Commit e0a989ab authored by Sandesh Venkatesh's avatar Sandesh Venkatesh
Browse files

Merge branch 'fxd_changes_related_to_float_updates_in_MR_797_788' into 'main' 

Fixed point Changes corresponding to float reference code updates made in MR 797,788 [allow regression]

See merge request !801
parents 6ce1f51f 95c647d8

lib_com/options.h

+2 −0
Original line number Diff line number Diff line
@@ -196,6 +196,8 @@ 
#define FIX_953_WRONG_ENERGY_RATIO_MASA_EXT     /* Nok: Fix 953 wrong energy ratio value after shift and cast to Word8 */

#define FIX_982_WRONG_DECODED_ENERGY_RATIO      /* Nokia: Fix 982 wrong energy in EXT mode and in second direction when present */

#define FIX_999_WRONG_ISM_EXTENDED_METADATA     /* VA: fix 999: fix ISM extended metadata decoding */

#define NONBE_FIX_1205_TD_STEREO_MOD_CT         /* VA: fix mismatch of coder_type (mod_ct) btw. TD stereo encoder and decoder */

#define NONBE_FIX_1204_MDCT_STEREO_NOISE_EST_SCALING    /* FhG: fixes for decoder-side noise level estimation in MDCT-Stereo to prevent noise bursts in stereo switching */

/* ################## End DEVELOPMENT switches ######################### */



/* clang-format on */

lib_dec/ivas_stereo_mdct_core_dec_fx.c

+39 −6
Original line number Diff line number Diff line
@@ -988,7 +988,7 @@ static void run_min_stats_fx( 
{

    Word16 ch, will_estimate_noise_on_channel[CPE_CHANNELS], save_VAD[CPE_CHANNELS];

    Word32 power_spec[L_FRAME16k];

    Word16 power_spec_16[L_FRAME16k], power_spec_e = 0;

    Word16 power_spec_e = 0;

    move16();

    Word32 *spec_in;

    Word16 spec_e;
@@ -1044,29 +1044,62 @@ static void run_min_stats_fx( 
            IF( ( EQ_16( will_estimate_noise_on_channel[0], will_estimate_noise_on_channel[1] ) ) || EQ_16( ch, 0 ) )

            {

                Word16 tmp16 = getScaleFactor32( spec_in, L_FRAME16k );



#ifdef NONBE_FIX_1204_MDCT_STEREO_NOISE_EST_SCALING

                Word32 power_spec_scale_fac;



                /* calculate power spectrum from MDCT coefficients and estimated MDST coeffs */

                power_spec_scale_fac = 20792; // 1.f / ( L_FRAME16k * L_FRAME16k ) in Q31

                move32();

                power_spec[0] = Mpy_32_32( W_extract_h( W_shl( W_mult_32_32( spec_in[0], spec_in[0] ), sub( tmp16, 4 ) ) ), power_spec_scale_fac ); /* 2 * (Q31 - x_e) + tmp16 - Q4 - Q31 */

                move32();

                power_spec[L_FRAME16k - 1] = Mpy_32_32( W_extract_h( W_shl( W_mult_32_32( spec_in[L_FRAME16k - 1], spec_in[L_FRAME16k - 1] ), sub( tmp16, 4 ) ) ), power_spec_scale_fac ); /* 2 * (Q31 - x_e) + tmp16 - Q4 - Q31 */

                move32();

#else

                /* calculate power spectrum from MDCT coefficients and estimated MDST coeffs */

                power_spec[0] = W_extract_h( W_shl( W_mult_32_32( spec_in[0], spec_in[0] ), sub( tmp16, 4 ) ) );                                                                                   /* 2 * (Q31 - x_e) + tmp16 - Q4 - Q31 */

                power_spec[L_FRAME16k - 1] = W_extract_h( W_shl( W_mult_32_32( spec_in[L_FRAME16k - 1], spec_in[L_FRAME16k - 1] ), sub( tmp16, 4 ) ) );                                            /* 2 * (Q31 - x_e) + tmp16 - Q4 - Q31 */

#endif

                FOR( Word16 i = 1; i < L_FRAME16k - 1; i++ )

                {

                    Word32 mdst;

                    mdst = L_sub( spec_in[i + 1], spec_in[i - 1] ); /* Q31 - x_e */



#ifdef NONBE_FIX_1204_MDCT_STEREO_NOISE_EST_SCALING

                    power_spec[i] = Mpy_32_32( L_add( W_extract_h( W_shl( W_mult_32_32( spec_in[i], spec_in[i] ), sub( tmp16, 4 ) ) ), W_extract_h( W_shl( W_mult_32_32( mdst, mdst ), sub( tmp16, 4 ) ) ) ), power_spec_scale_fac ); /* 2 * (Q31 - x_e) + tmp16 - Q4 - Q31*/

