International organisation for standardisation organisation internationale de normalisation
Task group activities Joint Meetings
Yüklə 8,63 Mb.
|
- Bu səhifədəki naviqasiya:
- MP4 FF extensions for Audio, with Systems, Thursday 9:00 – 10:00
- 3D Audio in Augmented Reality, with 3DV, Thursday 10:00 – 11:00
- 3D Audio Call for Proposals, with Requirements, Thursday 14:00 – 15:00
- Task Group discussions
- Support for MPEG Audio in the Marketplace
Task group activities
The Audio Chair presented
The contribution contained recommended changes to the Systems components to fully describe a raw audio or video signal. There as discussion on the tradeoffs of a simple Registration Authority description versus a simple Systems syntax. Audio experts will propose a revised, more complete Systems syntax that is compatible with a simple Registration Authority description. David Singer, Apple, noted that Motion JPEG 2000 may have Systems syntax for carriage of Linear PCM.
David Singer, Apple, presented
The presentation notes three issues in the MP4 FF and proposes solutions:
Issues to be addressed by Audio experts
It was agreed to incorporate the 32-bit sample rate capability into an existing issue this document as Technologies Under Consideration in 14496-12, ISO Base Media FF. David Singer, Apple, presented
The presenter used the opportunity of the joint meeting to draft the DoC on this item.
The Audio Chair and 3DV experts discussed what is needed in Augmented Reality. Experts noted that Audio technology coordinates are typically listener-centric, while Augmented Reality is scene-centric. Objects in a virtual scene, listener in virtual scene, generate objects position in listener coordinates and render. Possible Audio work: add directivity and distance rendering for objects.
The 3D Audio Call for Proposals was presented and discussed. Gregory Pallone, Orange Labs, reminded experts that Peter Grosche, Huawei, raised an inconsistency in the “envisioned standard” paragraph of the Call for Proposal (3.2), which was silent about HOA. That was fixed by adding the sentence that it “may support HOA inputs” The Call was approved by Requirements. In addition, the 3D Audio CE Methodology was reviewed and the encoder source code requirements for CE integration was highlighted.
Nikolaus Rettelbach, FhG-IIS, presented
The contribution zip archive contains an excel spreadsheet that contains an extensive report on interoperability of 7.1 channel bitstreams and legacy products. Only implicit signalling was investigated. Christof Fersch, Dolby, suggested that a further investigation of the problematic bitstream/product combinations, but now using explicit channel configuration. Daniel Fischer, FhG-IIS, presented
The contribution proposes fixes to the MPEG Surround Reference Software, which are described in the contribution and supplied as corrected software in the contribution zip archive. It was the consensus of the Audio subgroup to issue the contribution as a DCOR. Daniel Fischer, FhG-IIS, presented
The contribution proposes corrections to the specification text and the reference software Corrections to reference software only:
Corrections to reference software and text:
It was the consensus of the Audio subgroup to
Daniel Fischer, FhG-IIS, presented
The contribution has updates to the conformance spreadsheet and the conformance text. The text changes are:
The conformance spreadsheet is modified
Current status: 95% of defined are available and 50% of available are cross-checked.
A Liaison statement to DVB was reviewed and approved.
