Commit 1e268e41 authored by vaclav's avatar vaclav
Browse files

Merge branch 'basop-2404-harmonization-of-signal-classification' into 'main' 

harmonisation of signal_clas

See merge request !2785
parents 785e2f24 e104a5ec

lib_com/options.h

+1 −0
Original line number Diff line number Diff line
@@ -90,6 +90,7 @@ 
#define FIX_2280_REDUCTION_UNNECESSARY_SCALING          /* VA: reduction of unnecessary scaling */

#define FIX_2280_REDUCTION_UNNECESSARY_SCALING_NONBE    /* VA: reduction of unnecessary scaling, non-BE part */

#define FIX_2403_COMBINE_PITCH_OL                       /* VA : basop 2403, reusing common code between EVS and IVAS in pitch_ol */

#define FIX_2404_HARM_SIGNAL_CLAS                       /* VA: basop-2404, harmonize signal_clas and signa_ivas_clas */

#define HARMONIZE_ACELP_ENC                             /* VA: basop issue 2400: Remove duplicated main ACELP encoder function */



/* #################### End BE switches ################################## */

lib_enc/init_enc_fx.c

+1 −0
Original line number Diff line number Diff line
@@ -557,6 +557,7 @@ ivas_error init_encoder_fx( 
#else

        speech_music_clas_init_fx( st->element_mode, st->hSpMusClas );

#endif



        st->sp_aud_decision0 = 0;

        move16();

        st->sp_aud_decision1 = 0;

lib_enc/ivas_core_pre_proc_front_fx.c

+4 −0
Original line number Diff line number Diff line
@@ -1254,7 +1254,11 @@ void pre_proc_front_ivas_fx( 
     * TC frame selection

     *-----------------------------------------------------------------*/



#ifndef FIX_2404_HARM_SIGNAL_CLAS

    st->clas = signal_clas_ivas_fx( st, inp_12k8_fx, ee_fx, *relE_fx, L_look, tdm_SM_last_clas ); /* Q0 */

#else

    st->clas = signal_clas_fx( st, inp_12k8_fx, ee_fx, *relE_fx, L_look, tdm_SM_last_clas ); /* Q0 */

#endif

    move16();



    select_TC_fx( MODE1, st->tc_cnt, &st->coder_type, st->localVAD );

lib_enc/prot_fx_enc.h

+3 −2
Original line number Diff line number Diff line
@@ -511,7 +511,7 @@ Word16 signal_clas_fx( 
    const Word16 L_look,  /* i  : look-ahead                                        */

    Word16 *uc_clas       /* o  : temporary classification used in music/speech class*/

);



#ifndef FIX_2404_HARM_SIGNAL_CLAS

/* o  : classification for current frames  */

Word16 signal_clas_ivas_fx(

    Encoder_State *st,    /* i/o: encoder state structure                           */
@@ -521,6 +521,7 @@ Word16 signal_clas_ivas_fx( 
    const Word16 L_look,  /* i  : look-ahead                                        */

    Word16 *uc_clas       /* o  : temporary classification used in music/speech class*/

);

#endif



void speech_music_classif_fx(

    Encoder_State *st,            /* i/o: state structure                                   */
@@ -1127,7 +1128,7 @@ void coder_type_modif_fx( 


void speech_music_clas_init_fx(

#ifdef FIX_2405_HARM_SMC_INIT

    const Word16 element_mode, /* element mode to differentiate IVAS only init */

    const Word16 element_mode, /* i  : element mode                     */

#endif

    SP_MUS_CLAS_HANDLE hSpMusClas /* i/o: speech/music classifier handle   */

);

lib_enc/sig_clas_fx.c

+74 −39
Original line number Diff line number Diff line
@@ -31,9 +31,9 @@ 
#define K_SNR_FX 3541   /* Q15 .1111    */

#define C_SNR_FX -10921 /* Q15 -0.3333f */





#define THRES_EEN 514206 /* 251.077 =>  (10^(1/(K_EE*10)))  Q11*/





/*-------------------------------------------------------------------*

 * signal_clas_fx()

