Merge branch '571-reverb-not-activated-for-discrete-and-parametric-ism' of... (3fec8ace) · Commits · IVAS Codec Public Collaboration / IVAS Codec

lib_com/options.h

100755 → 100644

+2 −0

Original line number	Diff line number	Diff line
		@@ -154,6 +154,8 @@
		#define FIX_559_EXTL_IGF_MISMATCH /* VA: issue 559: fix mismatch between st->extl and st->igf observed as crash in PlanarSBA bitrate switching */
		#define FIX_571_REVERB_NOT_ACTIVATED_ISM /* Philips: Issue 571: Reverb not activated for discrete and parametric ISM */

		#define FIX_QMETA_SID_5k2 /* Nokia: Issue 137: enable using full 5.2k bitrate in MASA SID */

		/* ################## End DEVELOPMENT switches ######################### */
		/* clang-format on */

lib_dec/ivas_qmetadata_dec.c

+86 −8

Original line number	Diff line number	Diff line
		@@ -1181,6 +1181,9 @@ int16_t ivas_qmetadata_dec_sid_decode(
		float avg_direction_vector[3];
		float direction_vector[3];
		int16_t metadata_sid_bits; /* bits allocated to SID for metadata */
		#ifdef FIX_QMETA_SID_5k2
		int16_t bits_delta, bits_dir;
		#endif
		#ifdef DEBUG_MODE_QMETADATA
		static FILE *pF = NULL;
		static FILE *pF_azi = NULL;
		@@ -1204,8 +1207,12 @@ int16_t ivas_qmetadata_dec_sid_decode(
		}
		else
		{
		#ifdef FIX_QMETA_SID_5k2
		metadata_sid_bits = ( IVAS_SID_5k2 - SID_2k40 ) / FRAMES_PER_SEC - SID_FORMAT_NBITS;
		#else
		/* TODO: still use old sid frame size to keep bitexactness */
		metadata_sid_bits = ( 4400 /IVAS_SID_5k2/ - SID_2k40 ) / FRAMES_PER_SEC - SID_FORMAT_NBITS;
		#endif
		}

		start_index = *index;
		@@ -1247,16 +1254,80 @@ int16_t ivas_qmetadata_dec_sid_decode(
		{
		q_direction->not_in_2D = 1;
		}

		#ifdef FIX_QMETA_SID_5k2
		bits_dir = 0;
		if ( ivas_format != SBA_FORMAT )
		{
		/* Decode diffuseness*/
		for ( b = start_band; b < nbands; b++ )
		{
		diffuseness_index[b] = ivas_qmetadata_DecodeQuasiUniform( bitstream, index, DIRAC_DIFFUSE_LEVELS - 4 ) + 4;
		q_direction->band_data[b].energy_ratio_index[0] = diffuseness_index[b];
		q_direction->band_data[b].bits_sph_idx[0] = bits_direction_masa[diffuseness_index[b]];
		bits_dir += q_direction->band_data[b].bits_sph_idx[0];
		q_direction->band_data[b].azimuth_m_alphabet[0] = no_phi_masa[q_direction->band_data[b].bits_sph_idx[0] - 1][0];
		}

		bits_delta = metadata_sid_bits - ( start_index - index ) - bits_dir; / bit_diff is already read */

		if ( bits_delta > 0 )
		{
		while ( bits_delta > 0 )
		{
		for ( b = start_band; b < nbands && ( bits_delta > 0 ); b++ )
		{
		if ( q_direction->band_data[b].bits_sph_idx[0] < 11 )
		{
		bits_delta -= 1;
		q_direction->band_data[b].bits_sph_idx[0]++;
		}
		}
		}
		if ( q_direction->not_in_2D == 0 )
		{
		for ( b = start_band; b < nbands; b++ )
		{
		q_direction->band_data[b].azimuth_m_alphabet[0] = 1 << ( min( 8, q_direction->band_data[b].bits_sph_idx[0] ) );
		}
		}
		}
		else
		{
		while ( bits_delta < 0 )
		{
		for ( b = nbands - 1; b >= start_band && ( bits_delta < 0 ); b-- )
		{
		if ( q_direction->band_data[b].bits_sph_idx[0] >= 4 )
		{
		bits_delta += 1;
		q_direction->band_data[b].bits_sph_idx[0]--;
		}
		}
		if ( q_direction->not_in_2D == 0 )
		{
		for ( b = start_band; b < nbands; b++ )
		{
		q_direction->band_data[b].azimuth_m_alphabet[0] = 1 << ( min( 8, q_direction->band_data[b].bits_sph_idx[0] ) );
		}
		}
		}
		}
		}
		else
		{
		#endif