#else

                    power_spec[i] = L_add( W_extract_h( W_shl( W_mult_32_32( spec_in[i], spec_in[i] ), sub( tmp16, 4 ) ) ), W_extract_h( W_shl( W_mult_32_32( mdst, mdst ), sub( tmp16, 4 ) ) ) ); /* 2 * (Q31 - x_e) + tmp16 - Q4 - Q31*/

#endif

                    move32();

                }

                power_spec_e = sub( add( 4, shl( spec_e, 1 ) ), tmp16 );

            }



            Copy_Scale_sig32_16( power_spec, power_spec_16, L_FRAME16k, 0 ); /* exp(power_spec_e) */



#ifndef NONBE_FIX_1204_MDCT_STEREO_NOISE_EST_SCALING

            noisy_speech_detection_fx( st->hFdCngDec, st->VAD && st->m_frame_type == ACTIVE_FRAME, power_spec_16, sub( 15, power_spec_e ) );



            st->hFdCngDec->hFdCngCom->likelihood_noisy_speech = add( mult_r( 32440 /* 0.99 in Q15 */, st->hFdCngDec->hFdCngCom->likelihood_noisy_speech ), mult_r( st->hFdCngDec->hFdCngCom->flag_noisy_speech, 328 /* 0.01 in Q15 */ ) ); /* Q15 */

            move16();

            st->lp_noise = st->hFdCngDec->lp_noise; /* Q9.23 */

            move32();

#endif

        }



#ifdef NONBE_FIX_1204_MDCT_STEREO_NOISE_EST_SCALING

        IF( EQ_16( st->core, TCX_20_CORE ) )

        {

            Word16 x_fx_16[L_FRAME16k];

            Copy_Scale_sig32_16( x[ch][0], x_fx_16, L_FRAME16k, 0 ); /* exp(x_e) */



            test();

            noisy_speech_detection_fx( st->hFdCngDec, save_VAD[ch] && EQ_16( st->m_frame_type, ACTIVE_FRAME ), x_fx_16, sub( Q15, x_e[ch][0] ) );

            st->hFdCngDec->hFdCngCom->likelihood_noisy_speech = add( mult_r( 32440 /* 0.99 in Q15 */, st->hFdCngDec->hFdCngCom->likelihood_noisy_speech ), mult_r( st->hFdCngDec->hFdCngCom->flag_noisy_speech, 328 /* 0.01 in Q15 */ ) ); /* Q15 */

            move16();

            st->lp_noise = st->hFdCngDec->lp_noise; /* Q9.23 */

            move32();

        }

#endif



        test();

        test();

        IF( will_estimate_noise_on_channel[0] || will_estimate_noise_on_channel[1] || st->bfi )

lib_enc/enc_uv_fx.c

+3 −2
Original line number Diff line number Diff line
@@ -282,6 +282,7 @@ void encod_unvoiced_ivas_fx( 
{

    Word16 xn_fx[L_SUBFR];   /* Target vector for pitch search     */

    Word16 h1_fx[L_SUBFR];   /* Impulse response vector            */

    Word16 h2_fx[L_SUBFR];   /* Impulse response vector            */

    Word16 code_fx[L_SUBFR]; /* Fixed codebook excitation          */

    Word16 y2_fx[L_SUBFR];   /* Filtered algebraic excitation      */

    Word16 *pt_pitch_fx;     /* pointer to floating pitch buffer   */
@@ -348,7 +349,7 @@ void encod_unvoiced_ivas_fx( 
        find_targets_fx( speech_fx, hLPDmem->mem_syn, i_subfr, &hLPDmem->mem_w0, p_Aq_fx,

                         res_fx, L_SUBFR, p_Aw_fx, st_fx->preemph_fac, xn_fx, cn_fx, h1_fx );



        /*Copy_Scale_sig(h1_fx, h2_fx, L_SUBFR, -2);*/

        Copy_Scale_sig( h1_fx, h2_fx, L_SUBFR, -2 );

        Scale_sig( h1_fx, L_SUBFR, add( 1, shift ) ); /* set h1[] in Q14 with scaling for convolution */



        /* scaling of xn[] to limit dynamic at 12 bits */
@@ -390,7 +391,7 @@ void encod_unvoiced_ivas_fx( 
            // E_ACELP_innovative_codebook_fx( exc_fx, *pt_pitch_fx, 0, 1, gain_pit_fx, hLPDmem->tilt_code, acelp_cfg, i_subfr, p_Aq_fx, h1_fx, xn_fx, cn_fx, y1, y2_fx, (Word8) st_fx->acelp_autocorr, &prm, code_fx, shift, st_fx->L_frame, st_fx->last_L_frame, st_fx->total_brate, st_fx->element_mode );

            inov_encode_ivas_fx( st_fx, st_fx->core_brate, 0, L_FRAME, st_fx->last_L_frame,

                                 UNVOICED, st_fx->bwidth, st_fx->sharpFlag, i_subfr, -1, p_Aq_fx,

                                 gain_pit_fx, cn_fx, exc_fx, h1_fx, hLPDmem->tilt_code, *pt_pitch_fx, xn_fx, code_fx, y2_fx, &unbits_PI, L_SUBFR, shift, Q_new );