Thomas Sporer, FhG-IDMT, presented
The contribution presented ideas on how to test for audio quality in “non-sweet-spot” listening positions. It proposes a two-step approach:
The results are subject to this analysis:
It proposes to modify the “alternate listener positions” as specified in BS.1116
The contribution notes that sweet-spot evaluation is done in Test 1.1, and Test 1.2 therefore need only assess off-sweet-spot listener locations. Experts noted this proposal requires that a listener take both Test 1.1 and Test 1.2 and that it would be most robust if Test 1.1 immediately follows Test 1.2. The presenter noted that Test 1.2 should have the same objectives, e.g. assessment high quality audio, as Test 1.1. Further, he noted that the listener instructions should not require that one item (i.e. usually the hidden reference) be scored at 100. If need be, Test 1.2 could use fewer test items or fewer listeners at each off-sweet-spot position. There was large agreement to adopt the proposed lower distance from sweet-spot to off-sweet-spot locations. However, these were left as open issues:
Gregory Pallone, Orange Labs, presented
The presenter noted that Test 1.2, off-sweet-spot, is very important because it represents a very common use case of many people in a home theatre setting. The contribution proposes that Test 1.2 clearly state that it “will use same decoded wavefile as is used in Test 1.1.” It was the consensus of the Audio subgroup to
For Test 1.3, it does not support using a loudspeaker references since:
Four sites have contributed low-resolution BRIR: FhG-IIS, FhG-IDMT, Technicolor, Orange Labs. The presentation noted the advantages of low-resolution vs. high-resolution in binauralizing channel or object based content. The contribution proposes:
The contribution notes that this is equivalent to just choosing a HR-BRIR. Werner Oomen, Philips, stated that since the HRTF are not individualized, the low-resolution is sufficient when one uses HRTF interpolation. Thomas Sporer, FhG-IDMT, noted that based on the experience of his lab, BRIR is better in that it captures aspects of the room, and that LR-BRIR is all that we have. Note that “better” is the ability of a subject to correctly localize a sound in space. He further noted that BRIR often helps subjects to externalize the sound scene. Clemens Par, Suisaudec, asked the presenter if he had evaluated his proposal with respect to localization and coloration. The presenter said that the evaluation was performed both objectively with the figures shown during the presentation, and also with informal listening tests where the difference between HR-HRTF and LR-HRTF was audible. The presenter stated that the use of HR-BRIR instead of LR-BRIR is probably equivalent to what m28181 shows (a decrease of 30 MUSHRA points), but acknowledges that HR-BRIR are not available. However since HR-HRTF are available, they could be used to create the reference for some objects and HOA items (the ones that have the most reverberation for example). All the audio experts agreed that if there were HR-BRIR available, they would have selected them. The Chair noted that, as stated in the contribution, the LR-BRIR has deficiencies for object synthesis. Is this deficiency greater than the deficiency due to a “non-peronalized” BRIR? Presenter responded that LR-BRIR can give incorrect Inter-aural Time Differences (when interpolating between loudspeaker positions), and hence might be a greater deficiency as compared to the lack of personalization. This discussion topic came up again later, and Jean-Marc Jot, DTS, supported the use of HR-HRTF instead of LR-BRIR for items whose sources are available. Gregory Pallone, Orange Labs, stated that for several HOA items, he could provide the original mono sources and HOA panning laws. However, it was finally agreed to not use the HR-HRTF because they were:
The contribution notes that Test 1.4 is very important. The contribution has a MATLAB script in the Annex that can perform the randomization. It was agreed to have a break-out to agree on “sectors” of the 22.2. loudspeaker set and that there should be different “sectors” for the choice of 5 of 22 and 10 of 22. Jan Plogsties, FhG-IIS, presented
Test 1.3 (headphones): The contribution proposes that
Open issues
Clemens Par, Suissaudec, presented
The contribution proposes a process to determine the subjective listening test methodology for Test 1.3
Thomas Sporer, FhG-IDMT, noted that the proposed test assumes decorrelated loudspeaker signals, and that the 22.2 presentation is expected to have high correlation between adjacent loudspeaker pairs. He further noted that a test with headphones on or off is flawed in its ability to localize sound sources. Jan Plogsties, FhG-IIS, offered the anecdotal evidence that he cannot tell the difference between loudspeakers and headphones with personalized BRIR. Andreas Silzle, FhG-IIS, presented
The contribution presents the results of an experiment in object-based signal rendering. The experiment used point source objects rendered to the position of an actual loudspeaker as the reference with a phantom source created using VBAP between the four adjacent loudspeakers and an experimental renderer. Subjects were asked to judge
The presenter noted that the VBAP and experimental renderer were fully 30 MUSHRA points below the reference. The contribution proposes an additional test to specifically assess rendering under the restriction of audio objects with static locations in the sound stage. The test would consist of:
Werner Oomen, Philips, supported the test proposal since it is able to correctly test the rendering performance. Jean-Marc Jot, DTS, noted that there could be a multi-channel reference setup (i.e. multiple speakers and multiple sound objects). Experts suggested that this test could be incorporated into Test 1.1. It was observed that the proposed test item could be 22 channels of low-level pink noise plus 1 object that is the proposed test signal. Johannes Boehm, Technicolor, suggested that this proposed test be added to Test 1.4. Peter Grosche, Huawei, presented
Huawei feels that the Personal TV and Mobile TV use cases are very important. The presenter noted that all tests should have the listeners in the BS.1116 reference position. If off-sweet-spot listening, then the position must be clearly defined. For the use cases of Personal TV and Mobile TV, the contribution proposes to evaluate two loudspeaker configurations for Personal TV and Mobile TV, but using loudspeakers in the Home Theatre” setting. If it is necessary to reduce the test workload, Huawei proposes to remove 8.1 or 7.1 channel configurations. The most important aspect of the proposal is to test reproduction using only loudspeakers located only in the front of the subject. Henney Oh, WILUS STRC, stated support for the Huawei “front-speaker-evaluation” proposal. Clemens Par, Suissaudec, and Andreas Silzle, FhG-IIS, were sceptical that the test would give information that is relevant to real product implementations. Concerning the CfP, Huawei proposes to make the Phase 2 timetable use “symbolic” meeting numbers. On the Draft 3D Audio CE Methodology, Huawei proposes that,
This last proposal from Huawei is a comment on the “Draft MPEG Audio CE methodology for 3D Audio Work” which is a contribution authored by the Chair. This contribution is the product of considerable effort on his part to get a document that could gain the consensus approval of the Audio subgroup. In his opinion, it would be difficult to incorporate the first item (CE source code platform) and gain consensus of the group. 3D Audio Content Deep Sen, Qualcomm, presented
The contribution provides details of the Qualcomm HOA recording process and format. The presenter noted that Qualcomm proposed to eliminate the SFD2 format, since SFD1 is fully equivalent. Gregory Pallone, Orange Labs, presented
The contribution documents the items submitted by Orange Labs and the HOA file format parameters for all items, where some items are new and not previously documented. The renderer coefficients are available as part of the content. Jeongil Seo, ETRI, presented
The contribution describes the ETRI test items content. In addition, it proposes a clarification of audio object metadata and a description of the ETRI object renderer. Robert Steffens, Iosono, presented
The contribution documents a description of the Iosono/FhG-IDMT content. In particular, the new items are in OA3D format and are shortened to approximately 20 seconds. The presenter noted that if a few hundred ms of the Betty3 item are trimmed off, the number of active items in the digital never exceeds 26. This edited item can be made available this week. Thomas Sporer, FhG-IDMT, noted that many items may need to be modified and re-submitted. Andreas Silzle, FhG-IIS, reported that FhG-IIS has investigated digital headroom of the items and headroom after 5.1 and stereo downmix. In order to not clip for downmix, he proposed these options:
Johannes Boehm, Technicolor, presented
The contribution gives information on rendering for HOA content. Experts were in agreement that listeners should be instructed to grade down any differences in loudness. Robert Steffens, Iosono, presented
The contribution gave several examples of scenarios where object interactivity would be valuable. It proposes to add an interactivity flag to the object metadata in the 3D Audio file format. If further proposes to define a test to assess the ability of a technology to implement object interactivity, for example:
Chair noted that the range of interactivity could be from Tennis match (variable level of commentator and crowd) to Karaoke (infinite attenuation of main vocals). Oliver Wuebbolt, Technicolor, noted that dubbing into an alternate language is another use case that has infinite attenuation. Jan Plogsties, FhG-IIS, expressed that interactivity supports important use cases. Discussion of CfP There was extensive discussion on the CfP, particularly on the specification of the subjective performance tests. Hours of drawing on two flip charts finally converged to the text found in the CfP document. It was agreed that Phase 2 is essentially a CE with a very specific timeline and subjective performance tests. It was noted that a CE proponent should be able to choose the test set that is most appropriate for testing the proposed technology. Response to Swiss NB Comment The issued CfP document specifies subjective tests that assess the performance of presentation of 22.2 channel test items on loudspeakers over a broad range of bit rates: 1.2 Mb/s, 512 kb/s, 256 kb/s, 128 kb/s, 96 kb/s, 64 kb/s and 48 kb/s. In order to manage the workload of this broad range of tested bitrates, some of the lower bitrates (e.g. 96 kb/s and 48 kb/s) may have fewer listeners than the for the other bitrates, for example: 20 listeners for most rates, but only 10 listeners for 96 kb/s and 48 kb/s.
|