		/* Decode diffuseness*/
		for ( b = start_band; b < nbands; b++ )
		{
		diffuseness_index[b] = ivas_qmetadata_DecodeQuasiUniform( bitstream, index, DIRAC_DIFFUSE_LEVELS - 4 ) + 4;
		q_direction->band_data[b].energy_ratio_index[0] = diffuseness_index[b];
		q_direction->band_data[b].bits_sph_idx[0] = bits_direction_masa[diffuseness_index[b]];
		q_direction->band_data[b].azimuth_m_alphabet[0] = no_phi_masa[q_direction->band_data[b].bits_sph_idx[0] - 1][0];
		}
		#ifdef FIX_QMETA_SID_5k2
		}
		#endif
		for ( b = start_band; b < nbands; b++ )
		{
		q_direction->band_data[b].energy_ratio[0] = 1.0f - diffuseness_reconstructions[diffuseness_index[b]];
		@@ -1299,7 +1370,14 @@ int16_t ivas_qmetadata_dec_sid_decode(
		{
		for ( b = start_band; b < nbands; b++ )
		{
		#ifdef FIX_QMETA_SID_5k2
		if ( ivas_format == SBA_FORMAT )
		{
		#endif
		q_direction->band_data[b].azimuth_m_alphabet[0] = 1 << ( min( 5, q_direction->band_data[b].bits_sph_idx[0] ) );
		#ifdef FIX_QMETA_SID_5k2
		}
		#endif
		q_direction->band_data[b].azimuth_index[0] = ivas_qmetadata_DecodeQuasiUniform( bitstream, index, q_direction->band_data[b].azimuth_m_alphabet[0] );
		q_direction->band_data[b].azimuth_index[0] = ivas_qmetadata_dereorder_generic( q_direction->band_data[b].azimuth_index[0] ) + ( q_direction->band_data[b].azimuth_m_alphabet[0] >> 1 );
		avg_azimuth = 360.0f / (float) ( q_direction->band_data[b].azimuth_m_alphabet[0] ) * q_direction->band_data[b].azimuth_index[0] - 180;

lib_enc/ivas_qmetadata_enc.c

+94 −18

Original line number	Diff line number	Diff line
		@@ -833,6 +833,10 @@ ivas_error ivas_qmetadata_enc_encode_hr_384_512(
		}
		#endif

		#ifdef FIX_QMETA_SID_5k2
		q_direction->not_in_2D = 0;
		#endif

		/Coherence /
		if ( all_coherence_zero == 0 )
		{
		@@ -980,7 +984,11 @@ void ivas_qmetadata_enc_sid_encode(
		else
		{
		/* TODO: still use old sid frame size to keep bitexactness */
		#ifdef FIX_QMETA_SID_5k2
		metadata_sid_bits = ( IVAS_SID_5k2 - SID_2k40 ) / FRAMES_PER_SEC - SID_FORMAT_NBITS;
		#else
		metadata_sid_bits = ( 4400 /IVAS_SID_5k2/ - SID_2k40 ) / FRAMES_PER_SEC - SID_FORMAT_NBITS;
		#endif
		}

		#ifdef DEBUG_MODE_QMETADATA
		@@ -1036,12 +1044,77 @@ void ivas_qmetadata_enc_sid_encode(
		/Encode the quantized diffuseness in raw coding/
		bits_dir = 0;
		bits_diff = 0;
		#ifdef FIX_QMETA_SID_5k2
		if ( ivas_format != SBA_FORMAT )
		{
		for ( b = start_band; b < nbands; b++ )
		{
		q_direction->band_data[b].energy_ratio_index[0] = max( q_direction->band_data[b].energy_ratio_index[0], 4 );
		bits_diff += ivas_qmetadata_encode_quasi_uniform_length( q_direction->band_data[b].energy_ratio_index[0] - 4, DIRAC_DIFFUSE_LEVELS - 4 );
		q_direction->band_data[b].bits_sph_idx[0] = bits_direction_masa[q_direction->band_data[b].energy_ratio_index[0]];

		if ( q_direction->not_in_2D == 0 )
		{
		q_direction->band_data[b].azimuth_m_alphabet[0] = 1 << ( min( 8, q_direction->band_data[b].bits_sph_idx[0] ) );
		bits_dir += ivas_qmetadata_encode_quasi_uniform_length( q_direction->band_data[b].azimuth_m_alphabet[0] - 1, q_direction->band_data[b].azimuth_m_alphabet[0] );
		}
		else
		{
		bits_dir += q_direction->band_data[b].bits_sph_idx[0];
		}
		}

