Organisation internationale de normalisation
Received National Body Comments and Liaison matters
Yüklə 7,54 Mb.
|
24.7Received National Body Comments and Liaison mattersThere were none. 24.8Plenary DiscussionIt was the consensus of the Audio Subgroup to affirm the recommendations of the AhG, as recorded in Section 3.1. 25Record of AhG meetings25.1AhG Meeting on USAC -- Sunday 1000-1800Takeshi Norimatsu, Panasonic, presented
The contribution reports the results of a listening test. At the operating point of 8 kb/s and differential MOS score analysis, the results showed 3 items better and mean score better at the 95% level of significance. The document also reports that Panasonic verified the CE decoder software, in that the CE bitstreams decoded exactly to the CE waveforms. Heiko Purnhagen, Dolby, presented
The contribution reports the results of a listening test. At the operating point of 8 kb/s and differential MOS score analysis, the results showed 2 items better and mean score better at the 95% level of significance. Max Neuendorf, FhG, presented
The contribution reports the results of a listening test. At the operating point of 8 kb/s and differential MOS score analysis, the results showed 4 items better, 2 worse and mean score better at the 95% level of significance. Toru Chinen, Sony, presented
Predictive Vector Coding technology adds an alternave coding scheme for SBR envelope scale factors, one that has greatest performance at the lowest bit rates (i.e. 8 kb/s and 12kb/s). The technology used in this CE listening test uses adaptive factors in the encoder and decoder, e.g. having the SBR oversampling rate as a parameter rather than fixed as 2. In addition, PVC adds bitstream syntax elements to support SBR sinusoidal coding. Complexity of PVC typically leads to a .45 WMOPS increase in decoder complexity, or a 9% increase in decoder WMOPS. PVC requires 2070 bytes of ROM, or 2% increase in of decoder ROM. The contribution reports the results of a Sony listening test. At the operating point of 8 kb/s and absolute MOS score analysis, the results showed that the mean is better. At the operating point of 8 kb/s and differential MOS score analysis, the results showed 7 items better and mean score better. The contribution reports the results of pooling all listening test. At the operating point of 8 kb/s and differential MOS score analysis, the results showed 10 items better and mean score better. The highest differential scores were 4 MUSHRA points. When comparing consistency across listening test sites, 5 items showed consistent improvement for at least 2 out of the 4 test sites. Werner Oomen, Philips, requested time to check the pooled data, for example looking at pooling without considering the Sony data. It was agreed to bring this matter up for further discussion Tuesday after lunch. Taejin Lee, ETRI, presented
The contribution reports the results of a listening test. At the operating point of 8 kb/s mono and absolute MOS score analysis (WD7 – WD7+CE), the results showed 1 item better and mean score better. At the operating point of 8 kb/s mono and differential MOS score analysis (WD7 – WD7+CE), the results showed 10 items better and mean score better. Eunmi Oh, Samsung, presented
The contribution reports the results of a listening test. At the operating point of 8 kb/s mono and differential MOS score analysis (WD7 – WD7+CE), the results showed no statistically significant differences. Comment from listeners indicated that the CE technology provided an increased bandwidth but with increased distortion. Samsung also verified CE decoder executable: the CE bitstreams decoded exactly to the CE waveforms used in the listening test. Max Neuendorf, FhG, presented
The contribution reviews the CE technology, which is to permit a 4:1 SBR up-sampling by re-using existing SBR tools to the greatest extent possible. The presenter notes that such a 4:1 system exists in the DRM system. In the 4:1 system the core coder only codes the signal from the 16 lowest subband filters (as opposed to the lowest 32 for the 2:1 system). Two complexity comparisons were presented. The first assumes that the core coder runs at the same sampling rate. In this case, at 8 kb/s mono, the complexity of the 4:1 system as compared to the current 2:1 system is 142%, which is due to running the SBR synthesis filterbank at double the sampling rate and also running the FFT harmonic at a higher rate. The second assumes that the SBR synthesis filterbank runs at the same sampling