AhG on DRC and 3D Audio
The AHG on Dynamic Range Control (DRC) and 3D Audio and Audio Maintenance met Sunday October 27 1000-1800 hrs at the MPEG meeting venue.
3D Audio Binauralization CE
Listening Test Site Reports
Representatives from each listening test site presented
m31214
|
Qualcomm's Binaural CE Listening Lab – Conditions & Methodology
|
pxiang@qti.qualcomm.com, dsen@qti.qualcomm.com, npeters@qti.qualcomm.com, mmorrell@qti.qualcomm.com
|
m31272
|
Listening test report of ETRI for MPEG-H 3D Audio Binaural CE
|
Taejin Lee, Jeongil Seo, Kyeongok Kang
|
m31298
|
Listening test report of YSU for MPEG-H 3D Audio Binaural CE
|
Taegyu Lee, Henney Oh, Young-cheol Park, Dae Hee Youn
|
m31358
|
Samsung Listening Test for 3D Audio Binaural Core Experiment
|
Namsuk Lee, Sang Bae Chon, Sunmin Kim
|
m31424
|
Orange listening tests for the CE on RM0-CO binauralization
|
Gregory Pallone
|
m31435
|
Fraunhofer IIS Listening Test Results for Binaural CE for MPEG-H 3D Audio
|
Simone Füg, Jan Plogsties
|
m31467
|
Huawei listening test report for the binauralization CE
|
Peter Grosche, Panji Setiawan
|
The Chair presented a report on the combined data
m31414
|
MPEG-H 3D Audio Binaural CE Subjective Test Results
|
Schuyler Quackenbush
|
Technical Descriptions
Nils Peters, Qualcomm, presented
m31211
|
Technical Description of Qualcomm's Candidate for the Binaural Core Experiment
|
Pei Xiang, Nils Peters, Martin Morrell, Deep Sen
| -
Single band
-
Direct and Early reflection per loudspeaker
-
Late Reverberation is stereo
-
Qualcomm technology essentially same as Phase 1 submission - model (direct path = early reflection + hrtf) and late-reverb.
-
Only difference (from Phase 1) is - smaller reverb tail & adaptive reverb gain
Henney Oh, WILUS, presented both the ETRI and Yonsei/WILUS contributions
m31271
|
Description of ETRI proposal for MPEG-H 3D Audio Binaural CE
|
Jeongil Seo, Yong Ju Lee, Taejin Lee, Seungkwon Beack, Kyeongok Kang
|
m31297
|
Description of YSU proposal for MPEG-H 3D Audio Binaural CE
|
Taegyu Lee, Henney Oh, Young-cheol Park, Dae Hee Youn
|
ETRI/Yonsei/WILUS jointly propose a binaural rendering system, but two different versions were submitted with different tuning parameters.
Summary of submission technologies;
-
Frequency-varying filter-length for D&E (Direct & Early Reflection) - truncate in accordance with the RT20 of the BRIR
-
Integrated energy decay matching (EDM) for LR (Late Reverberation)
-
Inter-aural coherence matching (ICM) for LR
-
1-tap TDL (Tapped Delay Line) for SBR bands (optional)
Different parameter optimization Yonsei/WILUS and ETRIMultiband, QMF
Listening test results analysis based on difference scores with RM0;
-
Sys3 (Yonsei/WILUS, HQ mode) is the only system statistically better in overall sense
-
Sys3 is the unique system that is statistically no worse but 3 items are better
Computation complexity of submitted technologies;
-
FoM_complexity of Sys3 is 92.84
-
FoM_complexity of Sys5 is 94.62
Gregory Pallone, Orange, presented
m31421
|
Technical Description of the Orange proposal for the Binaural CE on RM0-CO
|
Gregory Pallone, Marc Emerit
| -
Same technology as in HOA-RM0
-
Single band
-
DE, LR split processing
Only want to standardize A (DE) and B (LR) coefficients, not means to get A, B from BRIR. Could provide informative code for getting A, B from BRIR.
Another contribution suggests a normative interface for the A, B.
Jan Plogsties, FhG-IIS, presented
m31437
|
Binaural Core Experiment - Fraunhofer IIS System Description
|
Simone Füg, Jan Plogsties
| -
Multiband, QMF, up to 48 bands (18 kHz), constant for all sub-bands
-
DE/LR boundary BRIR adaptive
-
LR is a stereo downmix processed by frequency-adaptive IIR filtering.