		/* Reduce bit demand by increasing diffuseness*/
		bits_delta = metadata_sid_bits - ( hMetaData->nb_bits_tot - bit_pos_start ) - bits_diff - bits_dir;
		if ( bits_delta > 0 )
		{
		while ( bits_delta > 0 )
		{
		for ( b = start_band; b < nbands && ( bits_delta > 0 ); b++ )
		{
		if ( q_direction->band_data[b].bits_sph_idx[0] < 11 )
		{
		bits_delta -= 1;
		q_direction->band_data[b].bits_sph_idx[0]++;
		}
		}
		}
		if ( q_direction->not_in_2D == 0 )
		{
		for ( b = start_band; b < nbands; b++ )
		{
		q_direction->band_data[b].azimuth_m_alphabet[0] = 1 << ( min( 8, q_direction->band_data[b].bits_sph_idx[0] ) );
		}
		}
		}
		else
		{
		while ( bits_delta < 0 )
		{
		for ( b = nbands - 1; b >= start_band && ( bits_delta < 0 ); b-- )
		{
		if ( q_direction->band_data[b].bits_sph_idx[0] >= 4 )
		{
		bits_delta += 1;
		q_direction->band_data[b].bits_sph_idx[0]--;
		if ( q_direction->not_in_2D == 0 )
		{
		q_direction->band_data[b].azimuth_m_alphabet[0] = 1 << ( min( 8, q_direction->band_data[b].bits_sph_idx[0] ) );
		}
		}
		}
		}
		}
		}
		else
		{
		#endif
		for ( b = start_band; b < nbands; b++ )
		{
		q_direction->band_data[b].energy_ratio_index[0] = max( q_direction->band_data[b].energy_ratio_index[0], 4 );
		bits_diff += ivas_qmetadata_encode_quasi_uniform_length( q_direction->band_data[b].energy_ratio_index[0] - 4, DIRAC_DIFFUSE_LEVELS - 4 );
		q_direction->band_data[b].bits_sph_idx[0] = bits_direction_masa[q_direction->band_data[b].energy_ratio_index[0]];

		if ( q_direction->not_in_2D == 0 )
		{
		q_direction->band_data[b].azimuth_m_alphabet[0] = 1 << ( min( 5, q_direction->band_data[b].bits_sph_idx[0] ) );
		@@ -1069,6 +1142,9 @@ void ivas_qmetadata_enc_sid_encode(
		}
		}
		}
		#ifdef FIX_QMETA_SID_5k2
		}
		#endif
		assert( ( bits_delta >= 0 ) && "Bit budget in Qmetadata SID is violated!!!" );

		/Code diffuseness/

tests/renderer/test_renderer_be_comparison.py

+26 −0

Original line number	Diff line number	Diff line
		@@ -145,6 +145,32 @@ def test_masa(test_info, in_fmt, out_fmt):
		test_info, in_fmt, out_fmt, in_meta_files=FORMAT_TO_METADATA_FILES[in_fmt]
		)

		@pytest.mark.parametrize("out_fmt", OUTPUT_FORMATS_BINAURAL)
		@pytest.mark.parametrize("in_fmt", INPUT_FORMATS_MASA)
		def test_masa_binaural_static(test_info, in_fmt, out_fmt):

		if out_fmt in ["BINAURAL_ROOM_IR", "BINAURAL_ROOM_REVERB"]:
		pytest.skip("Skipping binaural room outputs for MASA as unimplemented.")

		compare_renderer_vs_mergetarget(
		test_info, in_fmt, out_fmt, in_meta_files=FORMAT_TO_METADATA_FILES[in_fmt]
		)

		@pytest.mark.parametrize("trj_file", HR_TRAJECTORIES_TO_TEST)
		@pytest.mark.parametrize("out_fmt", OUTPUT_FORMATS_BINAURAL)
		@pytest.mark.parametrize("in_fmt", INPUT_FORMATS_MASA)
		def test_masa_binaural_headrotation(test_info, in_fmt, out_fmt, trj_file):

		if out_fmt in ["BINAURAL_ROOM_IR", "BINAURAL_ROOM_REVERB"]:
		pytest.skip("Skipping binaural room outputs for MASA as unimplemented.")

		compare_renderer_vs_mergetarget(
		test_info,
		in_fmt,
		out_fmt,
		trj_file=HR_TRAJECTORY_DIR.joinpath(f"{trj_file}.csv"),
		in_meta_files=FORMAT_TO_METADATA_FILES[in_fmt]
		)

		""" Custom loudspeaker layouts """