International organisation for standardisation organisation internationale de normalisation
Download 3.13 Mb.
|
- Navigate this page:
- 4.5.2Lossless Coding
- 4.5.3Spatial Audio
- 4.5.4MPEG-7
- 4.5.5Symbolic Music Representation
4.5Task Group discussions4.5.1MPEG-2/4 Audio IssuesAndreas Schneider, Coding Technologies, presented
New conformance test sequences and conformance criteria for the same have been proposed. Heiko Purnhagen, Coding Technologies, presented
This contribution was mostly editorial issues. There were two fixes to the signalling of the PS AOT. Andreas Schneider, Coding Technologies, presented
One previously undefined meaning of a byte alignment has been clarified. The new behaviour provides a high degree of flexibility while at the same time being consistent with the behaviour of the reference software for most cases. Werner Oomen, Philips, presented
These are mainly editorial and clarification issues. The main issue to clarify is how parameters are interpolated. Werner Oomen, Philips, presented
There are now 16 conformance test streams for SSC .The conformance data covers mono, parametric stereo, relevant parameterizations (i.e. tones, noise, and envelope), plus two accuracy classes, those being full and fixed point levels of accuracy. Sang-Wook Kim, Samsung, presented
Sang-Wook Kim, Samsung, presented
Sang-Wook Kim, Samsung, presented
The Chair voiced the opinion that the proposal in m12079 needs more study and discussion than would be afforded if this were accepted into e.g. the ALS FPDAM, in that its final ballot closes in one month. Hence it was the consensus of the Audio Subgroup to put this contribution plus m12081 into a separate output document for further study. Audio Coding Tool Repository Perhaps new work could leverage the existing body of Audio tools. 4.5.2Lossless CodingContinued discussion on the CBAC proposal Thomas Wiegand, HHI, indicated that he is withdrawing his support for the proposal since is in fact not the same algorithm as is used in AVC. Because of this, he feels that there is too much risk in producing a robust specification in the current timeline of ALS standardization. Takehiro Moriya, NTT, has checked the submitted source code implementation, and found that, for 24-bit word lengths, the decoder is slower than the current RM8 decoder. Xiaolin Wu, McMaster University, commented that he is willing to do the work to make this a viable proposal. However the Audio Chair noted that it is the meeting prior to the meeting in which the final text must be created, and feels that there may be too much risk in adopting technology that had not received sufficient review. Tilman Liebchen, TUB, agreed that the real problem with the proposal is that it comes very late in the standardization process. The consensus of the Audio Subgroup is to not proceed with this CE proposal. The reasons are that this proposal is too late in the standardization process, that the increase in performance and decrease in complexity of the complete system is quite small, and hence the risks outweigh the benefits to the degree that no action is warranted. RLS/ALS predictor Haibin Huang, I2R, presented
In this contribution, the baseline is equivalent to ALS RM12 without LTP or multichannel prediction. There was considerable discussion on the performance of this proposal, including the complexity of hardware-based systems (in which maximum complexity is relevant) and the complexity of general-purpose processor-based systems (in which average complexity is relevant). Takehiro Moriya, NTT, brought forward average complexity information. Ralf Geiger, FhG, brought forward cross-check information on the I2R CE proposal. It was the consensus of the Audio Subgroup to adopt CE13 into the ALS specification. However it is the understanding that the current short-term and LTP predictor technology is able to be put into its own profile whenever profiles are defined. If additional cross-check information (e.g. a cross-check from RealNetworks) raises significant new issues, then this decision may be revisited. Later in the week the Audio Chair presented on behalf of Yuriy Reznik, RealNetworks,
This contribution gave additional information on the performance and complexity of the adaptive predictor. Although it raised issues of complexity, it was the understanding of the Audio Subgroup that these issues will be addressed via profiles. For example, there may be a hierarchical set of two profiles, one of which (“low complexity”) contains only the current forward predictor, while another (“high performance”) contains both the forward predictor and the adaptive predictor. In this way, the marketplace can select between low-complexity and high-performance technology options. Inter-channel prediction The inter-channel prediction proposal showed modest but consistent improvement across the Fs/Wd subsets of the signal set. It shows significant improvement for a selected 8-channel audio signal, but inconsistent performance for 256-channel biomedical data. Tilman Liebchen, TUB, noted that it would be possible to make the M/S coding and interchannel prediction tool be active on a block-by-block basis, in which case either none, M/S or Interchannel prediction could be applied dynamically at every block. It was the consensus of the Audio Subgroup that this technology be incorporated into the ALS specification, with the understanding that this dynamic joint channel coding will also be incorporated into the specification. 4.5.3Spatial AudioSeven sites participated in the listening tests for “candidate RM0.” The Spatial Audio workplan, N6814, set out the following criterion that candidate RM0 must satisfy in order to be accepted as RM0:
