Fixes for crashes in +10dB and -10dB scaled inputs (b9e7c3a6) · Commits · SA4 / Audio / IVAS BASOP

lib_com/prot_fx.h

+2 −0

Original line number	Diff line number	Diff line
		@@ -10535,6 +10535,8 @@ void swb_pre_proc_ivas_fx(
		Word16 new_swb_speech, / o : original input signal at 32kHz - Q0 */
		Word32 new_swb_speech_fx, / o : original input signal at 32kHz - Q - q_reImBuffer */
		Word16 shb_speech, / o : SHB target signal (6-14kHz) at 16kHz- Q0 */
		Word32 shb_speech_32, / o : SHB target signal (6-14kHz) at 16kHz- Q - q_reImBuffer */
		Word16 use_shb32, / o : flags valid data in 32-bit shb buffer for better precision. */
		Word32 realBuffer[CLDFB_NO_COL_MAX][CLDFB_NO_CHANNELS_MAX], /* i : real buffer Q - q_reImbuffer */
		Word32 imagBuffer[CLDFB_NO_COL_MAX][CLDFB_NO_CHANNELS_MAX], /* i : imag buffer Q - q_reImbuffer */
		Word16 q_reImBuffer, /* i : scale data of real and imag CLDFB buffers */

lib_enc/ivas_core_enc.c

+15 −6

Original line number	Diff line number	Diff line
		@@ -91,7 +91,7 @@ ivas_error ivas_core_enc(
		STEREO_TD_ENC_DATA_HANDLE hStereoTD;
		float *inp[CPE_CHANNELS];
		float new_inp_resamp16k[CPE_CHANNELS][L_FRAME16k]; /* new input signal @16kHz, non pre-emphasised, used by the WB TBE/BWE */
		float old_syn_12k8_16k[CPE_CHANNELS][L_FRAME16k]; /* ACELP core synthesis at 12.8kHz or 16kHz to be used by the SWB BWE */
		float old_syn_12k8_16k[CPE_CHANNELS][L_FRAME16k] = { { 0 } }; /* ACELP core synthesis at 12.8kHz or 16kHz to be used by the SWB BWE */
		float shb_speech[L_FRAME16k];
		float hb_speech[L_FRAME16k / 4];
		float *new_swb_speech;
		@@ -106,6 +106,7 @@ ivas_error ivas_core_enc(
		#ifdef IVAS_FLOAT_FIXED
		Word32 *new_swb_speech_fx;
		Word16 shb_speech_fx[L_FRAME16k]; // Q_shb_spch
		Word32 shb_speech_fx_32[L_FRAME16k]; // Q_shb_spch
		Word16 Q_shb_spch;
		Word32 new_swb_speech_buffer_fx[L_FRAME48k + STEREO_DFT_OVL_MAX];
		Word16 new_inp_resamp16k_fx[CPE_CHANNELS][L_FRAME16k];
		@@ -847,6 +848,7 @@ ivas_error ivas_core_enc(
		Word32 realBuffer_fx[CLDFB_NO_COL_MAX][CLDFB_NO_CHANNELS_MAX];
		Word32 imagBuffer_fx[CLDFB_NO_COL_MAX][CLDFB_NO_CHANNELS_MAX];
		Word16 new_swb_speech_16_fx[L_FRAME48k] = { 0 };
		Word16 use_shb32 = 0;
		Word16 q_realImagBuffer = Q_factor_arrL( (float ) &realBuffer[n][0][0], CLDFB_NO_COL_MAX CLDFB_NO_CHANNELS_MAX );
		q_realImagBuffer = s_min( q_realImagBuffer, Q_factor_arrL( (float ) &imagBuffer[n][0][0], CLDFB_NO_COL_MAX CLDFB_NO_CHANNELS_MAX ) );
		floatToFixed_arrL32( (float ) &realBuffer[n][0][0], (Word32 ) &realBuffer_fx[0][0], q_realImagBuffer, CLDFB_NO_COL_MAX * CLDFB_NO_CHANNELS_MAX );
		@@ -858,7 +860,7 @@ ivas_error ivas_core_enc(
		#else
		for ( Word32 idx = 0; idx < ( L_FRAME48k + L_FRAME48k + NS2SA( 48000, DELAY_FIR_RESAMPL_NS ) ); idx++ )
		{
		st->hSignalBuf->input_buff[idx] = (Word16) st->hSignalBuf->input_buff_flt[0];
		st->hSignalBuf->input_buff[idx] = (Word16) st->hSignalBuf->input_buff_flt[idx];
		}
		#endif
		if ( st->element_mode == IVAS_CPE_DFT )
		@@ -895,7 +897,7 @@ ivas_error ivas_core_enc(
		}
		#endif
		#endif
		swb_pre_proc_ivas_fx( st, new_swb_speech_16_fx, new_swb_speech_fx, shb_speech_fx, realBuffer_fx, imagBuffer_fx, q_realImagBuffer, hCPE );
		swb_pre_proc_ivas_fx( st, new_swb_speech_16_fx, new_swb_speech_fx, shb_speech_fx, shb_speech_fx_32, &use_shb32, realBuffer_fx, imagBuffer_fx, q_realImagBuffer, hCPE );
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		fixedToFloat_arr( (Word16 ) &st->hBWE_TD->old_speech_shb_fx[0], (float ) &st->hBWE_TD->old_speech_shb[0], 0, L_LOOK_16k + L_SUBFR16k );
		fixedToFloat_arr( (Word16 ) &st->hBWE_FD->old_fdbwe_speech_fx[0], (float ) &st->hBWE_FD->old_fdbwe_speech[0], 0, L_FRAME48k );
		@@ -906,7 +908,14 @@ ivas_error ivas_core_enc(
		}
		fixedToFloat_arrL32( (Word32 ) &st->cldfbSynTd->cldfb_state_fx[0], (float ) &st->cldfbSynTd->cldfb_state[0], q_realImagBuffer, st->cldfbSynTd->p_filter_length );
		fixedToFloat_arr( new_swb_speech_16_fx, new_swb_speech, 0, L_FRAME48k );
		if ( use_shb32 == 0 )
		{
		fixedToFloat_arr( shb_speech_fx, shb_speech, 0, L_FRAME16k );
		}
		else
		{
		fixedToFloat_arrL( shb_speech_fx_32, shb_speech, q_realImagBuffer - 1, L_FRAME16k );
		}
		fixedToFloat_arrL32( (Word32 ) &realBuffer_fx[0][0], (float ) &realBuffer[n][0][0], q_realImagBuffer, CLDFB_NO_COL_MAX * CLDFB_NO_CHANNELS_MAX );
		fixedToFloat_arrL32( (Word32 ) &imagBuffer_fx[0][0], (float ) &imagBuffer[n][0][0], q_realImagBuffer, CLDFB_NO_COL_MAX * CLDFB_NO_CHANNELS_MAX );
		st->hBWE_TD->cldfbHBLT_flt = fix16_to_float( st->hBWE_TD->cldfbHBLT, Q13 );

