correct normalization of energy ratios at 384k MASA (73a5cdb7) · Commits · IVAS Codec Public Collaboration / IVAS Codec

lib_com/options.h

+1 −1

Original line number	Diff line number	Diff line
		@@ -180,7 +180,7 @@


		/* ##################### End NON-BE CR switches ########################### */

		#define NONBE_FIX_539_MASA_384K_CHIRP /* Nokia: issue 539, puts the normalization of the energy ratios at the correct place, affect MASA 384k only */
		/* clang-format on */

		#endif

lib_dec/ivas_qmetadata_dec.c

+57 −1

Original line number	Diff line number	Diff line
		@@ -908,6 +908,60 @@ int16_t ivas_qmetadata_dec_decode_hr_384_512(
		}
		if ( hQMetaData->no_directions == 2 )
		{
		#ifdef NONBE_FIX_539_MASA_384K_CHIRP
		float ratioSum;
		if ( bits_sph_idx == 16 )
		{
		for ( b = hQMetaData->q_direction[1].cfg.start_band; b < hQMetaData->q_direction[1].cfg.nbands; b++ )
		{
		for ( m = 0; m < hQMetaData->q_direction[1].cfg.nblocks; m++ )
		{
		hQMetaData->q_direction[1].band_data[b].energy_ratio[m] = 1.0f - diffuseness_reconstructions_hr[hQMetaData->q_direction[1].band_data[b].energy_ratio_index[m]];

		/* Scale energy ratios that sum to over one */
		ratioSum = hQMetaData->q_direction[0].band_data[b].energy_ratio[m] + hQMetaData->q_direction[1].band_data[b].energy_ratio[m];

		if ( ratioSum > 1.0f )
		{
		hQMetaData->q_direction[0].band_data[b].energy_ratio[m] /= ratioSum;
		hQMetaData->q_direction[1].band_data[b].energy_ratio[m] /= ratioSum;
		}
		}
		}
		}
		else
		{
		int16_t pos_2dir_band[MASA_MAXIMUM_CODING_SUBBANDS];
		d = 0;
		for ( b = hQMetaData->q_direction[0].cfg.start_band; b < hQMetaData->q_direction[0].cfg.nbands; b++ )
		{
		if ( hQMetaData->twoDirBands[b] == 1 )
		{
		pos_2dir_band[d] = b;
		d++;
		}
		else
		{
		pos_2dir_band[d] = 0;
		}
		}
		for ( b = hQMetaData->q_direction[1].cfg.start_band; b < hQMetaData->q_direction[1].cfg.nbands; b++ )
		{
		for ( m = 0; m < hQMetaData->q_direction[1].cfg.nblocks; m++ )
		{
		hQMetaData->q_direction[1].band_data[b].energy_ratio[m] = 1.0f - diffuseness_reconstructions_hr[hQMetaData->q_direction[1].band_data[b].energy_ratio_index[m]];

		ratioSum = hQMetaData->q_direction[0].band_data[pos_2dir_band[b]].energy_ratio[m] + hQMetaData->q_direction[1].band_data[b].energy_ratio[m];

		if ( ratioSum > 1.0f )
		{
		hQMetaData->q_direction[0].band_data[pos_2dir_band[b]].energy_ratio[m] /= ratioSum;
		hQMetaData->q_direction[1].band_data[b].energy_ratio[m] /= ratioSum;
		}
		}
		}
		}
		#else
		#ifdef FIX_566_2DIR_MASA_384K
		float ratioSum;
		for ( b = hQMetaData->q_direction[1].cfg.start_band; b < hQMetaData->q_direction[1].cfg.nbands; b++ )
		@@ -938,6 +992,7 @@ int16_t ivas_qmetadata_dec_decode_hr_384_512(
		}
		}
		}
		#endif
		#endif
		}

		@@ -1085,7 +1140,7 @@ int16_t ivas_qmetadata_dec_decode_hr_384_512(
		}
		}
		}

		#ifndef NONBE_FIX_539_MASA_384K_CHIRP
		/* Scale energy ratios that sum to over one */
		for ( b = 0; b < hQMetaData->q_direction[0].cfg.nbands; b++ )
		{
		@@ -1101,6 +1156,7 @@ int16_t ivas_qmetadata_dec_decode_hr_384_512(
		}
		}
		}
		#endif
		}

