Further discussion
Later in the week, Hyunkook Lee, LG, presented additional information on their CE. He showed a detailed performance table, and the average over all test items and operating modes is shown here:
-
Item
| -
Global Gain
| -
SF Only
| -
SF and Spectral
| -
All
| -
Average compression gain
| -
0.14%
| -
0.09%
| -
0.18%
| -
0.40%
|
Kristofer Kjörling, Dolby, noted that the proposal appears to offer limited compression advantage, and hence does not recommend any action at this time.
The Chair noted suggested that the group should consider
-
whether global gain should remain a 8-bit PCM value
-
whether there should be a reset between SF and Spectral entropy coding
-
the extent that it is advantageous to have one and only one entropy coder (i.e. arithmetic coding)
-
whether channel errors should be considered when changing bitstream elements
It was the consensus of the Audio Subgroup to study the above questions (bullet items) as part of the workplan on this CE. An additional element of the workplan is that LG will give RM proponent a training corpus to encode at 9 operating modes with bitstreams returned to LG.
Herve Taddei, Huawei, presented
This contribution proposes a pulse indexing technique for coding the excitation in the ACELP coding mode. The proposed technology is lossless and shows a bit savings of 32 to 64 bits per super-frame. It proposes to use the pulse indexing tool when there are more than 5 pulses per track, which typically occurs in the higher bitrates where ACELP is employed.
Phillipe Gournay VoiceAge, noted that a savings of 64 bits per superframe corresponds to a bitrate savings of much less than 3 kb/s. He further noted that the current RM0 bitstreams either do not ever use the modes cited in the contribution.
The Chair suggested it would be very desirable to report the bitrate savings as weighted by the relative frequency that the coding mode occurred in the RM0 bitstreams. What the group suggests as additional information can be captured in the Workplan.
Dostları ilə paylaş: |