 *
@@ -41,7 +41,8 @@ 
 * TC frames selection

 *-------------------------------------------------------------------*/



Word16 signal_clas_fx(                       /* o  : classification for current frames              */

/* o  : classification for current frames */

Word16 signal_clas_fx(

    Encoder_State *st,    /* i/o: encoder state structure                           */

    const Word16 *speech, /* i  : pointer to speech signal for E computation in Qx */

    const Word32 *ee,     /* i  : lf/hf E ration for 2 half-frames in Q6            */
@@ -55,6 +56,9 @@ Word16 signal_clas_fx( /* o : classification for current 
    Word16 i, clas, pc, zc, lo, lo2, hi, hi2, exp_ee, frac_ee;

    Word16 tmp16, tmpS;

    const Word16 *pt1;

#ifdef FIX_2404_HARM_SIGNAL_CLAS

    Word64 tmp64;

#endif

    Flag Overflow = 0;

    move32();
@@ -69,6 +73,31 @@ Word16 signal_clas_fx( /* o : classification for current 
    Ltmp = L_mult( st->voicing_fx[1], 16384 ); /* Q15*Q14->Q30 */

    mean_voi2 = mac_r( Ltmp, st->voicing_fx[2], 16384 );



#ifdef FIX_2404_HARM_SIGNAL_CLAS

    IF( st->element_mode != EVS_MONO )

    {

        tmp64 = W_mult0_32_32( ee[0], ee[1] );

        exp_ee = W_norm( tmp64 );

        Ltmp = W_extract_h( W_shl( tmp64, exp_ee ) ); // Q = Q6+Q6 + exp_ee - 32

        exp_ee = sub( 31, sub( add( Q12, exp_ee ), 32 ) );

        IF( EQ_16( BASOP_Util_Cmp_Mant32Exp( Ltmp, exp_ee, ONE_IN_Q31, 0 ), -1 ) )

        {

            een = 0;

            move16();

        }

        ELSE

        {

            /* mean_ee2 = 0.5f * 20.0f * (float)log10( tmp ); */

            /* een = K_EE_ENC * mean_ee2 + C_EE_ENC; */

            Ltmp = BASOP_Util_Log10( Ltmp, exp_ee );                            // Q25

            Ltmp = Mpy_32_32( Ltmp, 671088640 /*20.f in Q25*/ );                // Q25 + Q25 -Q31 = Q19 * 0.5 = Q20

            een = extract_l( L_shl( Mpy_32_16_1( Ltmp, K_EE_FX ), Q9 - Q20 ) ); // Q9

            een = s_min( s_max( een, 0 ), 512 );

        }

    }

    ELSE

#endif

    {

        /* average spectral tilt in dB */

        lo = L_Extract_lc( ee[0], &hi );

        lo2 = L_Extract_lc( ee[1], &hi2 );
@@ -97,6 +126,8 @@ Word16 signal_clas_fx( /* o : classification for current 
            een = round_fx( L_shl( Ltmp, 16 - 5 ) );   /* Q14 -> Q9 */

            een = mac_r( C_EE_FX, een, K_EE_FX );

        }

    }



    /* compute zero crossing rate */

    pt1 = speech + sub( L_look, 1 );

    tmpS = shr( *pt1, 15 ); /* sets 'tmpS to -1 if *pt1 < 0 */
@@ -114,6 +145,7 @@ Word16 signal_clas_fx( /* o : classification for current 
    pc = add( abs_s( sub( st->pitch[1], st->pitch[0] ) ), abs_s( sub( st->pitch[2], st->pitch[1] ) ) );

    st->tdm_pc = pc;

    move16();



    /*-----------------------------------------------------------------*

     * Transform parameters to the range <0:1>

     * Compute the merit function
@@ -241,8 +273,8 @@ Word16 signal_clas_fx( /* o : classification for current 
                BREAK;

