme flag in TD stereo OR LRTD primary channel */
		Word16 tdm_ratio_idx_SM / o : TDM ratio index for SM mode */
		);
		#endif

		#ifdef IVAS_FLOAT_FIXED
		/! r: projected azimuth index /
		Word16 ivas_dirac_project_azimuth_index(
		@@ -3043,4 +3053,14 @@ void tdm_ol_pitch_comparison_fx(
		Word16 pitch_fr[CPE_CHANNELS][NB_SUBFR], /* i/o: fractional pitch values, Q6 */
		Word16 voicing_fr[CPE_CHANNELS][NB_SUBFR] /* i/o: fractional pitch gains, Q15 */
		);

		#ifdef IVAS_FLOAT_FIXED
		void xtalk_classifier_td_fx(
		CPE_ENC_HANDLE hCPE /* i/o: CPE encoder structure */
		);

		void unclr_classifier_td_fx(
		CPE_ENC_HANDLE hCPE /* i/o: CPE encoder structure */
		);
		#endif
		#endif

lib_com/prot.h

+7 −4

Original line number	Diff line number	Diff line
		@@ -5986,9 +5986,12 @@ void core_switching_pre_enc(

		void core_switching_post_enc(
		Encoder_State st, / i/o: encoder state structure */
		const float old_inp_12k8, / i : old input signal @12.8kHz */
		const float old_inp_16k, / i : old input signal @16kHz */
		const float A[] /* i : unquant LP filter coefs. */
		// const float old_inp_12k8, / i : old input signal @12.8kHz */
		float old_inp_12k8, / i : old input signal @12.8kHz */
		// const float old_inp_16k, / i : old input signal @16kHz */
		float old_inp_16k, / i : old input signal @16kHz */
		// const float A[] /* i : unquant. LP filter coefs. */
		float A[] /* i : unquant. LP filter coefs. */
		);

		ivas_error core_switching_post_dec(