Skip to content

Level drop in ISMx formats at higher bitrates

When assessing the I/O level variances in the ISMx pass-through mode, observations from this document reveal that the level difference tends to approach 0 as the bitrate increases from 0 to 32 kbps. However, at 48 kbps, there is an abrupt drop in the level difference, typically ranging from 1 to 1.5 dB, deviating from the trend observed at lower bitrates. This distinctive pattern persists consistently, irrespective of the number of objects or the characteristics of the input file.

The I/O levels can be measured using the following procedure (exemplary case of ISM1):

bs1770demo -nchan 1 -lev -26 -conf 0 ltv48_1ISM.cut14.pcm
./IVAS_cod -ism 1 ltv48_1ISM.cut14.wav.0.csv BRATE 48 ltv48_1ISM.cut14.pcm bit
./IVAS_dec EXT 48 bit ltv48_1ISM.cut14.dec.pcm
bs1770demo -nchan 1 -lev -26 -conf 0 ltv48_1ISM.cut14.dec.pcm

Here is the input file with its metadata file.

The input level should be -27.116006 dB. Setting BRATE to 32000 and 64000 should result in -27.660894 dB and -28.64897, respectively. Thus, there is a drop of approx. -1 dB when comparing the output levels at the two bitrates.

The assumption is that the issue originates either within the core TCX encoder/decoder or during the subsequent processing steps following the TCX core decoder.

Edited by Ghost User