Skip to content

Deviations with MCT between BASOP and float for LTV 5.1 coded with 128kbps

Basic info

  • Float reference:
  • Fixed point:

Bug description

During verification of the LTV 5.1 at 128kbps, although there are no substantial issues or audible differences detected with the binaural subjective listening, there are segments that show larger signal differences per channel. Investigating this a bit further I found that for some segments the MCT algorithm, meaning number of channel pairs being coded jointly and channel pair combinations are deviating. As an example there is a segment of the frames 4031-4110 with large deviations that already are coming from the pre-processing. This can be confirmed from the number of bits per channel for the side-information coming from the pre-processing tools:

side_bits_per_chan

This already implies that there are differences with the signals that are input to the MCT algorithm affecting the whole processing chain henceforth:

Number of channel pairs selected from MCT:

blkCnt

Bits used for coding the MCT parameters:

mct_bits

Remaining bits to code the MCT channels:

availBits

Bits assigned to each channel:

bits_per_chan

I will investigate a bit further where the differences from the preprocessing are coming from.

Ways to reproduce

./IVAS_cod -mc 5_1 128000 48 ltv48_MC51.wav bit
./IVAS_dec 5_1 48 bit ltv48_MC51_128kbps.dec.wav