Panji Setiawan, Huawei, presented
m31468
|
Technical Description of the Huawei Binaural CE proposal
|
Simone Fontana, Karim Helwani, Peter Grosche, Panji Setiawan,
|
Comments
Taegyu Lee, Yonsei University, presented
m31311
|
Comments on the evaluation methodology for the computational complexity of binaural renderer
|
Taegyu Lee, Henney Oh, Young-cheol Park, Dae Hee Youn
|
This contribution presents comments and suggestions to help to understand and clarify the Figure of Merit (FoM) for binaural CE. It notes that the evaluation methodology for block-wise fast convolution consented in the 105th MPEG meeting has a bug. The revised evaluation methodology is presented.
The presentation gave complexity issue between processing domains.
-
When bitrate is 512kbps, the binaural processing in the QMF domain does not require 20 times QMF synthesis. Thus, binaural processing in QMF domain has significant computational gain for the QMF domain input signals.
-
When bitrate is 1.2Mbps, the 22 times QMF analysis and 2 times QMF synthesis is required to the binaural processing in QMF domain.
Using a revised evaluation methodology, revised FoMs of each submission of binaural CE are presented. Matlab script for calculation of each Binaural CE submission is provided.
The presenter provided Excel spreadsheets containing computational complexity of each submission for binaural CE and the chair made the spreadsheets available to the AhG members.
It was decided that Audio experts will check the spreadsheets.
Disussion
The Chair presented the combined listening test data, but now with proponent names.
-
Single band, time domain Qualcomm, Orange,
-
Multi band, SBR QMF ETRI, Yonsei, FhG, Huawei, RM0
Recommendation
-
That the Audio subgroup further analyse the architectures,
-
Discuss whether architecture is used for only a range of bitrates
-
Gather and check complexity numbers
-
Consider issues of architecture unification
Submissions to CfP on DRC
Toru Chinen, Sony, presented
m31359
|
Technical description of Sony proposal for Dynamic Range Control Technology
|
Hiroyuki Honma, Runyu Shi, Toru Chinen, Yuki Yamamoto, Mitsuyuki Hatanaka, Masayuki Nishiguchi
| -
Builds on the apple proposal (from 104th meeting)
Fabian Kuech, FhG-IIS, presented
m31384
|
Description of the Fraunhofer IIS Submission for the DRC CfP
|
Fabian Kuech, Christian Uhle
| -
DRC
-
Loudness control
-
Clipping Prevention (function of DRC and LC)
-
Guided limiter
-
(Peak limiter), if no gPL provided
-
Builds on the apple proposal (from 104th meeting), but adds gCP and PL.
-
gCP uses the same syntax and structures as a DRC gain sequence.
Frank Baumgarte, Apple, presented
m31471
|
Description of Apple's proposal considering the CfP on Dynamic Range Control Technology
|
Frank Baumgarte, David Singer
|
m31472
|
Apple's Dynamic Range Control Proposal: Listening Test Results
|
Frank Baumgarte, Fabian Kuech
|
m31473
|
Apple's Dynamic Range Control Proposal: Evaluation
|
Frank Baumgart
|
Apple’s submission for the DRC CfP includes an improved an enhanced DRC tool compared to the proposal from the 104th meeting. It is a universal tool that supports time-domain and sub-band domain DRC. Multiband DRC is also supported in both domains. The gain interpolation is based on splines to achieve smooth DRC gain transitions. The corresponding enhancements in the Sample Entry of the file format (m31470) include a list of descriptors of the DRC effect to enable an informed choice of the most appropriate DRC at the decoder. Moreover, various configurations can be supported such as pre- and/or post-downmix DRC.
Listening test results at two testing sites (Apple, Fraunhofer IIS) show that the proposed DRC tool delivers transparent quality while the current MPEG DRC can introduce distortions in AAC when a fast-acting DRC is applied.
Evaluations of the proposed tool show a minor bit rate increase for the coded DRC gains versus the current MPEG standard. The complexity is low if the tool operates without using the time-domain DRC filterbank. Drc config
Recommendations
-
Seems that FhG-IIS proposal can be supported by Apple proposal, but need to check.
-
What is the impact of the Sony proposal considering the current Apple proposal . Need to check.
Recommendations and review of AhG Report
The AhG members reviewed the AhG report and agreed on the report’s recommendations made to the Audio subgroup.
Dostları ilə paylaş: |