lib_enc/swb_pre_proc.c

+13 −1

Original line number	Diff line number	Diff line
		@@ -282,6 +282,8 @@ void swb_pre_proc_ivas_fx(
		Word16 new_swb_speech, / o : original input signal at 32kHz - Q0 */
		Word32 new_swb_speech_fx, / o : original input signal at 32kHz - Q - q_reImBuffer */
		Word16 shb_speech, / o : SHB target signal (6-14kHz) at 16kHz- Q0 */
		Word32 shb_speech_32, / o : SHB target signal (6-14kHz) at 16kHz- Q - q_reImBuffer */
		Word16 use_shb32, / o : flags valid data in 32-bit shb buffer for better precision. */
		Word32 realBuffer[CLDFB_NO_COL_MAX][CLDFB_NO_CHANNELS_MAX], /* i : real buffer Q - q_reImbuffer */
		Word32 imagBuffer[CLDFB_NO_COL_MAX][CLDFB_NO_CHANNELS_MAX], /* i : imag buffer Q - q_reImbuffer */
		Word16 q_reImBuffer, /* i : scale data of real and imag CLDFB buffers */
		@@ -308,9 +310,11 @@ void swb_pre_proc_ivas_fx(
		Word32 lbEner;
		Word16 exp_lbEner = 0;
		Word16 Sample_Delay_SWB_BWE32k, lMemRecalc32k, dft_ovl32k;
		Word32 shb_speech_32[L_FRAME16k];
		Word32 one_by_50_Q31 = 42949673;

		*use_shb32 = 0;
		move16();

		lMemRecalc32k = NS2SA_FX2( 32000, L_MEM_RECALC_NS );
		move16(); // exp_lbEner
		move16(); // lMemRecalc32k
		@@ -783,6 +787,8 @@ void swb_pre_proc_ivas_fx(

		// Copy_Scale_sig_16_32((const Word16 )&shb_speech[0], (Word32 )&shb_speech_32[0], L_FRAME16k,Q16);
		cldfbSynthesis_ivas_fx( realBufferFlipped, imagBufferFlipped, shb_speech_32, -1, st->cldfbSynTd );
		*use_shb32 = 1;
		move16();
		Copy_Scale_sig_32_16( (const Word32 ) &shb_speech_32[0], (Word16 ) &shb_speech[0], L_FRAME16k, negate( sub( q_reImBuffer, 1 ) ) );
		}

		@@ -888,6 +894,8 @@ void swb_pre_proc_ivas_fx(
		IF( ( EQ_16( st->bwidth, FB ) \|\| st->core == ACELP_CORE ) && ( st->element_mode == EVS_MONO ) )
		{
		InitSWBencBufferStates_fx( st->hBWE_TD, shb_speech );
		*use_shb32 = 0;
		move16();
		}
		ELSE
		{
		@@ -932,6 +940,8 @@ void swb_pre_proc_ivas_fx(
		}

		Decimate_allpass_steep_fx( spchTmp, hBWE_TD->state_ana_filt_shb_fx, L_FRAME32k, shb_speech );
		*use_shb32 = 0;
		move16();

		Copy( shb_speech + L_FRAME16k - ( L_LOOK_16k + L_SUBFR16k ), hBWE_TD->old_speech_shb_fx, L_LOOK_16k + L_SUBFR16k );

		@@ -971,6 +981,8 @@ void swb_pre_proc_ivas_fx(
		shb_speech[i] = mult_r( mult_r( i, 983 /0.03f Q15/ ), shb_speech[2 * delay - 1 - i] );
		move16();
		}
		*use_shb32 = 0;
		move16();
		}

		return;