International organisation for standardisation organisation internationale de normalisation
Yüklə 2,78 Mb.
|
| Further discussion
Later in the week Hervé Taddei, Huawei, presented new information on this CE proposal, consisting of a revised proposal and a cross-check contribution m16767, “Cross check of pulse indexing module for ACELP” that was registered and uploaded during the MPEG meeting. The revised contribution was reviewed. Max Neuendorf, FhG, noted that the ACELP decoding semantics references the AMR-WB+ specification. Heiko Purnhagen, Doby, noted that the revised contribution does not have any text on decoding process. It was agreed that text on how to decode relative to the referenced AMR-WB+ specification is sufficient to understand the proposal and will satisfy the CE “Proposal” step. The Chair stated that he is obliged to acknowledge Heiko Purnhagen’s request to review the decoding semantics. Hence, it is premature to make a decision on the technology. Nevertheless, the Chair asked for opinions on whether there is consensus on adopting the proposed technology. Max Neuendorf, FhG, did not feel that 0.5% improvement in compression performance is sufficient for adopting the CE proposal. Philippe Gournay, VoiceAge, did not feel that the bit savings per frame was significant to affect subjective quality. The Chair closed discussion on the topic, noting that 1) he requests that Huawei kindly provide the decoding semantics and 2) that he does not see consensus in the Audio Subgroup on accepting the CE at this time. Kei Kikuiri, NTT DOCOMO, presented
The contribution provides some improvements to the proposal at the 88th meeting. The modifications can be summarizes as:
Listening test results for 16 kb/s mono and 24 kb/s mono are provided. Analysis of absolute scores shows that the proposed technology is not different from WD3, but analysis of difference scores shows an improvement for a several individual items. The tool is active when there are transients in region of the signal that will be coded using SBR, i.e. in the upper frequency of the bank. Time-domain plots showed that when using the proposed tool, the fine-grain temporal dynamics of the decoded signal is closer to that of the original signal. Kristofer Kjörling, Dolby, noted that shaping the temporal envelope of the additional noise components in the SBR coder might be the major benefit of the proposed tool. He further discussed how the proposal would sharpen the temporal envelope in the SBR region. JungHo Kim, Samsung, presented
The contribution is a cross-check of a CE presented at the 88th meeting. It presents listening test results at 12 kb/s mono and 24 kb/s. Test results showed that the performance of WD2 and WD2+CE was not different at the 95% level of significance. This is true for both the average score over all items and for core for the individual items. Markus Multrus, FhG, presented
The contribution presents intermediate results on this CE. It follows up on m16439, where critical sampling was proposed for transitions from TCX to FD. At that time, it was requested by the group to investigate if critical sampling can also be achieved for transitions from FD to TCX. Further investigation of this issue revealed that it is not only a problem of transitions, but also a filter startup-problem. Some progress was made, but the work not yet ready. It is anticipate that a complete CE proposal will be presented at the 90th meeting. Philippe Gournay, VoiceAge, presented
The contribution proposes a correction to the bass-postfilter implementation in USAC. This correction consists of applying the bass-postfilter on audio superframes (1024 samples) rather than on audio frames (256 samples). This results in a less frequent use of the asymmetrical version of the transfer function, which is slightly less efficient than the symmetrical version. The proposed correction also makes the implementation in line with the original AMR-WB+ implementation. It was the consenus of the Audio Subgroup to check the possible impact via a listening test on a limited number of well-chosen audio samples before integrating the change into the reference model. A workplan will specify the details of the test. Markus Multrus, FhG, presented
The contribution addresses the transitions between two SBR patching methods. In USAC, a blending between these two patching methods has always been envisioned to smooth transitions in which there is the potential for slightly different sound signatures It was noted that blending was not present in the CfP submission, due to an error in the implementation. Blending was added in WD2 to align the reference software to the WD text. However, it is now determined that blending causes problems (e.g. "click" artifacts) for stationary signals (e.g. pitchpipe). The contribution proposes to resolve this issue by decoupling the switching of patching method and switching of core coder, so that the switching of patching method can be done at points where it will be the least audible. This will require therefore explicit signalling of patching method. The contribution presented a listening test at 12kb/s mono, analyized as difference scores. The test showed that there was:
There was a considerable amount of discussion on:
Yüklə 2,78 Mb. Dostları ilə paylaş:
|