International organisation for standardisation organisation internationale de normalisation

8.2.3MPEG-D Unified Speech and Audio Coding

Core Experiments

Hervé Taddei, Huawei, presented

m16587

Memory and complexity reduction CE proposal for USAC LPD mode

Wei Xiao, Fuwei Ma, Hervé Taddei, Minjie Xie

The contribution proposes to use a new indexing scheme for the AVQ quantized data. The new coding scheme is lossless relative to the current coding scheme, and so its contribution is not enhancing compression but rather:

• 
  Required memory for the AVQ decoding is reduced by 43%
  
• 
  Complexity of decoding the AVQ vector is reduced
  

It was noted that when the MPEG Reference Encoder is given a 48 kHz sampling rate mono file to code at e.g. 12 kb/s it prints an “informational message” and exits. Further discussion confirmed that the MPEG Reference Encoder requires an input file of lower sampling rate in order to code at 12 kb/s. It was decided to create a shell script that illustrates the required resampling, if any, for all coding modes when reading and writing 48 kHz signal files.

There was considerable discussion concerning the complexity reduction of the proposal versus potential additional complexity reduction in the WD3 implementation. Philippe Gournay, VoiceAge, noted that, in the context of the entire decoder, the proposal provides a ROM savings of 0.42%. He further stated that the decoder complexity is 20 WMOPS and in this context the proposal provides a WMOPS savings of 0.085%.

The Chair noted that the proposal still requires as part of the CE Proposal phase:

• 
  The RM changes (syntax and description) which are necessary to incorporate the proposed technology.
  

and requires a cross-check as part of the CE Check phase of the CE process.

He further noted that, considering the large amount of discussion, it would be beneficial for experts to further study the proposal prior to further discussion later in the week.