Organisation internationale de normalisation
Symbol System description
Yüklə 7,54 Mb.
|
Hence, meaningful difference statistics are:
Roch Lefebvre, University of Sherbrook/VoiceAge, requested that pooled results using only cross-check sites be presented. It was decided to continue discussion on this topic Tuesday after lunch. Seungkwon Beack, ETRI, presented
The contribution presented the results of a listening test. At an operating point of 12 kb/s mono and differential MUSHRA analysis (WD+CE – WD) showed 2 items worse and mean worse. Jeongook Song, Yonsei University, presented
The contribution reviewed most recently proposed CE technology, which shown in the following block diagram: An The contribution presented the results of two listening tests. At an operating point of 12 kb/s mono and differential MUSHRA analysis (WD+CE – WD) the Yonsei University test site showed 2 items better and mean better. Test results from ETRI showed 3 items worse and mean worse, which is not consistent with results from the Yonsei test site. The presenter noted that eLTP technology should, in theory, provide coding gain but acknowledged that this was not evident in the listening tests. Considering the short timeline to final standardization, the proponent elected to withdraw the CE. Takeshi Norimatsu, Panasonic, presented
This contribution presented the results of a listening test. At the operating point of 16 kb/s mono, analysis of MUSHRA differential scores showed that for CGG 1 item was worse and mean worse; for CGG+POST 1 item better, 1 item worse and mean worse. The contribution also reported on a cross-check of the CE decoder: all CE bitstreams were decoded to the CE waveforms to within +/- 1 lsb, except for one item which was within +/- 3 lsb. Jeongook Song, Yonsei University, presented
This contribution presented the results of a listening test. At the operating point of 16 kb/s mono, analysis of MUSHRA differential scores showed that for CGG 1 item was worse and mean worse; for CGG+POST 3 item worse and mean worse. Markus Multrus, FhG, presented
The contribution noted that a single, unified global gain syntax element in USAC would be a very efficient means to adjust signal level in the compressed domain. The current and proposed state of gain syntax elements is shown in the following figure: Note that the rightmost column indicates the net change in gain syntax bits associated per frame is zero. This contribution presented the results of a listening test. At the operating point of 16 kb/s mono, analysis of MUSHRA differential scores showed that for CGG there was no difference; for CGG+POST there was no difference. When pooling all listening data, analysis of MUSHRA differential scores showed that for CGG there were 4 worse and mean worse; for CGG+POST there were 2 worse, mean worse. The presenter acknowledged that the CE did not maintain audio quality. However, in the time since the listening test items were prepared a bug was identified and a correction implemented in the CE decoder. It is anticipated that this fix will have a significant positive impact on the CE performance, and so the proponent requested that a workplan be drafted to repeat the CE and bring new evidence to the next meeting. Henney Oh, Yonsei University, suggested that if the CE listening test is repeated, it be done with a higher bitrate. The Chair stated that he was very supportive of this work, as it is associated with bitstream issues that he previously identified as areas that he would like the group to work as the standard nears completion. Nevertheless, he proposed that the issue of drafting the CE workplan be taken up after all CE contributions are reviewed. Miyoung Kim, Samsung, presented
The contribution reports on a listening test at the16 kb/s stereo operating point. The systems under test were: Sys1 RM8 (as represented by the reference bitstreams) Sys2 RM8+IPD (as represented by explicitly activating IPD) Sys3 RM8+CE In differential analysis of the MUSHRA scores: For (RM8+CE - RM8), RM8+CE better in the mean For (RM8+CE - RM8+IPD), RM8+CE is better for 1 item. Heiko Purnhagen, Dolby, presented
The contribution reports on a listening test at the20 kb/s stereo operating point. In differential analysis of the MUSHRA scores: For (RM8+CE - RM8), RM8+CE better for 1 item. For (RM8+CE - RM8+IPD), RM8+CE is worse for 1 item. Grégory Pallone, Orange Labs, presented
The contribution reports on two listening tests, one at 16 kb/s and the other at 20 kb/s stereo operating point. In differential analysis of the 16 kb/s MUSHRA scores: For (RM8+CE -RM8), RM8+CE worse for 3 item. For (RM8+CE - RM8+IPD), RM8+CE is worse for 2 item, mean worse and 1 better. In differential analysis of the 20 kb/s MUSHRA scores: For (RM8+CE -RM8), RM8+CE is worse for 1 item and better for 1 item. For (RM8+CE - RM8+IPD), RM8+CE is worse for 1 item and 1 better for 1 item. In addition, the contribution reports on the cross-check of the CE deoder. It was verified that the CE bitstreams decoded exactly to the CE waveforms. Julien Robilliard, FhG, presented
The contribution reports on two listening tests, one at 16 kb/s and the other at 20 kb/s stereo operating point. In differential analysis of the 16 kb/s MUSHRA scores: For (RM8+CE -RM8), RM8+CE better for 2 items and worse for 1 item. For (RM8+CE - RM8+IPD), RM8+CE is worse for 3 item and mean worse. In differential analysis of the 20 kb/s MUSHRA scores: For (RM8+CE -RM8), RM8+CE is better for 7 items and mean better. For (RM8+CE - RM8+IPD), RM8+CE is worse for 4 items and mean worse. David Virette, Huawei, presented
The CE technology proposes to fix the OTT box for the case that CLD=0dB and ICC=−1, where the current system synthesizes the output based only on the decorrelated mono signal. The contribution reports on two listening tests, one at 16 kb/s and the other at 20 kb/s stereo operating point. In differential analysis of the 16 kb/s MUSHRA scores: For (RM8+CE -RM8), RM8+CE better for 1 items and worse for 1 item. For (RM8+CE - RM8+IPD), there is no difference. In differential analysis of the 20 kb/s MUSHRA scores: For (RM8+CE -RM8), RM8+CE is better for 1 item. For (RM8+CE - RM8+IPD), RM8+CE is better for 1 item. It also presents results pooled over all test sites. In differential analysis of the 16 kb/s MUSHRA scores: For (RM8+CE - RM8), RM8+CE better for 2 items and worse for 2 items. For (RM8+CE - RM8-IPD), RM8+CE is worse for 3 items. In differential analysis of the 20 kb/s MUSHRA scores: For (RM8+CE - RM8), RM8+CE is better for 1 item, worse for 1 item and better in the mean. For (RM8+CE - RM8-IPD), RM8+CE is worse for 2 items. The presenter also reviewed the consistency of test results across all test sites The presenter acknowledged that the IPD system appears to be superior to the CE technology. However, the CE technology would still be interesting for very low bit rates where IPD is not transmitted. Jullian Robilliard, FhG, noted that transmission of IPD consumes 0.4% of total bitrate over all items at 16 kb/s. Hence, there is no reason to disable IPD at this rate. Markus Multrus, FhG, observed that at 12 kb/s and 8 kb/s, it would likely be most beneficial for the core coder to operate in mono, so that these tools would not be used. It is the consensus of the Audio Subgroup that the CE technology does not show sufficient performance to be incorporated into USAC, particularly since the IPD technology is already available in USAC. Yüklə 7,54 Mb. Dostları ilə paylaş:
|