        }

    }

    /* Onset classification */



    /* Onset classification */

    /* tc_cnt == -1: frame after TC frame in continuous block of GC/VC frames */

    /* tc_cnt ==  0: UC frame */

    /* tc_cnt ==  1: onset/transition frame, coded by GC coder type */
@@ -266,9 +298,11 @@ Word16 signal_clas_fx( /* o : classification for current 
        st->tc_cnt = -1;

        move16();

    }



    return clas;

}



#ifndef FIX_2404_HARM_SIGNAL_CLAS

Word16 signal_clas_ivas_fx(                       /* o  : classification for current frames              */

                            Encoder_State *st,    /* i/o: encoder state structure                           */

                            const Word16 *speech, /* i  : pointer to speech signal for E computation in Qx */
@@ -498,6 +532,7 @@ Word16 signal_clas_ivas_fx( /* o : classification for cur 
    }

    return clas;

}

#endif



/*-------------------------------------------------------------------*

 * select_TC_fx()
@@ -519,6 +554,7 @@ void select_TC_fx( 
         * Select TC coder type for appropriate frames which is in general VOICED_TRANSITION,

         * VOICED_CLAS or ONSET frames following UNVOICED_CLAS frames

         *---------------------------------------------------------------------*/



        test();

        IF( localVAD != 0 && GE_16( tc_cnt, 1 ) )

        {
@@ -540,21 +576,21 @@ void select_TC_fx( 
    return;

}





/*-------------------------------------------------------------------*

 * coder_type_modif_fx()

 *

 * Coder type modification

 *-------------------------------------------------------------------*/



void coder_type_modif_fx(

    Encoder_State *st, /* i/o: encoder state structure                     */

    const Word16 relE  /* i  : frame relative E to the long term average   */

)

{

    Word16 unmod_coder_type, vbr_generic_ho;



    SC_VBR_ENC_HANDLE hSC_VBR = st->hSC_VBR;





    IF( st->Opt_SC_VBR )

    {

        vbr_generic_ho = hSC_VBR->vbr_generic_ho;
@@ -582,8 +618,8 @@ void coder_type_modif_fx( 
        test();

        test();

        test();

        if ( ( st->element_mode == 0 && GT_32( st->total_brate, ACELP_9k60 ) && EQ_16( st->coder_type, UNVOICED ) ) ||

             ( st->element_mode > 0 && GT_32( st->total_brate, MAX_UNVOICED_BRATE ) && EQ_16( st->coder_type, UNVOICED ) ) )

        if ( ( st->element_mode == EVS_MONO && GT_32( st->total_brate, ACELP_9k60 ) && EQ_16( st->coder_type, UNVOICED ) ) ||

             ( st->element_mode > EVS_MONO && GT_32( st->total_brate, MAX_UNVOICED_BRATE ) && EQ_16( st->coder_type, UNVOICED ) ) )

        {

            st->coder_type = GENERIC;

            move16();
@@ -609,8 +645,7 @@ void coder_type_modif_fx( 
        test();

        test();

        test();

        if ( st->localVAD == 0 && ( (

                                        EQ_16( st->coder_type, UNVOICED ) && ( ( st->Opt_SC_VBR == 0 ) || ( ( EQ_16( st->Opt_SC_VBR, 1 ) ) && vbr_generic_ho == 0 && GT_16( st->last_coder_type, UNVOICED ) ) ) ) ||

        if ( st->localVAD == 0 && ( ( EQ_16( st->coder_type, UNVOICED ) && ( ( st->Opt_SC_VBR == 0 ) || ( ( EQ_16( st->Opt_SC_VBR, 1 ) ) && vbr_generic_ho == 0 && GT_16( st->last_coder_type, UNVOICED ) ) ) ) ||

                                    EQ_16( st->coder_type, TRANSITION ) || EQ_16( st->coder_type, VOICED ) )



        )
@@ -657,7 +692,7 @@ void coder_type_modif_fx( 
            }

        }



        IF( st->element_mode == 0 )

        IF( st->element_mode == EVS_MONO )

        {

            /* At higher rates and with 16kHz core, allow only GC and TC coder type */

            test();