                                 gain_pit_fx, cn_fx, exc_fx, h2_fx, hLPDmem->tilt_code, *pt_pitch_fx, xn_fx, code_fx, y2_fx, &unbits_PI, L_SUBFR, shift, Q_new );



            E_ACELP_xy2_corr( xn_fx, y1, y2_fx, &g_corr, L_SUBFR, Q_xn );

lib_enc/ivas_cpe_enc.c

+7 −4
Original line number Diff line number Diff line
@@ -678,8 +678,11 @@ ivas_error ivas_cpe_enc_fx( 
#ifdef IVAS_FLOAT_FIXED

#ifdef IVAS_FLOAT_FIXED_CONVERSIONS

        Word16 input_fx[2][L_FRAME48k];

        floatToFixed_arr16( sts[1]->input, input_fx[0], 0, input_frame );

        floatToFixed_arr16( sts[0]->input, input_fx[1], 0, input_frame );

        Word16 tmpppp;

        tmpppp = s_min( Q_factor_arr( sts[1]->input, input_frame ), Q_factor_arr( sts[0]->input, input_frame ) );



        floatToFixed_arr16( sts[1]->input, input_fx[1], tmpppp, input_frame );

        floatToFixed_arr16( sts[0]->input, input_fx[0], tmpppp, input_frame );

#endif

        Word16 tdm_SM_flag;

        IF( hCPE->hStereoTD->tdm_LRTD_flag == 0 )
@@ -695,8 +698,8 @@ ivas_error ivas_cpe_enc_fx( 
        stereo_tdm_downmix_ivas_fx( hCPE->hStereoTD, input_fx[0], input_fx[1], input_frame, tdm_ratio_idx, tdm_SM_flag, tdm_ratio_idx_SM );



#ifdef IVAS_FLOAT_FIXED_CONVERSIONS

        fixedToFloat_arr( input_fx[0], sts[0]->input, 0, input_frame );

        fixedToFloat_arr( input_fx[1], sts[1]->input, 0, input_frame );

        fixedToFloat_arr( input_fx[0], sts[0]->input, tmpppp, input_frame );

        fixedToFloat_arr( input_fx[1], sts[1]->input, tmpppp, input_frame );

#endif

#else

        stereo_tdm_downmix( hCPE->hStereoTD, sts[0]->input, sts[1]->input, input_frame, tdm_ratio_idx, ( ( hCPE->hStereoTD->tdm_LRTD_flag == 0 ) ? tdm_SM_or_LRTD_Pri : 0 ), tdm_ratio_idx_SM );

lib_enc/ivas_decision_matrix_enc.c

+26 −1
Original line number Diff line number Diff line
@@ -498,10 +498,24 @@ void ivas_decision_matrix_enc_fx( 
    IF( EQ_16( st->core, TCX_20_CORE ) && LT_32( st->total_brate, STEREO_TCX_MIN_RATE ) )

    {

        st->core = ACELP_CORE;



#ifdef NONBE_FIX_1205_TD_STEREO_MOD_CT

        test();

        test();

        test();

        /* In TD stereo below 24.4 kbps we cannot overwrite the `coder_type` when it is set to TRANSITION, */

        /* as it is used for TD stereo bit allocation. To ensure consistent bit allocation, it must remain unchanged on the decoder side. */

        if ( st->idchan == 0 && !( LT_32( element_brate, IVAS_24k4 ) && EQ_16( st->coder_type, TRANSITION ) && EQ_16( st->element_mode, IVAS_CPE_TD ) ) )

        {

            st->coder_type = AUDIO;

            move16();

        }

#else

        st->coder_type = AUDIO;

        move16();

#endif

        st->sp_aud_decision2 = 0;



        move16();

        move16();

        move16();
@@ -844,6 +858,16 @@ void ivas_signaling_enc_fx( 


                IF( EQ_16( st->core, ACELP_CORE ) )

                {

#ifdef NONBE_FIX_1205_TD_STEREO_MOD_CT

                    /* write coder type */

                    push_indice( hBstr, IND_ACELP_SIGNALLING, st->coder_type, 3 );



                    IF( GE_32( element_brate, FRMT_SHP_MIN_BRATE_IVAS ) )

                    {

                        /* write sharpening flag */

                        push_indice( hBstr, IND_SHARP_FLAG, st->sharpFlag, 1 );

                    }

#else

                    IF( LT_32( element_brate, FRMT_SHP_MIN_BRATE_IVAS ) )

                    {

                        push_indice( hBstr, IND_ACELP_SIGNALLING, st->coder_type, 3 );
@@ -856,6 +880,7 @@ void ivas_signaling_enc_fx( 
                        /* write sharpening flag */

                        push_indice( hBstr, IND_SHARP_FLAG, st->sharpFlag, 1 );

                    }

#endif



                    /* write extension layer flag to distinguish between TBE (0) and BWE (1) */

                    IF( st->extl_brate > 0 )