		#ifdef DEBUG_MODE_QMETADATA

lib_enc/ivas_qmetadata_enc.c

+65 −0

Original line number	Diff line number	Diff line
		@@ -1552,6 +1552,70 @@ static void ivas_qmetadata_quantize_diffuseness_nrg_ratios_hr_512(

		if ( hQMetaData->no_directions == 2 )
		{
		#ifdef NONBE_FIX_539_MASA_384K_CHIRP
		float ratioSum;
		if ( bits_dir_hr == 16 )
		{
		for ( j = hQMetaData->q_direction[1].cfg.start_band; j < hQMetaData->q_direction[1].cfg.nbands; j++ )
		{
		for ( k = 0; k < hQMetaData->q_direction[1].cfg.nblocks; k++ )
		{
		index = masa_sq( 1.0f - hQMetaData->q_direction[1].band_data[j].energy_ratio[k], diffuseness_thresholds_hr, HR_MASA_ER_LEVELS );
		push_next_indice( hMetaData, index, MASA_BITS_ER_HR );
		hQMetaData->q_direction[1].band_data[j].energy_ratio_index[k] = index;
		hQMetaData->q_direction[1].band_data[j].energy_ratio[k] = 1.0f - diffuseness_reconstructions_hr[index];

		ratioSum = hQMetaData->q_direction[0].band_data[j].energy_ratio[k] + hQMetaData->q_direction[1].band_data[j].energy_ratio[k];
		if ( ratioSum > 1.0f )
		{
		hQMetaData->q_direction[0].band_data[j].energy_ratio[k] /= ratioSum;
		hQMetaData->q_direction[1].band_data[j].energy_ratio[k] /= ratioSum;
		}

		needed_bits[1] += MASA_BITS_ER_HR;
		hQMetaData->q_direction[1].band_data[j].bits_sph_idx[k] = bits_dir_hr;
		}
		}
		}
		else
		{
		int16_t pos_2dir_band[MASA_MAXIMUM_CODING_SUBBANDS];
		k = 0;
		for ( j = hQMetaData->q_direction[0].cfg.start_band; j < hQMetaData->q_direction[0].cfg.nbands; j++ )
		{
		if ( hQMetaData->twoDirBands[j] == 1 )
		{
		pos_2dir_band[k] = j;
		k++;
		}
		else
		{
		pos_2dir_band[k] = 0;
		}
		}
		for ( j = hQMetaData->q_direction[1].cfg.start_band; j < hQMetaData->q_direction[1].cfg.nbands; j++ )
		{
		for ( k = 0; k < hQMetaData->q_direction[1].cfg.nblocks; k++ )
		{
		index = masa_sq( 1.0f - hQMetaData->q_direction[1].band_data[j].energy_ratio[k], diffuseness_thresholds_hr, HR_MASA_ER_LEVELS );
		push_next_indice( hMetaData, index, MASA_BITS_ER_HR );
		hQMetaData->q_direction[1].band_data[j].energy_ratio_index[k] = index;
		hQMetaData->q_direction[1].band_data[j].energy_ratio[k] = 1.0f - diffuseness_reconstructions_hr[index];

		ratioSum = hQMetaData->q_direction[0].band_data[pos_2dir_band[j]].energy_ratio[k] + hQMetaData->q_direction[1].band_data[j].energy_ratio[k];

		if ( ratioSum > 1.0f )
		{
		hQMetaData->q_direction[0].band_data[pos_2dir_band[j]].energy_ratio[k] /= ratioSum;
		hQMetaData->q_direction[1].band_data[j].energy_ratio[k] /= ratioSum;
		}

		needed_bits[1] += MASA_BITS_ER_HR;
		hQMetaData->q_direction[1].band_data[j].bits_sph_idx[k] = bits_dir_hr;
		}
		}
		}
		#else
		#ifdef FIX_566_2DIR_MASA_384K
		float ratioSum;
		#endif
		@@ -1580,6 +1644,7 @@ static void ivas_qmetadata_quantize_diffuseness_nrg_ratios_hr_512(
		hQMetaData->q_direction[1].band_data[j].bits_sph_idx[k] = bits_dir_hr;
		}
		}
		#endif
		}

		return;