Inseon Jang, ETRI, presented
The results were mixed, in that RM0 satisfied the “no worse than” criterion for only two of the three tests. David Virette, France Telecom, presented
France Telecom conducted test 1a, in which RM0 satisfied the “no worse than” criterion. Juergen Herre, FhG, presented
In the FhG test results, RM0 satisfied the “no worse than” criterion for all three tests. In addition, this criterion was satisfied for each test item in each test. FhG conducted an “extended T1a” test, in which three parameterizations of RM0 were tested: RM0, RM0 high-rate, RM0 low-rate. The results showed that RM0 high-rate (called “high quality) is better than RM0 at the 95% level of significance, while RM0 low-rate is not different from RM0 at the 95% level of significance. Finally, the presentation notes that the average side information rate for candidate RM0 is significantly lower that than of the original FhG/Agere or CT/Philips proponent systems. Werner Oomen, Philips, presented
The results showed that in test 1a, 2a and 3, RM0 satisfied the “no worse than” criterion. On a per-item basis, RM0 shows significant improvement over CT/P proposal for the applause items. In a separate test that compares RM0, RM0 low-rate and RM0 high-rate, RM0 and RM0 low-rate are not different, while RM0 high-rate has distinctly better performance at the 95% level of significance. Kristofer Kjörling, CT, presented
The results showed that in test 1a, 2a and 3, RM0 satisfied the “no worse than” criterion. The presenter noted that candidate RM0 had significantly lower bitrate than either of the original FhG/Agere or CT/Philips proponent systems. Coding Technologies also tested RM0, RM0 high-rate, RM0 low-rate in an additional test. The results showed that RM0 low-rate is no worse than RM0 at the 95% level of significance. Itaru Kaneko, TPU, presented
Tokyo Polytechnic University conducted an “extended” test 1a, in which all of RM0, RM0 low-rate and RM0 high-rate were all included in the test. Kurt Jacobson, University of Miami, presented
The results showed that in test 1a, 2a and 3, RM0 satisfied the “no worse than” criterion. Werner Oomen, Philips, presented
The contribution notes that candidate RM0 has the property that the spatial side information has a bit-rate scalable structure. In order to demonstrate this, three parameterizations of RM0 were made available to the test sites: RM0, RM0 low-rate and RM0 high-rate (to be referred to as “high-quality”). For test 1a, low-rate parameterization has a side-information rate of less than 6 kb/s, which is half the rate of the RM0 rate of approximately 12 kb/s. Discussion Performance Werner Oomen, Philips, made a presentation of an analysis with pooled and post-screened test data. Two post-screening rules were applied:
He showed a plot of hidden reference score and 95% confidence interval, with the data sorted by decreasing hidden reference score, and this plot motivated the cut-off of hidden reference score of 90 as the post-screening process. In some cases (e.g. T1a and T2a), this post-screening resulted in removal of a significant fraction of the listener population. He presented graphs of the performance of the systems under test after the post-screening process. The pooled and post-screened results showed that candidate RM0 satisfied the “no worse than” criterion for all tests. Bit rates The following table (from m12005), summarizes the average bit rates of candidate RM0, other parameterizations of RM0, and the two proponent systems on which it is based.
Conclusions It is the consensus of the Audio Subgroup that “candidate RM0” passes all acceptance criteria set forth in N6814, and hence that it become RM0. WD text will be available Friday as an output document. A workplan for Spatial Audio Coding work will indicate the location of the reference code that implements RM0. The workplan also indicates a schedule for the work, but this proposed schedule did not receive unanimous support, and hence will continue to be discussed.
Juergen Herre, FhG, presented a draft of a revised procedure for MPEG-4 Audio core experiments. There was much good discussion leading to a document that all felt served the upcoming Spatial Audio Coding work. 4.5.4MPEG-7Matthias Gruhne, FhG, presented
The goal of this proposal is to expand the applicability and performance of the current AudiosignatureDS to signals that are highly distorted, for example as a result of GSM audio coding and possible associated error mitigation of transmission errors. Three different test scenarios demonstrated that the EnhancedAudioSignatureDS provided significantly better query performance than the AudioSignatureDS. The Audio Subgroup agrees to accept this as a core experiment, and a cross-check will be expected at the next MPEG meeting. 4.5.5Symbolic Music RepresentationGorgio Zoia, EPFL, presented
This contribution details how the Capella proposal might have better presented its technology in the CfP process. Although this is interesting information, and informal inspection suggests that this new information would not have altered the CfP results. In any case, it is not appropriate to alter those CfP results. Parties that have an interest in Capella are welcome to incorporate components and capabilities of that technology into the SMR WD via the core experiment process. Directory: sc29 -> open -> 29view 29view -> International organisation for standardisation organisation internationale de normalisation 29view -> International organisation for standardisation organisation internationale de normalisation Download 3.13 Mb. Share with your friends:
|