International organisation for standardisation organisation internationale de normalisation

Additional discussion

Herve Taddei, Huawei, presented additional information on the performance of this proposed tool. Philip Gournay, VoiceAge, noted that the proposed complexity reduction is modest when put in the context of the complexity of the entire decoder. The Chair noted that it would be very useful information to show the benefits delivered by the proposed tool in the context of operating the USAC system over the test set at some operating mode (bitrate for mono or stereo). Bernhard Grill, FhG, noted that if these modes are never used then removing them would give an even bigger complexity reduction. Kristofer Kjörling, Dolby, commented that the proposed CE is improving on a mode that is not used at any time for the set of operating modes and test items.

It was agree that the RM proponent will run the USAC RQE from 12 to 64 at 4kbps increments for the CfP test set and report the relative frequency of the AMR-WB+ coding modes. This will be captured in the USAC Workplan.

Ralf Geiger, FhG, presented

m16439

Proposed improvements to WD2 of USAC

Jeremie Lecomte Max Neuendorf Ralf Geiger Markus Multrus

The contribution proposes a new method for handling the transitions between wLPT (i.e. TCX) mode in LPF to FD coding modes. Currently transitions from LPD to FD have an overhead of 128 samples and necessitate at 1152 length MDCT, which is more complex than a 1024 MDCT.

LPD (ACELP) to FD requires a 128 sample overhead for overlap. This transition is not affected by the proposal.

LPD (TXC) to FD discards 128 samples and overlaps 128 samples. The discarded 128 samples are not needed if the TDAC would actually work (i.e. TDAC by the frequency domain data). The contribution shows that by exchanging the order of MDCT and LP filtering, a proper TDAC overlap can be achieved.

A listening test at 24 kb/s mono for test items that actually entailed a TCX to FD transition showed no audible difference at the 95% level of significance. It was noted that these transitions occur infrequently. In those frames affected, it showed an average a bit savings of 2.5%, and overall showed a savings of 500 bits per second. In a similar sense of the comment made by the Chair for m16338, Huawei suggested that 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.

Experts noted that, with this contribution, the transition from LPD to FD can be done in a critically sampled way, but that the transition from FD to LPD is still done in a non-critically sampled way. Kristofer Kjörling, Dolby, noted that a solution that permits the T/F coefficients to remain in critically sampled mode would be very desirable, even if bitrate savings were modest. This topic will continue to be discussed.