rate. In this case, at 8 kb/s mono, the complexity of the 4:1 system as compared to the current 2:1 system is 73%, which is due to running the core coder and the FFT harmonic transposer at half the sampling rate. It reports the results of a test at the FhG site, at the operating point of 8 kb/s mono and differential MOS score analysis (WD7 – WD7+CE), the results showed 6 items better and mean better. The results of a test at Dolby site, at the operating point of 8 kb/s mono and differential MOS score analysis (WD7 – WD7+CE), the results showed 5 items better and mean better. It also reports results for all data pooled, in which 9 items are better and mean better. Three items are better at the level of 6 MUSHRA points. When looking at consistency across all test sites, 8 items are reported better for at least 2 of the 4 sites and also show improvement in the mean for the pooled data. There was some discussion, and it was noted that the two complexity comparisons show that the 4:1 SBR tool provides a “middle ground” between 2:1 system at low output sampling rate and hence low audio bandwidth (but reasonable audio quality at 8 kb/s) and the 2:1 system at high output sampling rate and hence high audio bandwidth(but most likely very poor audio quality at 8 kb/s). David Virette, Huawei, asked whether the consistency of just cross-check sites could be presented. Heiko Purnhagen, Dolby, had done those calculations and presented that data as both a table of “items better” and as MUSHRA differential score plots. The Chair noted that some items show improvement of 6 or even 8 MUSHRA points. It was the consensus of the AhG to recommend to the Audio Subgroup to incorporate this CE technology into the USAC DIS. Takeshi Norimatsu, Panasonic, presented
The contribution reports the results of a listening test. At the operating point of 8 kb/s mono and differential MOS score analysis (WD7+2:1SBR+PCV – WD7+4:1SBR+PVC), the results showed differences at the 95% level of significance. It also reported on the number of listeners who preferred one system to the other (or vice versa) for each item. This preference was nearly evenly split amongst the listeners for every item, suggesting that the true preference distribution may be bi-modal due to strong individual preferences. Toru Chinen, Sony, presented
The contribution reviews the motiviation for the CE: that it should be checked that PVC also works with the proposed 4:1 SBR configuration. Note that for this CE, the parameterization of PVC factors is used (e.g. RATE). A favourable outcome would indicate that PVC can be used in combination with 4:1 SBR without any negative interaction. It noted that there is negligible increase in complexity for combining PVC with the 4:1 SBR system. Any increase in complexity is due to running PVC over 48 rather than 32 subbands in the high frequency generation. The contribution reports the results of pooling the Sony, FhG and Dolby listening test data. At the operating point of 8 kb/s mono and absolute MOS score analysis (WD7+2:1SBR+PCV – WD7+4:1SBR+PVC), the results showed the mean better at the 95% level of significance At the operating point of 8 kb/s mono and differential MOS score analysis (WD7+2:1SBR+PCV – WD7+4:1SBR+PVC), the results showed 14 items better and mean better at the 95% level of significance. When Sony, FhG, Dolby and Panasonic listening test data are pooled, the results show, for absolute score analysis, the mean is better and for differential score analysis, 8 items better and mean better. David Virette, Huawei, presented
The contribution reports the results of a listening test. At the operating point of 32 kb/s stereo and differential MOS score analysis, the results showed no differences at the 95% level of significance. Philippe Gournay, VoiceAge, presented
The contribution reports the results of a listening test. At the operating point of 16 kb/s stereo and differential MOS score analysis, the results showed 1 item better at the 95% level of significance. At the operating point of 32 kb/s stereo and differential MOS score analysis, the results showed no differences at the 95% level of significance. In addition, it verifies that the CE decoder executable was able to decode the CE bitstreams and produce the CE waveforms to within +/- 1 lsb. Julien Robilliard, FhG, presented
The contribution reviewed the CE technology: The decorrelator is comprised of three stages:
The CE proposes the following changes:
The CE complexity is
Yüklə 7,54 Mb. Dostları ilə paylaş:
|