Further discussion
Hervé Taddei, Huawei, presented a revised document m16587. A review by the group confirmed that it was a complete CE Proposal. Jeff Huang, Qualcomm, presented m16766, “Crosscheck report of Huawei’s CE proposal for memory and complexity reduction of USAC LPD.”
Philippe Gournay, VoiceAge, noted that alternate implementations may produce equivalent savings in memory. Hervé Taddei, Huawei, requested that he please product the alternate implementation or justify the basis of for his position. Mohammed Raad, Raadtech, noted that the technology provides ROM table savings and there is no reason to ignore this savings. Bernhard Grill, FhG, noted that it is typical for an audio coder to have 1% to 3% redundancy for a given signal.
The Chair noted that there was considerable discussion on this proposal and the tone of the discussion suggests that there is no consensus on the CE proposal at this time.
Oliver Wuebbolt, Thomson, presented
m16638
|
Thomson Core Experiment Proposal for USAC
|
Oliver Wuebbolt
|
The presentation describes an alternative context adaptive arithmetic coder for coding the USAC FD spectrum that uses one preceding spectrum but different generation and definition of contexts.
The principal advantage of the new proposal is that it reduces decoder RAM and ROM requirements. It notes that, in this context, computational complexity is difficult to assess in a consistent manner, and so no comparison is given. In addition, it provides some additional compression in the lossless representation.
JungHo, Samsung, noted that the evidence was supplied only for FD mode (and not TCX mode). The presenter indicated that Thomson used an HE-AAC encoder to develop the PDF, and so no platform for TCX was available.
It is proposed that this work be progressed via a workplan to coordinate next development steps, e.g. getting WD3 bitstreams to calculate more robust PDFs and getting third parties to do cross-checks on the experiment.
JungHoe Kim, Samsung, presented
m16598
|
Proposed CE on Spectral Noiseless Coding in USAC
|
KiHyun Choo, JungHoe Kim, Eunmi Oh
|
The contribution proposes a new tool for noiseless coding of the FD and TCX spectral coefficients. The primary value of the tool is that it provides greater coding efficiency by having a larger number of contexts. In addition, it has smaller arithmetic coding tables sizes. The tool uses the same contexts as WD3 and adds MSB and LSB context-dependent coding. Arithmetic coding PDF was estimated using 28 minutes of music signal and 25 minutes of speech signal. The experiment used the WD3 bitstreams but inserted into them the new arithmetic coding data. Bit reservoir constraints were always met and, in the case that the new tools provided some additional compression, fewer bits were written into the bitstream.
Over the nine CfP operating modes, the tool provides a bitrate savings of from 0.5% to 1.13%, when considering only the frames coded in FD or TCX modes. Arithmetic coding table sizes are reduced by a factor of 1.7.
Hervé Taddei, Huawei, asked how the complexity of the proposed tool compares to that of WD3. The presenter noted that in WD3 arithmetic coding is a small component with respect to the overall coder complexity, and so suspects that this will not be an issue. However, complexity figures will be reported at the next meeting.
Max Neuendorf, FhG, noted that in WD3 the PDF for the LSB is flat, and hence provides no basis for lossless coding gain. The presenter indicated that the proposed tool provides a transformation of the symbol during the coding process such that the LSB PDF is no longer flat.
The Chair asked that Samsung and Thomson experts work together to draft a workplan for these two CEs, and in particular, that they agree on a common framework for reporting results at the next meeting.
Philippe Gournay, VoiceAge, presented
m16667
|
Comments on Huawei's CE proposal on ACELP Indexing for USAC
|
Philippe Gournay, Bruno Bessette
|
The contribution revises a table given in m16338 from Huawei in which some bitrates were corrected. In addition, the contribution notes that WD3 can be optimized such that it has complexity comparable to that of the new tool by Huawei.
Philippe Gournay, VoiceAge, presented
m16668
|
Report on Mode Usage for ACELP in USAC
|
Philippe Gournay, Bruno Bessette, Roch.Lefebvre
|
The contribution reports that relative frequency of selection of coding modes at two operating modes:
Dostları ilə paylaş: |